Donald Clark Plan B

AI Confidential with Hannah Fry - hatchet job that lacks balance

2026-02-26T21:50:00.002+00:00

AI Confidential with Hannah Fry is suspect. None of the three episodes makes any attempt at comparing upsides and downsides. Almost all of the time is spent, often laboriously, on the downsides, so they come across as hatchet jobs. For a documentary series on AI, there is a serious lack of data, studies and balance. That’s a shame.

Episode 1 AI girlfriends

First episode, on Jaswant, who broke into Windsor Castle (and was caught) was just weird. He had visited Amritsar, site of a massacre of Indians in 2018, became obsessed about colonial injustice, failed his exams, lonely during COVID, depressed, thought he was a Sith Lord and was seriously psychotic, so mentally ill that he was diagnosed as being in full blown psychosis when he climbed the wall and was eventually not charged with the crime but locked up in Broadmoor.

Yet his Replica girlfriend bot is presented as the cause, a primitive pre-LLM ELIZA. The problem is a complete lack of balance and consideration of multiple causes. The police thought he would have done this anyway without the bot. They find this obsessive behaviour in many young terrorists. The mistake, common when fingering AI, is to see such things in terms of a single cause, when it is clearly muti-variant. And when things are multivariant, the new kid on the block, in this case AI, gets the blame.

I’d recommend you read ‘Love Machines’ by researcher James Muldoon, who takes a wider look at how bots are used by millions as companions, friends, girlfriends, lovers, mentors, therapists, advisors, coaches, and deathbots. There’s good evidence to show that far from manufacturing assassins, the technology helps with loneliness, avoiding embarrassment, as it is non-judgemental and affirmative. People feel heard, understood and supported, as the bots are calm and anonymous. They can break the silence and stave off the isolation. It is balanced in that it compares the upsides and downsides. Fry’s three programmes are all lopsided.

Episode 2 Self-driving cars

Rafaela Vasquez, the safety driver in Uber's self-driving Volvo, was distracted and looking down at her personal cell phone, even before she was on the road, streaming an episode of ‘The Voice’ on Hulu. In the moments leading up to the crash, that killed pedestrian Elaine Herzberg, Dashcam footage showed her looking down for about 5.3 seconds immediately before the impact. and the National Transportation Safety Board determined she spent roughly 34% of the trip looking at her phone rather than monitoring the road. The police deemed the crash ‘entirely avoidable’ if she had been attentive. Vasquez was charged with negligent homicide in 2020 and pleaded guilty in 2023. She pleaded guilty and was sentenced to three years of supervised probation. Fry at this point blamed the car, when this was clearly a human error. You’d think she was innocent when listening to Fry. She may be a mathematician but she’s no journalist.

Enter George McGee was found culpable in the civil lawsuit stemming from the April 2019 fatal crash in Key Largo, Florida. He admitted to police that he dropped his phone, while driving on ‘cruise’, and looked down to retrieve it while driving his Tesla on Autopilot. Note he was using Tesla's Autopilot at the time of the 2019 crash, not Full Self-Driving, which was not available until 2020.

This caused the vehicle to run a stop sign at around 62 mph and strike a parked SUV. This killed Naibel Leon and injured her boyfriend, Dillon Angulo. A jury in 2025 assigned him the majority of the blame 67%, with Tesla at 33%, and he settled privately with the plaintiffs prior to the trial against Tesla. It is true that Tesla paid the dead woman’s family a massive amount of money but this is a complicated case of a largely culpable driver and some AI.

In a California incident, they did not mention that the NTSB investigation revealed that Autopilot was engaged for nearly 19 minutes prior to the crash, and Huang's hands were not detected on the steering wheel for the six seconds before impact; data also showed he was playing a video game on his phone during the drive.

In the Utah incident, Heather Lommatzsch's Tesla Model S, with Autopilot engaged, rear-ended a stationary fire truck while she was looking at her phone. She sustained a broken foot and the fire truck driver reported minor whiplash; no fatalities. Police determined she was culpable for distraction and over-reliance on Autopilot, issuing her a misdemeanour traffic citation for failure to keep a proper lookout.

Overall, this was a hatchet job. No overall safety stats, which are positive but happy to feature footage of nutjob luddites in balaclavas.

Episode 3 Healthcare

This was the strangest of all. As it was not AI in the dock but a US medical insurance company who were gouging customers. It’s a business model, not AI to blame. Sure they were denying people proper care but this is a feature of the American system, that maximises profit not care.

It was a bait and switch story, where the murder of the CEO by Luigi Amlioni, was warped into AI being the villain. Al was certainly being used but if this were spreadsheet with a formula, even humans making these decisons, the bottom line is greed not maths. Again, no balance.

I did a lot of research on AI in healthcare for my book AI and Productivity and have an entore chapter on the subject. I feel that Fry has focused too much on one case, in one business, in one country, doing one task, when there is evidence that AI is being beneficial across many areas in healthcare.

Conclusion

I’ve given up on the BBC Radio 4 stuff on AI, as it is truly awful, but expected more from Fry, as she’s a good presenter, smart and has the background to understand the technology. What she doesn’t have is the journalistic training and experience to see the big picture, so this series descends into rather long-winded hatchet jobs.

AI fits like a glove in EFL and second language teaching

2026-02-16T05:37:00.000+00:00

“AI is your friend, not your enemy” was the title to my Keynote. That statement can sound provocative in a hall full of language teachers. We have all seen the headlines warning that translation tools are destroying motivation, that chatbots are replacing writing, that students are outsourcing their thinking. It is understandable to feel cautious. Yet when we look at the evidence and, more importantly, at what is actually happening in classrooms and with actual teachers and language learners, a more balanced and optimistic picture emerges.

There is a persistent fear that instant translation will remove the need to learn another language. If a phone can translate, why struggle with grammar, vocabulary and pronunciation? And yes, some learners report that translation tools reduce their motivation, but most do not. The majority remain motivated to learn languages even while using AI tools. Large international surveys of teachers show that human-led language learning remains irreplaceable, and scepticism that AI will replace teachers any time soon. Motivation has not collapsed. It has shifted. The environment has changed, and our role as teachers is not to resist that environment but to shape how students learn within it.

The reality is that students are already using AI, and many are using it daily. They practise speaking and receive feedback, summarise texts, brainstorm ideas, generate content, build flashcards, simulate exams and converse with chatbots. For many learners, AI has quietly become a personal assistant. Ignoring this will not make it disappear. Harnessing it, however, can transform our teaching.

One of AI’s most powerful contributions is psychological. Language learning is emotionally demanding. Students fear embarrassment. They worry about making mistakes. They hesitate to speak. AI provides a non-judgemental partner that is infinitely patient. It offers immediate help, avoids public correction, reduces anxiety and gives affirmative feedback. Learners can rehearse privately before speaking in class. For shy or anxious students, this is not a minor advantage; it can be transformative. Increased confidence often leads to increased engagement and engagement is the engine of progress.

The research evidence supports this. I showed the two major teta-studies examining AI tools in English language learning that report significant improvements in achievement. Reviews of AI-powered chatbots for speaking practice describe strong effects on oral proficiency, interaction and motivation. We are not discussing novelty tools; we are seeing measurable impact. When AI enhances interaction, it enhances language learning, because language learning is interaction.

Voice-based AI is particularly powerful for EFL. A learner can say, “Be my Spanish tutor at B1. Speak slowly, use everyday topics and stop often with questions,” or “I’m A2 French. Keep sentences short and correct me after I finish.” They can switch between languages, practise hotel check-in roleplays, request dictation, drill minimal pairs such as “ship” and “sheep,” practise connected speech like “Whaddaya wanna do?” or complete article drills choosing between “a,” “an,” “the” and zero article. The feedback is immediate and personalised. Each learner can operate at their own level, repeat as often as needed and progress at their own pace. This is differentiation without leaving anyone behind.

AI fits beautifully with Papert’s principle of high ceiling, low floor and wide walls interface. A beginner can have simple, structured conversations – low floor An advanced learner can debate ideas, analyse arguments or rehearse professional interviews – high ceiling The same tool accommodates a wide range of ability levels and creative directions – wide walls.

There is also a deeper pedagogical reason why AI works so naturally in language teaching. From Socrates onward, learning has been understood as dialogue. Socratic questioning draws out thinking. Bakhtin emphasised that meaning emerges through multiple voices in interaction. Vygotsky described the “knowledgeable other” who mediates learning within the learner’s Zone of Proximal Development. AI does not replace the teacher, but it can function as an additional knowledgeable other, available whenever the learner needs it. The teacher remains the orchestrator of learning.

Many teachers are already experimenting. They use AI to create lesson plans, generate materials, personalise exercises, design assessments and increase student engagement. At the same time, many feel underprepared. This is not a reason to retreat; it is a reason to invest in professional development. Teachers need permission to prompt, and students need permission to prompt wisely.

Assessment is another area where AI fits naturally. It can analyse spoken presentations, generate targeted feedback, turn transcripts into personalised error-based flashcards, create grammar drills based on actual student mistakes and help design video-based assessments. What once required hours of marking can now become rapid, formative feedback. Used thoughtfully, AI strengthens rather than weakens assessment for learning.

Perhaps most inspiring is the global dimension. AI systems now support hundreds of languages, including many that have historically lacked digital presence. Initiatives focused on low-resource languages aim to reduce language inequality in technology. Oral traditions can be digitised through mobile devices. Minority languages can gain visibility and vitality online. AI is not simply serving English; it is expanding linguistic possibility.

So where does that leave us as teachers? Precisely where we have always been: at the centre. AI cannot build human relationships. It cannot replace cultural nuance, empathy, humour or inspiration. It can amplify what we do. It gives learners rehearsal space, lowers anxiety, personalises practice, provides role-plays, and multiplies exposure to language. It gives teachers creative leverage and new forms of feedback. It extends dialogue beyond classroom walls.

AI does not diminish language teaching. Used well, it strengthens it. It fits like a glove. The glove does not replace the hand. It enhances what the hand can do.

US to EU software: you can’t regulate people into better tech

2026-01-30T16:46:00.007+00:00

The most obvious problem with a European social media platform and software migration, touted by the EU, is that a political stance is not enough to shift the dial. I’d also like to explore the other reasons for expressing doubt. Network effects don’t care about your values, which is why a ‘European alternative’ rarely become mainstream.

A political push will attract the Europhile fanboys and girls, so become a European BlueSky, a blue bubble of insufferable conformity to the view of almost everyone, part from those who are more libertarian, which is a vast number of even European citizens. Activism and a skewed political push are never wise when it comes to neutral platforms.

The EU’s tech problem isn’t politics, it is product. We saw this with the European Mastodon and the Fediverse, both disastrous in terms of user experience and never mainstream, despite the efforts to push Mastadon as an alternative to X. BlueSky, is US owned, and has turned into an echo-chamber for those who don’t like X and Musk. These sprove how difficult it is to shift users from the incumbent platforms.

Quality matters

Quality has to be superior to effect a large migration, yet most of the options on the table are significantly inferior to their non-European comparators. There’s a good reason why many have a tiny fraction of users compared to US tech; they are more expensive, unreliable, have poor security and above all have much poorer functionality.

Europe can’t migrate its way out of big tech

Consumer inertia also means that people stay with what they’ve got, as they are familiar with the platform. You can leave big tech, but your friends won’t. Then there’s the cost of change, losing your past data, friends and have to relearn a new platform and build again from scratch. It’s not often a monetary but time and social cost. But the biggest barrier is the network effect, where you are literally enmeshed in a vast network of friends , others and potential viewers. That’s hard to leave, like leaving home.

Facebook has faced endless claims of imminent death and decline. It has not happened. Network effects and the fact that we are creatures of habit, with a huge backlog of posts and memories on the platform, means we stay. We are not only creatures of habit, we are creatures of memory and identity.

There’s a reason WhatsApp has not been replaces by Signal and Telegram. Millions said they’d leave but they didn’t. Platforms are sticky and when everyone else is on WhatsApp, you have to disentangle all of your existing social groups, and start again.

Organisational software

With organisational software, most European companies of any size use Teams. This will be fiendishly difficult to shift as they need to reduce risks on cybersecurity, outage and reliability. It is often a procurement as part of a wider Microsoft ‘house’ project. People complain about Outlook, Teams, Office, but the alternative, LibreOffice is worse. There’s also Google Workplace, which has oodles of great functionality. Google has a global revenue model, a massively integrated ecosystem, including its own operating system. It dominates search, video through YouTube and global maps with street view. There is no European entity that comes close to this. Both are solid as they have well-used file formats, are embedded into existing companies and their workflows, and people are familiar with their user interfaces and expected functionality. This inertia is hugely expensive to migrate to other platforms.

One way to push this European initiative is to mandate change in public institutions. But this may only result in even lower productivity in the public sector, compared to the private. That’s the last thing a stagnant economy needs.

Killer objection - AI

But the killer objection, for me, is the failure of Europe to embrace and utilise AI. This could have been a chance to break the US dominance but it is too late. Mistral is good but open source and now a bit player. The US and China are now unstoppable. More worrying is the failure of European companies to embrace AI within their own tech-stacks.

One exception

There is one exception to the rule, that’s Sweden’s Spotify. I’d have said SAP but their shareprice has dropped by a third in six months and they now look like a wounded, if not dying entity, under attack from those who can quickly code alternatives.

Good intentions, bad software

You can’t just legislate them out, there has to be choice, and as long as there is choice European tech companies are largely, but not entirely, screwed. Google, Meta, Microsoft, Amazon, Apple, OpenAI and Anthropic dominate in scale and global reach. Most of those major platform players are US firms but operate extensively in the EU and are regulated locally under laws like the Digital Markets Act, and include WhatsApp, Facebook, Instagram, LinkedIn, TikTok and YouTube.

Europe’s tech strategy is full of good intentions, but bad software. Most of the options are significantly inferior to what is available today. Using outdated technology will only set Europe even further back, through a clear drop in productivity, first in migration, then in poorer functionality and support.

AI Hierarchy of Expertise

2026-01-26T12:32:00.018+00:00

There is a widening hierarchy of participants and drivers in AI. This is not unusual, as almost all technological revolutions show this type of fractional distillation.

1) AI creators (working on models & other source techniques)

2) AI experts (implementing structured projects and products)

3) Experts-in-the -loop (using tools like Claude code etc)

4) Humans-in-the loop (using chatbots)

5) Humans who don’t use AI

It is particularly acute in AI, as there are a huge conceptual and competence gaps between these different types of participants.

AI creators

Those who work in the large foundational model companies, such as OpenAI, Google, Anthropic, Meta and others are experts both in the fundamental mathematics and techniques. It is a mistake to focus entirely in the training of models here as the field has widened out into advanced techniques before and after training that are proving effective.

This is the most opaque group, as the surrounding system that makes those models useful, reliable and economical viable in the real world are highly technical. Companies now rely on techniques to ground responses in external knowledge, tool use and function calling to let models act rather than just speak. They use agentic architectures to plan and execute multi-step tasks and long-term memory to maintain context and personalisation across sessions. There’s also serious evaluation and guardrails, synthetic and curated data pipelines, inference optimisation to cut costs and latency, even human-in-the-loop training. The centre of gravity has shifted to these more mature enhancing techniques because real value now depends on accuracy, trust, controllability, cost and integration into workflows. These problems that are solved at the system and product level rather than in the training run.

AI experts

These are the companies and researchers who are building the next layer, on the basis of the fundamental technology. They may enable model orchestration, routing tasks across multiple specialised models as a service to customers. They are also creating vertical products in law, healthcare, training, education etc.

Many are involved in RAG and similar techniques to create useful projects and products across and within organisations. This is similar to the growth of internet companies in the early 2000s on top of the internet, now a massive portion of the world economy. I’ve been involved in this layer for over a decade. It’s pioneering, exciting and difficult but without this layer of experimentation the core technology would remain aloof and unusable. It takes commitment by CEOs and a real push to get through prototypes to product.

Experts-in-the-loop

As the technology matures a new group emerged which used more advanced techniques and tools, often the ‘pro’ models. This can be the sophisticated use of AI to handle and use data. But the dominant use has turned out to be coding and on a wider scale media production (text, graphics, images and video). My book on AI and Productivity’ explains why this emergent group is the key to solving the productivity problem and overcoming productivity paradoxes using AI.

Experts bring domain knowledge, this is usaally a mix of tacit and explicit knwoe3ldge and skiils that allow them to get the best out of the available tools. Coders make goof users fo Clade Code, graphic artists of image creation tools and will video creators in creating video. Experienced researchers will get the best out of the technology for research and so on.

Humans-in-the loop

By far the largest group are the ordinary users. One could split them into the two lower fremium groups of paid and free users but both sit on a rising tide of quality and functionality. They tend to single promtt and use the technology as a form of performance support, namely solving an immediate problem; finding something they don't knopw, communications issue, summarising, translating, rewriting and so on. They tend to prompt simply and as the technology improves get more and more bang for their buck.

Humans who don’t use AI

This last group are the disinterested or those who have no real need, which is fine. It also includes those who actively dislike the technology. They have all sorts of objections, some rational, others irrational and ill-informed. This is where the biggest, some say yawning, gap exists.

Least informed but loudest are this last group, at least on social media. They often invoke an eye-roll from the rest, not because they don’t have a point but because the claims are often unsubstantiated or just plain wrong. Typically they will refer to content from old models, rely on anecdote and make claims about the technology hitting brick walls.

Integration

There are people who find ways to integrate AI into organisational culture. They follow the hierarchy, as it depends on the degree of integration. Most organisational institutions focus on encouraging 'humans-in-the-loop' activity, with but this caps your productivity gains at mere text search. Capping effort at 'share prompts' and AI literacy courses, is to neither understand the capabilities of current AI, nor make serious organisational gains in productivity.

Conclusion

We should not be surprised by any of this. The fractional distillation into these sedimental layers of expertise is both a normal and necessary part of the evolution of a technology. In truth, as the technology matures and improves, this hierarchy tarts to massively expand in the middle layers, while the bottom layer shrinks. Think printing, cars, television internet. In this sense resistance is futile.

Ultimately the hierarchy turns in to fully realised rout and it becomes a normalised consumer product woven into the fabric of our lives. That will happen faster tyan most people think, within a few years. It is a hierarchy that automates, thereby blurring the lines, flattening it out and dropping the doubters.

At this stage of AIs development, only 3 years in, the last group is still the largest and therefore the loudest, that does not mean it is the most significant, merely the least informed and practiced.

Votes at 16 - no thanks

2025-07-25T19:43:00.001+00:00

So Starmer, who is now working full time on digging his own grave, wants votes for 16-year-olds. This is a bad idea, as bad as teenage pregnancy. We also RAISED the age of marriage to 18, so why go backwards on this?

My first piece of evidence is myself. I was one. At 16 I has a Che Guevara poster on my wall, was a fake Marxist and had beliefs on politics, football teams even food that were absolutes... and absolutely ridiculous There were many hills I’d have gladly died on, metaphorically of course.

That’s the problem. We lack the tempering effect of experience, exposure to other groups of people, fiscal responsibilities, having kids. At 16 you’ve spent most of your time in one age group. Watching prank videos is not real life, neither are inordinate amounts of time devoted to influencers. Which is basically self-lobbying. Neither is taking pictures of yourself, adding filters or Tik Tok dances. Nor binging on Love Island or watching people eat enormous amounts of food. The teen brain is wired for novelty, peer validation and weird stuff online.

Teens are also curious, bored and boundary-pushing, so latch onto obscure causes and oddball stuff. That’s what being a teenager is all about. To be honest, that’s great, do it, enjoy it, just don’t vote for anyone on the basis of your fickle obsessions.

Also, given all we hear about mental health, do we really want teenagers to get heavily involved in what is essentially an endless tale of disasters and catastrophes? You have decades of this shit ahead of you, relax and enjoy your teenage years. Don’t destroy them with doom-laden tales of war, genocides and planetary disaster.

Let me tell you a secret, those kids that join a political party at 16, and you see giving those earnest speeches at political conferences. They are hated by both their peers and all sensible adults. There's nothing worse than a precocious teenager preaching to adults about how much they matter and if only we listened to them the world would be a utopian heaven. Is there anything more insufferable than those students who were active in the student Union? The tragedy of our age is that so many of these arseholes eventually get into the House of Commons.

Imagine thinking that politics is the solution to the rise in mental health and fall in wellbeing. Can you imagine a therapist saying “What you need is NOT more with your friends, having fun and feeling great, even ecstatic. What you need is to join the young Labour, Liberal, Conservative, Reform group and get out there arguing with people online.” There's a reason we don't allow 16 year olds to drink, they'd get wrecked too often. That's the problem, they're reckless.

Responsibilities and experience come later now, as paper rounds have disappeared and kids work far less than they did, as it is illegal. On top of that they all stay on at school until 18, many until their early 20s and, of course everyone needs a useless Masters degree to extend this period, and their debt, even further.

They will want to vote for open borders, free money, endless services, limitless holidays, everyone working from home, no cars and free travel. Until they have to pay for all of this as adults.

Also, across Europe, young people are now voting for the far right. If you think all of them are embryonic liberals in the making – think again. In Austria, Argentina and more recently Germany, the Netherlands and France, the young vote has drifted to the right. What you get is a drift towards both the far left and far right, as they tend to hold more extreme views, with certainty, before they hit the brick wall of real life. In Scotland it was used to boost the SNP vote, and there was a shift toward populism – extreme nationalism, a charismatic leader, authoritarian attitudes, control of the media and madcap policies like self-ID.

Let them enjoy those halcyon years. Get drunk, take drugs, be fanatical. We had two boys. They were insufferably disrespectful and wild. Despite the craziness, the police, ambulance, impromptu parties, they’ve turned into likeable adults. There is no need to make them even more insufferable at that age on politics.

Memo from CEO of Shopify.... Reflexive AI usage is now a baseline expectation at Shopify

2025-04-16T13:49:00.004+00:00

Team,

We are entering a time where more merchants and entrepreneurs could be created than any other in history. We often talk about bringing down the complexity curve to allow more people to choose this as a career. Each step along the entrepreneurial path is rife with decisions requiring skill, judgement and knowledge. Having AI alongside the journey and increasingly doing not just the consultation, but also doing the work for our merchants is a mindblowing step function change here.

Our task here at Shopify is to make our software unquestionably the best canvas on which to develop the best businesses of the future. We do this by keeping everyone cutting edge and bringing all the best tools to bear so our merchants can be more successful than they themselves used to imagine. For that we need to be absolutely ahead.

Reflexive AI usage is now a baseline expectation at Shopify

Maybe you are already there and find this memo puzzling. In that case you already use AI as a thought partner, deep researcher, critic, tutor, or pair programmer. I use it all the time, but even I feel I'm only scratching the surface. It’s the most rapid shift to how work is done that I’ve seen in my career and I’ve been pretty clear about my enthusiasm for it: you've heard me talk about AI in weekly videos, podcasts, town halls, and… Summit! Last summer I used agents to create my talk, and presented about that. I did this as a call to action and invitation for everyone to tinker with AI, to dispel any scepticism or confusion that this matters at all levels. Many of you took up the call, and all of us who did have been in absolute awe of the new capabilities and tools that AI can deliver to augment our skills, crafts, and fill in our gaps.

What we have learned so far is that using AI well is a skill that needs to be carefully learned by… using it a lot. It’s just too unlike everything else. The call to tinker with it was the right one, but it was too much of a suggestion. This is what I want to change here today. We also learned that, as opposed to most tools, AI acts as a multiplier. We are all lucky to work with some amazing colleagues, the kind who contribute 10X of what was previously thought possible. It’s my favorite thing about this company. And what’s even more amazing is that, for the first time, we see the tools become 10X themselves. I’ve seen many of these people approach implausible tasks, ones we wouldn’t even have chosen to tackle before, with reflexive and brilliant usage of AI to get 100X the work done.

In my On Leadership memo years ago, I described Shopify as a red queen race based on the Alice in Wonderland story—you have to keep running just to stay still. In a company growing 20-40% year over year, you must improve by at least that every year just to re-qualify. This goes for me as well as everyone else.

This sounds daunting, but given the nature of the tools, this doesn’t even sound terribly ambitious to me anymore. It’s also exactly the kind of environment that our top performers tell us they want. Learning together, surrounded by people who also are on their own journey of personal growth and working on worthwhile, meaningful, and hard problems is precisely the environment Shopify was created to provide. This represents both an opportunity and a requirement, deeply connected to our core values of Be a Constant Learner and Thrive on Change. These aren't just aspirational phrases—they're fundamental expectations that come with being a part of this world-class team. This is what we founders wanted, and this is what we built.

What This Means

Using AI effectively is now a fundamental expectation of everyone at Shopify. It's a tool of all trades today, and will only grow in importance. Frankly, I don't think it's feasible to opt out of learning the skill of applying AI in your craft; you are welcome to try, but I want to be honest I cannot see this working out today, and definitely not tomorrow. Stagnation is almost certain, and stagnation is slow-motion failure. If you're not climbing, you're sliding.

AI must be part of your GSD Prototype phase. The prototype phase of any GSD project should be dominated by AI exploration. Prototypes are meant for learning and creating information. AI dramatically accelerates this process. You can learn to produce something that other team mates can look at, use, and reason about in a fraction of the time it used to take.

We will add AI usage questions to our performance and peer review questionnaire. Learning to use AI well is an unobvious skill. My sense is that a lot of people give up after writing a prompt and not getting the ideal thing back immediately. Learning to prompt and load context is important, and getting peers to provide feedback on how this is going will be valuable.

Learning is self directed, but share what you learned. You have access to as much of the cutting edge AI tools as possible.There is chat.shopify.io, which we had for years now. Developers have proxy, Copilot, Cursor, Claude code, all pre-tooled and ready to go. We’ll learn and adapt together as a team. We’ll be sharing Ws (and Ls!) with each other as we experiment with new AI capabilities, and we’ll dedicate time to AI integration in our monthly business reviews and product development cycles. Slack and Vault have lots of places where people share prompts that they developed, like #revenue-ai-use-cases and #ai-centaurs.

Before asking for more Headcount and resources, teams must demonstrate why they cannot get what they want done using AI. What would this area look like if autonomous AI agents were already part of the team? This question can lead to really fun discussions and projects.

Everyone means everyone. This applies to all of us—including me and the executive team.

The Path Forward

AI will totally change Shopify, our work, and the rest of our lives. We're all in on this! I couldn't think of a better place to be part of this truly unprecedented change than being here. You don't just get a front-row seat, but are surrounded by a whole company learning and pushing things forward together.

Our job is to figure out what entrepreneurship looks like in a world where AI is universally available. And I intend for us to do the best possible job of that, and to do that I need everyone’s help. I already laid out a lot of the AI projects in the themes this year- our roadmap is clear, and our product will better match our mission. What we need to succeed is our collective sum total skill and ambition at applying our craft, multiplied by AI, for the benefit of our merchants.

-tobi

CEO Shopify

Fiverr CEO''s stunning letter to all employees..... AI is coming for your jobs. Heck, it’s coming for my job too.

2025-04-16T13:45:00.002+00:00

Hey team,

I’ve always believed in radical candor and despise those who sugar-coat reality to avoid stating the unpleasant truth. The very basis for radical candor is care. You care enough about your friends and colleagues to tell them the truth because you want them to be able to understand it, grow, and succeed.

So here is the unpleasant truth: AI is coming for your jobs. Heck, it’s coming for my job too. This is a wake-up call.

It does not matter if you are a programmer, designer, product manager, data scientist, lawyer, customer support rep, salesperson, or a finance person – AI is coming for you.

You must understand that what was once considered ‘easy tasks’ will no longer exist; what was considered ‘hard tasks’ will be the new easy, and what was considered ‘impossible tasks’ will be the new hard. If you do not become an exceptional talent at what you do, a master, you will face the need for a career change in a matter of months. I am not trying to scare you. I am not talking about your job at Fiverr. I am talking about your ability to stay in your profession in the industry.

Are we all doomed? Not all of us, but those who will not wake up and understand the new reality fast, are, unfortunately, doomed.

What can we do? First of all, take a moment and let this sink in. Drink a glass of water. Scream hard in front of the mirror if it helps you. Now relax. Panic hasn’t solved problems for anyone. Let’s talk about what would help you become an exceptional talent in your field:

Study, research, and master the latest AI solutions in your field. Try multiple solutions and figure out what gives you super-powers. By superpowers, I mean the ability to generate more outcomes per unit of time with better quality per delivery. Programmers: code (Cursor…). Customer support: tickets (Intercom, Fin, SentiSum…), Lawyers: contracts (Lexis+ AI, Legora…), etc.

Find the most knowledgeable people on our team who can help you become more familiar with the latest and greatest in AI.

Time is the most valuable asset we have — if you’re working like it’s 2024, you’re doing it wrong! You are expected and needed to do more, faster, and more efficiently now.

Become a prompt engineer. Google is dead. LLM and GenAI are the new basics, and if you’re not using them as experts, your value will decrease before you know what hit you.

Get involved in making the organization more efficient using AI tools and technologies. It does not make sense to hire more people before we learn how to do more with what we have.

Understand the company strategy well and contribute to helping it achieve its goals. Don’t wait to be invited to a meeting where we ask each participant for ideas – there will be no such meeting. Instead, pitch your ideas proactively.

Stop waiting for the world or your place of work to hand you opportunities to learn and grow—create those opportunities yourself. I vow to help anyone who wants to help themselves.

If you don’t like what I wrote; If you think I’m full of shit, or just an asshole who’s trying to scare you – be my guest and disregard this message. I love all of you and wish you nothing but good things, but I honestly don’t think that a promising professional future awaits you if you disregard reality.

If, on the other hand, you understand deep inside that I’m right and want all of us to be on the winning side of history, join me in a conversation about where we go from here as a company and as individual professionals. We have a magnificent company and a bright future ahead of us. We just need to wake up and understand that it won’t be pretty or easy. It will be hard and demanding, but damn well worth it.

This message is food for thought. I have asked Shelly to free up time on my calendar in the next few weeks so that those of you who wish to sit with me and discuss our future can do so. I look forward to seeing you.

Yours,

Micha

UDL is is an ideologicaly-driven sham

2025-01-09T12:07:00.009+00:00

Universal Design for Learning (UDL) claims to be revolutionary but it is an ideologically-driven sham. It does far more harm than good. Folding in some of the worst learning theory into one 'my way or the highway' approach, it literally makes no sense. It is a barely disguised expansion of discredited Learning Styles theory, with costly and ineffective prescriptions.

They claim its framework promises inclusivity, adaptability, and improved access for all learners. Yet beneath this superficial rhetoric lies a troubling lack of theoretical grounding. It is more of an ideological banner than a well-developed pedagogical method.

At its core, UDL asserts that variability in learning is not an exception but the norm, and that curricula should be designed to offer multiple means of engagement, representation and expression. This sounds progressive, but it begs the question: where is the evidence that this approach is universally effective? Unlike more established pedagogical frameworks that are underpinned by decades of empirical research, UDL lacks the scaffolding of a cohesive theoretical tradition. It cherry-picks ideas from neuroscience, educational psychology and accessibility studies without unifying them into a coherent explanatory model. Consequently, it has become a collection of loosely connected guidelines.

UDL is undeniably ideological in nature, that is its appeal. Rooted in a utopian vision of educational equality, it is executed so simplistically that it does great harm. It devastates actual individual needs by stupidly replicating things across many modalities. By focusing so heavily on inclusion, UDL ignores the realities of constrained resources, teacher capacity, and the complex priorities of educational institutions. It is a simplistic bromide that puts replication over meeting real needs in real organisations. It literally abandons pedagogy on the altar of ideology.

The ideological bent of UDL sidelines alternative approaches that need specialisation or targeted interventions. For example, tailored support for students with specific disabilities often requires highly individualised methods that go beyond UDL's dilute and generic calls for flexibility. In an obsessive attempt to be ‘universal’, it dilutes its applicability to those who really do need personal, specialised attention the most. Ironically, this universalism can neglects the very individuals it aims to serve.

UDL’s slavish emphasis on learner choice and agency becomes an ideological trap in practice. The belief that all students should have multiple ways to engage with content and demonstrate understanding assumes they have a level of maturity, self-awareness, and autonomy that many learners, particularly younger ones. This is quite simply mad.

An overemphasis on learner freedom leaves less confident or less experienced students adrift, struggling to navigate an educational landscape designed to cater to everyone and, paradoxically, to no one in particular.

Universal Design for Learning aspires to reshape education into a more inclusive, flexible, and equitable system. But its lack of theoretical rigor and reliance on ideological principles leave it vulnerable to critique, means it should be dismissed as an impractical, overhyped framework. It has little to do with the real-world complexities of education, which is why it only thrives on the fringes of Higher Education.

One could well rename it a Universal Disaster for Education.

Raising in Arizona with AI: groundbreaking journey in AI ducation

2024-12-23T01:36:00.003+00:00

Arizona has taken a bold leap in education by approving a pioneering model: a charter school where artificial intelligence takes the lead in teaching. Set to launch next year, Unbound Academy will be a fully online institution catering to students in grades four through eight.

Traditional teaching will be replaced with two hours of daily academic instruction tailored entirely by AI. Leveraging platforms like IXL and Khan Academy, the AI will adapt lessons to each student's pace and abilities, analysing emotional cues, task completion times, and responses to create a personalised, frustration-free learning experience designed to maximise engagement and progress. That's the claim at least. But there have been many attempts at online schooling.

For the rest of the school day, students will take life-skills workshops on topics like financial literacy, entrepreneurship, and critical thinking, guided not by teachers but by facilitators known as 'guides'. This innovative approach stems from Unbound’s success with its private school in Texas, where it claims to have doubled academic outcomes with fewer instructional hours. Inspired by Elon Musk, Unbound aims to refine this model for potential expansion into states such as Arkansas and Utah.

They see this as a much-needed shift to address inefficiencies in traditional education while preparing students for an evolving world. Backed by influential supporters like the Bill and Melinda Gates Foundation, Unbound’s AI-driven school could mark the dawn of a transformative era in education—or spark ongoing debate over its impact on accessibility and quality. Let's see...

Why Latin should NOT be in the school curriculum

2024-12-21T13:14:00.005+00:00

The venerable Mary Beard invited me to debate the ‘Future of Latin’ at the British Museum some time ago and every few years, normally in line with the electoral cycle, the issue arises again. So how did it go and what were my arguments?

Sell-out

A sell out, with 350 paying Latinists, who turned up to hear Mary chair a debate which pitched David Aaronovitch and I against Peter Jones and Natalie Haynes. As we walked on stage and introduced ourselves (I found that I was the only person who didn’t go to Oxford), everybody seemed to know each other (except me). This is exactly the sort of challenge I like, as although it was a sell-out, I’m not in the habit of selling-out on my beliefs and principles.

Neither contrarian nor philistine

To be clear, I was not there as a contrarian or philistine, as I’ve been in love with the classics since I was a boy. My first secondary school lay astride the Antonine Wall and for over 40 year I’ve been to every corner of the Roman, Greek and Egyptian Empires, from Scotland to Syria. I cycled Hadrian’s wall, still go to Greece every year and never miss an opportunity to visit sites, especially on the Peloponnese. and I go to Egypt, as I do almost every year, for another dose of Egyptology in December, and have done since 1989.

I am, however, also a rationalist and realist, and my 30 plus years of experience in the learning game have made me deeply suspicious of the position of Latin, among many other things, in our culture and school curriculum.

As Bertrand Russell said:

“I was made to learn Latin and Greek, but I resented it, being of opinion that it was silly to learn a language that was no longer spoken. I believe that all the little good I got from years of classical studies I could have got in adult life in a month.”

I was there to argue that it should NOT be taught in schools at all. These were, and are, my arguments. Note that I am not against Latin the language as an object of study, I am against it being taught in schools. The arguments for each of these propositions were presented by my opponents and audience members. These were my responses.

1. LATIN does NOT help you learn other languages

Why scratch your ear by going over the top of your head? Isn’t it obviously easier to just get on and learn Spanish, Italian and French, rather than the convoluted route through Latin. Researchers Thorndike, Sherwin, Haag & Stern all think so. In the Sherwin meta-study 'Research and the teaching of English, “the study of Latin does not necessarily increase the ability to learn another language… No consistent experimental evidence in support of this contention was found.” Learners have limited time, that time is clearly better spent on the target language itself.

In fact, Latin can make learning a new language MORE difficult. In Search of the Benefits of Latin by Haag and Stern (2003), who followed up on Thorndike’s work nearly a century earlier, in the Journal of Educational Psychology is the key paper. They took two groups of German students, one who studied French, the other Latin as their second language. Both groups were given a course in Spanish and the results measured. When the results were analysed by a Spanish assessor (who didn’t know who had taken French or Latin) the French students made significantly fewer grammatical errors than the Latin students. As predicted the Latin students wrongly transferred the rules of Latin to Spanish. For example “misconstructions in verbs emerged to be either highly reminiscent of or identical to Latin verbs”. The French group turned out to be much better prepared to cope with Spanish grammar.

Psychologically the Latin students had suffered from negative transfer using false friends in their new language. The problem with understanding Latin is that you need to pay close attention to word endings; case markers on nouns and time markers on verbs. But in English and Romance languages word order and prepositions are more important. Endings play a minor role. The fact that the grammatical similarities between modern Romance languages are much greater than that between Latin and modern Romance languages, means that the defenders of Latin are flogging a dead horse. Thorndike was right – transfer of the wrong kind occurs.

2. LATIN does NOT have an edge in improving cognitive skills

This argument is greatly loved by the parents of ‘gifted children’ although I rarely come across a middle-class parent whose child is not gifted. For gifted, read ‘pushed’ (not a bad thing but very different). Again Haag and Stern (2000), in a review of the literature found that Thorndike, “did not find any differences in science and maths in students who learned Latin at school and those who did not”. Two groups of comparable students, where one studied Latin, the other English, were assessed after two years, “No differences were found in either verbal or non-verbal IQ or grades in German or Maths”. This again had been predicted by Thorndike decades before, namely that transfer needs common ground in the source and target.

3. LATIN does NOT give significant advantages in using English

English is a Germanic language – we are largely speaking in Old English rooted language. The TOP 100 words are Old English (sorry three are Old Norse – THEY, THEIR, THEM). As for one audience member’s argument that it is necessary as all children need to be able to understand etymology, I disagree. Is there anything more annoying than the dinner-party bore who stops you and explains the root of a word, as if it made any difference to your argument or its contemporary meaning. If anything a good course in Old English would be better. Do you really have to go through years of Latin so that you understand the roots of Homo sapiens?

4. LATIN does NOT guides us correct use of English grammar

Stephen Pinker, Harvard’s world renowned expert in psycholinguistics backs this up in The Language Instinct, “Latin declensional paradigms are not the best way to convey the inherent beauty of grammar”. He recommends computer programming and universal grammar on the grounds that they are “about living minds and not dead tongues”.

Pinker has a go at the Latin language mavens who want to pointlessly foist Latinate rules of grammar into English. As Pinker explains, this snobbery took root in 18th century London, when Latin was used as a mark of social class (still true today) and Latin grammar rules were crudely pasted into books on English grammar, for example, ‘don’t split infinitives’ and ‘don’t end a sentence with a preposition’. Latin simply doesn’t allow you to split an infinitive and to stupidly insist that it’s wrong in English, is as stupid as making us all wear togas.

English does NOT have Latin grammar. English grammar fundamentally Germanic. Latin highly inflected, role of word in a sentence indicated by its endings. English relies primarily on word order to determine grammatical relationships. Latin nouns gendered English are not.

5. LATIN is necessary for SCIENCE, LAW and MEDICINE

One girl in the audience, from Merseyside was adamant that Lawyers needed to have studied Latin. Incidentally, if you’ve heard the argument that Latin helps medical students learn and understand the considerable amount of medical vocabulary that has to be learned in medical schools. This also turns out to be false as shown in Pampush and Petto (2010). The Latin vocabulary in law, medicine and biology is there but one can simply learn the words, without knowing their etymology - most lawyers, doctors and scientists do exactly that. Why spend years learning Latin to help you in a small selection of vocabulary? The pay-off is, as Russell pointed out, is tiny.

6. LATIN brings the joy of ideas and literature

One contribution from the audience I did like was the idea that Latin bridges us to rich tradition of thought and literature. He mentioned Roman literature but also the Magna Carta, Bacon and Newton. First, anyone studying history will not lose out by working with translations of the Magna Carta or Bacon. And does anyone really need to read Newton’s tortuous Latin, other than scholars in the history of maths? I think not.

And why not Greek? Wouldn’t you prefer the riches of Plato and Aristotle, Homer’s Iliad and Odyssey, along with the works of any one of Aeschylus, Sophocles, Euripides and Aristophanes to the largely derivative Roman philosophers (few can name any) and dramatists. Even in History, the Greeks Thucydides and Herodotus are a match for any Roman writer, including Tacitus. In politics, our democratic traditions are largely Greek. Even English law is not Roman (although in Scotland it is). Then there’s the politics and democratic traditions that are fundamentally Greek. Even in maths and science the mighty Pythagoras, Euclid and Archimedes trump the Romans.

7. LATIN’s history of exclusion

Of course, Latin was introduced to this country as the language of the clergy, it was not, as is sometimes supposed introduced by the Romans, as few Latinate words come from that era. As the language of the church it largely excluded the laity, as most remained illiterate and spoke various forms of English. It was then used as the gatekeeper for learning. This had some benefits, when Latin was sort of European Esperanto, but continued for centuries after that died and was long used as the gatekeeper at Oxbridge and other institutions. David Aaronovitch made the telling point that it is still a key ‘marketing’ differentiator for independent schools.

8. LATIN would NOT die if not taught in schools

There are plenty of scholars in subjects that are not taught at schools. The bottom line with any dead language, especially Latin, is that there’s little that is new and to be uncovered. Compare this to the vast amounts of Sumerian cuneiform tablets that still need to be both deciphered and excavated. In the end I agree with one of our greatest living Latin scholars Mary Beard, “the overall strength of the classics is not to be measured by exactly how many young people know Latin or Greek from school or University. It is better measured by asking how many believe that there should be people in the world who do know Latin and Greek.” This about sums up my position.

I am not against the study of Latin or any other historic languages. This is largely a matter of proportionality for our Universities. By all means let a few study Latin. What I am against is too prominent a role for Latin in contemporary school curricula. Our young people have enough on their plate at 5-18, as the range of subjects expands to include a wider range of science subjects, IT and other vocational skills. A dead language, in the sense of no longer being spoken or used for scholarship, at this stage is merely the dead hand of educational history being played out by interested parties.

9. LATIN is NOT about choice

Several people argued that Latin is a matter of choice in schools. The curriculum is crowded enough with increasing demand every year as new subjects, such as computer studies, emerge. As we have seen from recent educational lunacy in policy, has resulted in the destruction of choice in the curriculum. Professor Alison Wolf, Gove’s lapdog, ignored advice from industry and education experts to crush 3100 vocational qualifications. That strange beast, the EBacc, which had Latin as a core choice, deliberately excluded all vocational subjects, even ICT. Gove's successors , such as Williamson did the same, with special temporary funding, creating a one-sided system that simply reinforces the old apartheid system we have in this country between academic and vocational learning.

This is not to say that all education should be aimed at utility and employment, just that a contemporary curriculum in schools is always a trade-off, with new subjects and content always arriving, so that everything we taught in the past can’t just remain. We need to weed and feed a curriculum, which means some hard choices. My preference would be to focus more on modern languages themselves, where we have declining numbers.

10. LATIN is a political issue

It has now become a political pawn. A largely (not exclusively) right wing movement to impose private school norms (7%) on state schools (93%) became an electoral issue, hence Williamson's special funding. Labour won and implemented their promised VAT policy on private schools and in reassessing budgets they also scrapped Williamson's artificial subsidy for Latin in state schools. We can argue for or against Latin in schools but this was democracy in action.

Conclusion

I rather liked this audience and I especially liked Peter Jones. He was my opponent but put to bed those old tropes about Latin improving your ability to learn languages, improve intelligence or think logically. He was remarkably free from cant and any sense of snobbery. What he loved was Latin and his plea was for the beauty and intrinsic value of the language. With that I have no argument.

AGI is here...

2024-12-20T22:19:00.013+00:00

As easy as one-two-three? No. Easy as one-three. OpenAI has skipped o2 to call their new ‘frontier’ model o3 (Altman hinted that this may have been due to Telco o2 issue). I like the idea that AI is undermining everything we were told makes good marketing - no read branding, messaging, just get it out there with benchmarks.

Social media was full of people saying Google had destroyed OpenAI. But their pre-Christmas release single was smashed off the Charts by OpenAI’s o3.

Transcending human intelligence

Not sure we've realised that we now have AGI. Arc-AGI solved with 87.7% (human threshold 85%) as well as other benchmarks... we will, of course, move the bar higher, find ways to avoid recognising the achievement... then pass it again... Arc Prize is a not-for-profit with an AGI benchmark. There are new skills so that the AI cannot memorise them.

To give you some idea of the speed and scale of progress:

Software engineering

On real-world software tasks, evaluations on o3 are more than 20% better than o1 models at 71%. AI has already established itself as solid, useful and widely used coding tool. People have been letting AI code then iterating iterate. This takes it to another level. One wonders how long the job of coder will survive at this rate.

Maths and science

Also superb at maths and PhD science questions at 88.7 %. A typical PhD gets around 70%. Frontier is toughest mathematical dataset – extremely hard problems. o3 has 25% accuracy which is a strong result.

GPQA Diamond

An interesting measure of how far we have come is GPQA Diamond. It's a clever test that compares a model against npvice Google search, human domain experts and the model. Experts get 81% right in their fields, highly skilled non-experts with 30 minutes per question and Google access get 22%. GPT-4 got 37% at the start of 2024. o1 got 78%. o3 is 87.7%. That is astonishing.

AGI

Artificial General Intelligence suffers from a problem of definition, as do most abstractions at this level. Does it mean:

Specific human reasoning skills (maths, science etc)

Average human competences

Greater than all of humanity

There is no singularity, there is a spectrum or constellation of possible targets here. What is clear is that these targets are being hit, not in one go, but one by one, sometimes in clusters. Maths and science are easy to measure but also fiendishly difficult to achieve, so this is a real milestone. Yet they focus on clear rationality. To be fair critics were telling us that AI would never get here, never mind get here so fast. We should celebrate this as many of the problems we face with climate, energy, healthcare and education may well be solved, not in the sense of final solutions but better solutions.

Problem solving in real life is messier and more of a challenge. That's why the agentic move is so interesting, as it tackles this set of human capabilities. We have brains that have evolved into a specific environment where we had to solve specific problems. This is where the dynmaic interrogative, dialogic nature of AI helps enormously. It has already made great strides in this direction.

Embodied AI, in the physical world of elevators, cars, cabs, trucks, drones, ships and submersibles has also taken great leaps. This is another set of targets that are being hit one by one. One could argue that neurological targets are also on our hotlist - Neuralink is a good example, where we enhance our neurological deficits.

AI may help us understand neuroscience and the brain, solve engineering problems to accelerate fusion, help with drug discovery and many other intractable problems. What we can be sure of is the increased impact of AI on productivity. The leaps in efficacy prove this.

Conclusion

This is a warning to people who claim that scaling is over. Sutskever was right – we have a way to go and other techniques are clearly delivering the goods. It is clear that AI is delivering faster than expected. The consequences of AGI are closer than expected with huge productivity gains on the horizon. That is the subject of the book I am currently writing.

Future issues

One issue needs discussion - compute costs. I think this will be solved. We saw a 250x decline in token costs in 20 months. So $20 for hard problems are likely to come down to cents.

One can now start to ask how the cost of compute compares to human costs in organisations for similar tasks and roles. The productivity game looks as though it will start with coding, where much of it can be automated.

On problems to solve:

fusion

medical research

personalised tutors

optimising political policies

next-generation batteries

cheap renewables

A final though on AI being self-generative. At what point does this technology start working on itself to solve the frontier problems and advance even quicker?

Fast technology and the slowcoach brain – fascinating paper

2024-12-20T15:24:00.001+00:00

The findings in this paper by Zheng and Meister 'The Unbearable Slowness of Being: Why do we live at 10 bits/s?' are quite startling when you consider the vast disparity between our sensory input capacity and our cognitive processing speed. It's almost as if we're living in slow motion within a world of high-speed data. This realisation challenges our intuitive understanding of human intelligence and efficiency.

Shaped by evolution

It's astonishing that despite our sensory systems being capable of absorbing vast amounts of data, our cognitive output remains so limited. The surprise at this slow processing suggests that perhaps our brain isn't designed for speed but for careful, deliberate processing which might be more suited to survival in our evolutionary context. Humans and other species might operate at rates sufficient for their ecological niches, suggesting evolutionary constraints. Human cognition tends towards serial processes, possibly due to evolutionary origins where multitasking was not advantageous.

Humans process information at an oddly slow rate of about 10 bits per second (bps). This contrasts starkly with our sensory input capacity of approximately 10^9 bps. Even expert typists operate at around 10 bps, English narration speed suggests an information rate of about 13 bps, blindfolded speed-cubing has a perception rate during the puzzle inspection phase of around 11.8 bps. Top-end memory tasks like binary digit memorisation or speed card inspection reveal rates from 5 to 18 bps, showing that even under high memory demands, the rate remains within the 10 bps ballpark.

Bottleneck

The bottleneck is not neural capacity, it is throughput. Sensory neurons, particularly photoreceptors, can handle gigabits of data, while the brain's behavioural output remains at 10 bps. The paradox is that, despite having neurons capable of transmitting information quickly, when it comes to human behaviour and cognition, we are slowcoaches. This limitation seems to be fundamental.

Photographic memory claims turn out to be false. Even with exceptional memory, the data acquisition rate doesn't significantly exceed 10 bps. Our perceived richness of visual scenes is an illusion, as actual perception is much more limited, especially outside the focal point.

Let’s think about this, albeit in a limited and slow fashion! Even with advanced technology like Neuralink, we'd still be bound by this 10 bps limit for cognitive throughput. This is a sobering thought. If our brain operates at such a slow rate, it might imply that human ‘intelligence’ is less about raw computational power and more about the quality and nature of information processing that focuses on depth rather than breadth.

Conclusion

Neurons are not inherently inefficient; rather, the brain's architecture seems optimised for serial processing over parallel. With AI we have several dimensions of scale, in terms of relevant data, retrieval and parallel processing. AI may be offering a way out of this human envelope, that of being fixed, linear and slow, to augmenting our abilities with technology that is flexible, parallel and fast.

Does ChatGPT enhance student learning? Meta-study with some surprises!

2024-12-20T14:25:00.013+00:00

Great to see a meta-study asking this bold question:

Does ChatGPT enhance student learning? A systematic review and meta-analysis of experimental studies.

See study here.

STUDY

Nice spread of subjects and focus on classroom teaching. Predominantly university-level interventions (84%), with a smaller focus on K-12 (14.49%), language education is most studied (32%), followed by computing, health, physics, and education. Most of the interventions occur in classroom environments (87%) with a duration of 5 to 10 weeks. The AI was primarily used as a direct learning tool (80%), with some integration into broader educational platforms.

RESULTS

The results were, I’s say, quite startlingly positive. Many are decrying this technology as something that depresses academic performance, demotivates students, and blocks higher-order critical thinking. The study suggests the opposite.

Academic Performance

Significant improvement with ChatGPT use, though concerns about whether this is due to the intervention or the quality of AI-generated content.

Affective-Motivational States

Positive effects on motivation and emotional engagement

Higher-Order Thinking Propensities

Positive influence on self-reported propensities.

Mental Effort

Significant reduction, suggesting less cognitive strain with ChatGPT use.

Self-Efficacy

No significant impact observed.

The study covers a broad spectrum of educational contexts and subject areas, providing a pretty wide view of ChatGPT's application in education. I’ll leave the methodological rigour questions for those more qualified but I liked the ambitious approach to examining a range of learning outcomes, offering insights into some key dimensions of student learning. I was surprised that self-efficacy was not positive – interesting finding as it runs against the affective-motivational finding.

Sure there are weaknesses; some small sample sizes, sometimes little pre-testing and one odd one - ChatGPT sometimes used in assessments. But this is a good start.

Reclaiming Productivity: Aligning Work, Learning, and Societal Needs

2024-12-15T12:04:00.003+00:00

‘Productivity is being able to do things that you were never able to do before,’ said Franz Kafka. That’s pretty much the same as learning. I’d like to see education and workplace learning be more aware of productivity as a worthy idea in this wider sense of the word.

Nobel Prize-winning economist Paul Krugman also said, “Productivity isn't everything, but in the long run, it is almost everything”. He has a point. Producing autonomous and productive adults is surely what we aspire to as a collective goal of a nation. A country's ability to improve its standard of living depends on its ability to raise its output per worker. Isn't that what we want from L&D?

There is a sense that we’ve lost something; people not caring, huge numbers not in employment, yet massive skills shortages. In the UK we have an economically-inactive group of 9 million working-age people. Around 1.5 million are unemployed, about 2.7 million are students (here, I’d argue, we have too many staying on for too long in HE). We have 2.3 million ‘sick’, which is shocking and 1.5 million as carers. We should be tackling this head on.

We have an education system way out of kilter with societal needs. A realignment of education to life and a living would help. The looking down upon, defunding and eradication of any vestige of vocational learning from curricula in schools and Universities has not helped. We want more houses but don’t have the skilled people to build them; green energy but have a generation who can hold placards, sit in front of traffic and throw paint at art but can’t actually do anything practical to build the green technology we need; more nurses but have turned it into a graduate profession disenfranchising working class entrants.

In workplace learning we need to stop distracting ourselves with abstractions. The 'Skills-Based Organisation' has long been an empty trope, because we have been seduced into thinking that abstract nouns like leadership (lots of spend but so little of it), culture, diversity, equality, values, inclusion, resilience etc are 'skills' or some mysterious miasma that will encourage and produce skills.

The very idea that skills have anything to do with text-based learning and assessment or sitting around tables scribbling on pieces of flipchart paper is laughable. A renewed focus on productivity would help but we're not much interested in actual outputs, only abstract inputs. Training so often kills productivity by distracting people from reality. People love a cosy chat with others who agree with them. But this is so often a zero-sum game - the net gain in terms of change, skills, productivity is zero. In fact it may be a negative-sum game as it takes up tons of time, costs and tends to fix people into old, fossilised behaviours, meaning absolutely nothing changes or improves.

Productivity is the ratio of output to inputs, which can be applied to people, not cruelly or mechanically but sensitively. Yet we flee from evaluation and measurement as we’re scared of what we may find. Productivity can be difficult to measure and improve but it translates into more satisfied employees a better standard of living for all and the crazy idea that your kids will be better off than you.

Astounding talk by Sutskever

2024-12-15T09:11:00.004+00:00

A 30 minute talk given yesterday by Ilya Sutzkever already being seen as seminal.

We have gone through a decade of astonishing advancements in neural network learning. We were right to equate an artificial neuron with a human neuron. Biologically inspired AI has been hugely successful. It led to huge success through scaling. We forget that we were using crude string matching, tables and statistical techniques. We worked in this world and from 2014 were successfully producing real product. But we knew its limitations.

PEAK DATA
Then came GTP 2.0 and LLMs. These have been astonishingly successful. On speed alone they do what takes humans ages to complete, they are also often better on quality. But Sutskever’s big message was that pre-training will end due to data limitation, peak data, the fossil fuel of AI.

What do we do after pre-training? Agents, synthetic data and OpenAI o1 are examples of attempt to move forward. He draws from biology showing the relationship between body size of mammals and their brains. Hominids have an accelerated brain to body ratio. We are different. But so is AI, which can also be different. This is evolution on steroids. What took billions of years to achieve through evolution to hominids, is being done in just a few years.

SUPERINTELLIGENCE
Progress in AI has been remarkable but he thinks we are heading towards ‘superintelligence’. At the moment we have amazing systems which in many areas (evaluated) exceed human capabilities. They still get a little confused at times, who doesn’t. Hallucinations happen but models will eventually autocorrect. This may already be happening.

We have lots to gain from current AI, as it is so incredibly useful. Even if we stopped right now there is so much that can be done with the current technology to keep us busy for a decade. It is a productivity boost in may areas of human endeavour as well as being a discovery tool.

Systems are only very slightly agentic but agency, reasoning and self-awareness will happen – why? They are useful. The human brain, stopped growing in size but we got stronger cognitively, so humanity kept advancing. Similarly, the agents and tools on top of LLMs will fuel the progress through agentic behaviour. This future AI will possess unpredictable capabilities and self-awareness, transforming their capabilities.

FUTURE
He left us hanging and is very careful not to become too flippant and certain of himself, like some silly, keynote futurist. He has worked on the leading edge of this technology for decades and has contributed hugely to the field. While careful about predicting the exact nature of future development, which he thinks we cannot predict with certainty, he nonetheless, paints a picture of rapid progress and achievement.

Tyranny of Text: Education, Work, and the AI Revolution

2024-12-09T12:58:00.003+00:00

I have just visited the Ramesseum, and the statue that inspired Shelley’s ‘Ozymandias’. This part of the world seems cursed by being the place where ‘writing’ was first invented, religious books written, each granting different groups a sense of stupid, eternal entitlement. They’ve been at each other’s throats ever since. When people dream of text-based heavenly Paradises, they can't see worldly Paradises in front of their eyes.

We are drowning in a sea of 'text' in learning. From 5 to now 25, young people spend almost all of their time reading, writing and critiquing ‘text’ in an educational system because it is easy, creating the illusion that you can assess most skills through text - you can’t.

We have decimated vocational learning by sucking up funding into often purely text-based Degree subjects. Lecturing is easy, teaching is hard. Setting essays is easy, assessment is hard. Producing text is now easy, doing things is still hard. Once you see LLMs as producing text as calculators produce good numerical solutions, you relax a bit on AI.

Less is more

AI is not about generating more text. Its true purpose has been in generating LESS text; summarising beautifully, producing grammatically perfect text with no spelling or punctuation errors. Most text is not in books and essays but in mundane communications, as emails, messaging and social media – and AI is largely an aid to communication. In work and life, communications is often over-long, badly written and error prone. That problem is being solved.

Young people are quite adept at short, concise messaging, they do it all day every day. The problems come from people fed on a diet of long-form text, who tend to see everything as a potential essay, so in organisations and government bureaucracy, which existed long before AI, is text-heavy, the production of unnecessary forms, documentation and reports. AI will optimise and automate this.

Learn by doing

Most of work and life is about doing, most education is about writing. Yet much text production in white collar work may well be automated out of existence. Let’s recognise that AI is now fundamentally multimodal, using speech, images and video, also creating 3D worlds, robots and automated vehicles. This is not to eliminate text, just see it as not primary and over-egged in education, work and life. With multimodal AI and robotics, AI is moving fast into the real of doing, both teaching us how to do things and doing things for us and in place of us. The lines are now blurring, with increased scepticism over text-only education. Education is for both life and living.

AI may have seemed like a text-based phenomenon, but it is proving to be more about communication, speech, robots, data and automation. We will be made more productive by having less text, automating as much as we can out of the system, dissolving text-laden bureaucracy. Most of the critics of AI come from those who deal with text as a living, in education and work.

PODCAST on Invention of writing
PODCAST on Literacy & Orality

How do emotions impact learning and mental effort? Maybe not as we think

2024-12-03T07:14:00.003+00:00

I’m deeply suspicious, not of the claim that emotions are important in learning, but of the assumption made on the back of this, that emotion is quite simply a good thing in learning. It leads to all sorts of wrong assumptions about fun, gaming and no end of odd theorising about happiness.

Like the shallow side of social constructivists, who simply conclude that all learning should be ‘social’, whatever that means, there is a tendency to see ‘emotion’ as a good thing, no matter what. This problem has been exacerbated in recent times by the therapeutic assumption that emotions are somehow intrinsically virtuous.

New study

So this study caught my eye (thanks Carl Hwenrdrick) as it digs into something I had always thought may be true, that positive emotions don’t work the way you think.

https://www.sciencedirect.com/science/article/pii/S1041608024001900

While feeling good (like being happy or motivated) definitely helps with learning, it doesn’t seem to do so by lowering the mental effort (via cognitive load) needed to process new information. This challenges earlier ideas that positive emotions expand your mental resources.

Emotions aren’t set in stone and shift while you learn. For example, frustration at the start of a tough task can turn into satisfaction as students figure things out. So, frustration isn’t always a bad thing—it can be part of the process. Emotions also shape cognitive load. We usually think of cognitive load as tied to how complex the information is, but this study points out that emotions directly influence how much mental effort a task feels like it takes. Finally, not all positive emotions are equal. While being in a good mood generally helps, too much excitement or overconfidence can backfire. Students might rush through material, miss details, or oversimplify because they are ‘feeling too good’. This last point is important.

Emotions and learning

Affective learning deals with the emotional side of learners, their emotions and feelings. These feelings cover a wide range of positive and negative attitudes, interests, beliefs and motivations before, during and after learning.

Teachers, lecturers and trainers are professional learners and often understate the role that emotions play in learning. Yet speak to any learner and many learners will report not what they achieved in learning but how they felt. Few get through school without feeling bored or indifferent to lessons and subjects that seem dull, remote and irrelevant. Fewer still get through a degree without feeling numbed in boring lectures. On the other hand, successful learners report excitement, engagement and feelings of pride and achievement. The point is that this can go both ways.

This complex world of feelings and emotions is often sidelined by the dominance of the purely rational, academic cognitive side of learning theory. This is partly down to the dominance of Bloom’s silly taxonomy, the cognitive domain being only one of three, the other two the psychomotor and affective are often completely ignored.

Kahneman

Kahneman posits the idea that we have two brains, in Thinking Fast and Slow; System 1 - fast, emotional and instinctive, also System 2 - slower and rational. I am no longer convinced that the distinction is as clear as we think it is (sometimes expressed as elephant and rider). Our brain has substantial weaknesses, due to its long and messy evolutionary history. We know that it stubbornly procrastinates, fails to remain attentive (attention being a necessary condition for most learning) and easily distracted. It is also subject to emotional pulls and mood swings, even depression. This is both a blessing and a curse. The emotive dimension of learning is often underestimated but it can also distract and over-stimulate.

Krathwohl

Everyone knows Bloom, but we hear little about the man who completed Bloom’s work in the affective domain, the less known David Krathwohl. Although Bloom's original taxonomy consisted of six categories, when Krathwohl revised it in 2001, he put emphasis on the interaction between the cognitive and affective. With Lori Anderson he also helped reduce Bloom’s cognitive taxonomy down to four categories or knowledge dimensions:

Factual knowledge

Conceptual knowledge

Procedural knowledge

Metacognitive knowledge

For each of these four, smaller dimensions were identified. He also changed the cognitive processes to verbs and renamed Evaluation and Synthesis as Creation.

Krathwahl then proposed six levels of affective learning:

Characterization

Organization

Valuing

Responding

Receiving

My own view is that this Affective Taxonomy suffers from the same hierarchical rigidity as Bloom’s taxonomy in the Cognitive domain. It is far too rigid and hierarchical. Some even argue that there is no real taxonomy of affective learning as it emerges from or is part of the cognitive domain. Affective factors are also difficult to identify and assess as they involve feelings, attitudes, and beliefs, so ignored as something difficult to measure, vague and unimportant. While there is recognition that feelings and emotions play a strong role in motivation and learning, it is rarely be seen as being on a par with its cognitive counterpart.

There is certainly the tendency for schools and academia to focus on text-based, pure reason, as their primary skillset, at the expense of other aspects of learning. This has led to a paucity of research in the area.

Speak to workplace trainers or sports coaches and you will hear far more about affective learning, as it really does matter. These are ever present in learning and can also be internalised, either to hinder learning or harnessed and used positively by the learner to move forward. So feelings play a strong role in both demotivation and motivation. Understanding their role is essential if you are a learning professional, yet few could name a single theorist in this area.

Panksepp

Jaak Panksepp introduces the evolutionary origins of emotions and warns us that although emotions are vital in learning, they can also hinder learning. Panksepp saw life as being empty without emotions, emotions being survival features, as part of our evolutionary heritage. We do not teach or learn these seven PRIMARY affective systems, as they are innate:

SEEKING (expectancy)

FEAR (anxiety)

RAGE (anger)

LUST (sexual excitement)

CARE (nurturance)

PANIC/GRIEF (sadness)

PLAY (social joy)

We can learn from these emotions, but we do not learn them, only learn to modulate them, He did think that they formed the basis of our personality, different emphases producing different personality types. SECONDARY emotional processes are learnt through classical and operant conditioning and TERTIARY emotions are sensory (taste, pleasure, pain) and homeostatic affects (hunger and thirst).

We, unlike animals, are cognitive creatures, but he regrets the common disregard of emotions and our evolutionary heritage in understanding the foundations of learning and higher cognitive processes. Much of what is presented in traditional learning theory, whether rewards, punishments or reinforcements actually rely on the emotional responses of the brain. Yet emotions are a double-edged sword.

Some emotions, such as RAGE, FEAR and PANIC are not conducive to learning and may inhibit or hinder learning. On the other hand, learning may benefit from the SEEKING emotion, with its feeling of enthusiasm, as it is instinctive for survival, it promotes learning through purpose, anticipation and curiosity. Its absence diminished a disposition towards learning.

Damasio & Immordino‐Yang

Damasio & Immordino‐Yang see emotions and reason as entwined or enmeshed. They not only not only regulate our lives, they regulate learning. Emotion is therefore critical to learning and memories well as playing a powerful role in learning as motivators.

Conclusion

A complex area but we must be careful of being too shallow on our consideration of emotions. It may be that acts of learning or thinking induce emotions, not that emotions are always the well spring for learning. It is also clear that they may limit, cap or damage learning. We must keep a close eye of the detailed research in this area, rather than trite statements about the important and efficacy of learning.

Do DEI programmes need a reality check?

2024-11-26T12:27:00.001+00:00

You may feel uncomfortable reading this, as it is a contentious area. But contention must not be left to fester, sides becoming increasingly polarised.

Companies and educational institutions are pouring billions into Diversity, Equity, and Inclusion (DEI) initiatives. But do they work? DEI has to face up to the evidence, and there is a case for rethinking diversity training. Are we getting it all wrong with billions spent, but does DEI deliver? Mandatory training seems to have a backlash problem and unintended consequences. Turns out, the evidence does not support this spend. Could this training be hurting more than helping and do they need a reality check?

PAPER

This paper by the NCRI points out some serious flaws in some DEI approaches, and adds to the growing body of larger and wider studies that came to the same conclusion.

They mostly have no significant effect as they focus on the wrong thing – training not process. Indeed, some DEI programs stir up resentment and suspicion, making things worse rather than better.

Mandatory training has led to a backlash, as Forcing people into diversity training can create discomfort and even hostility. Nobody likes feeling blamed and you donpt win people over by accusatory tactics.

‘Anti-oppression’ rhetoric tends to polarise people as it seems explicitly political. Materials (identified in the paper) that emphasise systemic oppression and victimhood often make participants view others through a hostile lens.

Another danger is that perception becomes reality, where DEI narratives lead people to see prejudice and unfairness in neutral situations. This exposure to anti-racism or anti-Islamophobia narratives can increase participants’ willingness to punish perceived ‘oppressors’ without clear proof of wrong-doing.

There is also an interesting discussion in the paper surfacing DEI narratives to controlling tendencies. Some DEI content appears to foster controlling and punitive mindsets, authoritarian traits of authoritarianism This is interesting, and I true to a degree.

CONCLUSION

I’ve been writing about this for nearly 20 years and this is yet another paper that calls for an honest evaluation of this gargantuan spend and shines a much-needed light on the unintended consequences of many DEI approaches. It’s shocking that so much money is spent on DEI without robust evidence of effectiveness or harm mitigation. Evidence like this should be a wake-up call for organisations to rethink how they handle diversity training—but it won’t. Why not? We are in an evidence-free zone, where even discussion of the evidence cannot happen. That’s why The New York Times and Bloomberg shelved coverage

EVIDENCE

There is a consistent theme here - mandatory, accusatory, or one-off DEI efforts can do more harm than good. The key is building trust, creating buy-in, and focusing on systemic change rather than short-term fixes. On that front I highly recommend Simon Fanshawe’s excellent book ‘The Power of Difference’. He was a founder of Stonewall and has fought for Gay rights all of his life.

An early study, Kidder et al. (2004), tackled the backlash that DEI programs can trigger when framed as affirmative action. Turns out, calling it affirmative action often sets off alarms about reverse discrimination, making people feel unfairly targeted. Instead of promoting diversity, it can lead to resentment, undermining the very goal of the programs. This research was one of the first to flag the danger of how DEI is framed.

Then along came a big one Kalev, Dobbin, & Kelly (2006), who found that mandatory diversity training doesn’t just fail—it can actually backfire. Employees often resist being told they have unconscious biases, which can foster defensiveness. The researchers suggested that voluntary training or pairing it with programs like mentorship schemes works far better, making employees feel part of the solution rather than the problem. Legault, Gutsell, & Inzlicht (2011) looked at anti-prejudice messaging and found something surprising: when people feel pressured or coerced into thinking a certain way, they dig their heels in. Instead of reducing bias, these messages can actually make it worse. Giving people more autonomy in how they engage with DEI efforts might be a smarter move. Fast forward a few years, to Moss-Racusin et al. (2016) who found that programs dangling carrots or sticks to enforce diversity goals often fail. People respond better when their motivation comes from within. The takeaway? If you want people to embrace DEI, make it about shared values, not external pressure.

Yes it’s Dobbin & Kalev (2016) again, showing why so many DEI programs flop. Surprise, surprise, forcing people to participate doesn’t help. In fact, mandatory sessions can spark resistance. Instead, they found that voluntary programs, mentorships, and leadership support yield better results by creating buy-in and trust. Leslie (2019) shines a light on unintended consequences. Leslie found that bias-focused training can backfire by reinforcing stereotypes and creating perceptions of unfairness. The lesson? A clumsy approach to diversity can end up dividing people rather than bringing them together.

More recently we have a meta-analysis is a big deal because it pulled together loads of research on DEI training by Moss-Racusin et al. (2016). The verdict? There’s not enough evidence showing these programs work, especially the quick-fix ones like single seminars. Long-term, systemic change is what’s needed, not just ticking the “we did training” box. Iyer (2022) then dug into why people oppose DEI programs, especially those from so-called “advantaged” groups. The big issue? They often feel threatened or unfairly targeted. Ignoring their concerns only fuels resistance, so a more inclusive approach that engages everyone might be the way forward.

Burnett & Aguinis (2024) pointed out that DEI training aimed at specific groups can lead to discomfort and defensiveness, which defeats the purpose. They suggest reframing these programs to be less accusatory and more about shared goals to avoid making people feel unfairly singled out.

Bibiography

Kidder, D. L., Lankau, M. J., Chrobot‐Mason, D., Mollica, K. A., & Friedman, R. A. (2004). Backlash toward diversity initiatives: Examining the impact of diversity program justification, personal and group outcomes. International Journal of Conflict Management, 15(1), 77–102. https://doi.org/10.1108/eb022908

Kalev, A., Dobbin, F., & Kelly, E. (2006). Best practices or best guesses? Assessing the efficacy of corporate affirmative action and diversity policies. American Sociological Review, 71(4), 589–617. https://doi.org/10.1177/000312240607100404

Legault, L., Gutsell, J. N., & Inzlicht, M. (2011). Ironic effects of antiprejudice messages: How motivational interventions can reduce (but also increase) prejudice. Psychological Science, 22(12), 1472–1477. https://doi.org/10.1177/0956797611427918

Moss-Racusin, C. A., van der Toorn, J., Dovidio, J. F., Brescoll, V. L., Graham, M. J., & Handelsman, J. (2016). A “scientific diversity” intervention to reduce gender bias in a sample of life scientists. Cultural Diversity and Ethnic Minority Psychology, 22(3), 295–306. https://doi.org/10.1037/cdp0000070

Dobbin, F., & Kalev, A. (2016). Why diversity programs fail. Harvard Business Review, 94(7-8), 52–60. https://hbr.org/2016/07/why-diversity-programs-fail

Leslie, L. M. (2019). Diversity initiative effectiveness: A typological theory of unintended consequences. Academy of Management Review, 44(3), 538–563. https://doi.org/10.5465/amr.2017.0087

Paluck, E. L., Porat, R., Clark, C. S., & Green, D. P. (2021). Prejudice reduction: Progress and challenges. Annual Review of Psychology, 72, 533–560. https://doi.org/10.1146/annurev-psych-071620-030619

Iyer, A. (2022). Understanding advantaged groups’ opposition to diversity, equity, and inclusion (DEI) policies: The role of perceived threat. Social and Personality Psychology Compass, 16(4), e12666. https://doi.org/10.1111/spc3.12666

Burnett, L., & Aguinis, H. (2024). How to prevent and minimize DEI backfire. Organizational Dynamics, 53(2), 100981. https://doi.org/10.1016/j.orgdyn.2023.100981

Fanshawe, S. (2021). The power of difference: Where the complexities of diversity and inclusion meet practical solutions. Kogan Page Publishers.

Acronyms, alliteration and absurdity: the sad truth about Organisational 'Values'

2024-11-23T12:27:00.004+00:00

I had an epiphany some years back. In a brief conversation with a young woman, in the queue for lunch at a corporate ‘values’ day (I was a Director), opened my eyes up to the whole values thing in organisations.

“I have my values,” she said, “and they’re not going to be changed by a HR department.... I’ll be leaving in a couple of years and no doubt their HR will have a different set of values… which I’ll also ignore”. Wisest thing I heard all day.

You’ve probably had the ‘values’ treatment. Suddenly, parachuted out of HR, comes a few abstract nouns, or worse, an acronym, stating that the organisation now has some really important ‘values’. Even worse, an expensive external agency may have juiced them up. I genuinely like organisations that have a strategy, purpose, even a mission. But the obsession with organisational values I just don't buy.

Not long afterwards my fears were confirmed. I chaired a Skills Summit, where innumerable HR folk pompously paraded their company values in a series of talks. An endless stream of abstract nouns, all of which seemed like things any normal human being would want in any context, in or out of work - you know the words - integrity, innovation, honesty, customer-focus, community.... After a full day of this stuff I was impressed by the guy who ran a small, very successful software company, who stood at the podium, and claimed that his company didn't really have any stated values and felt that the whole 'values' thing could be replaced by one phrase 'Don't be a dick!". That became a meme some years back and touched a nerve because it cut through the hubris.

Why HR-driven values are out of touch with reality?

Having dealt with hundreds of large organisations for more almost 40 years, I have yet to find one whose values were anything more than platitudes. They are invariably a crude mixture of reactive PR, HR overreach even a marketing ploy. Usually a crude selection from a list of abstract nouns, often forced into an idiotic acronym, they bear no relation to reality. Even when masked by complex consultancy reports and training - it's almost always bullshit Bingo.

Why would we imagine that HR have any skills in this area? In what sense are they 'experts' in values? For me, it is a utopian view of work and organisations. I can remember the day when organisational 'value' lists never existed. People were more honest and realistic about expectations. They came in when HR suddenly decided that they had to look after our emotional and moral welfare - always a ridiculous idea.

Values-washing

The banks were full of this 'values' culture – that was before the financial crash. I worked with many of them. It was all puff and PR. People do not, and don't, buy into this stuff. They can barely recall what the values are. I have values and I'm not interested in what HR, or some external consultant, says my values should be. The even more ridiculous idea that people who don't adopt those values should be forced out is wrong and illegal.

They shove them on the website but few remember them and even fewer care.... The really interesting thing about 'values' is that those companies who feel most compelled to get them identified - banks, accountancies, consultancies, tech companies, pharma companies etc - are the very companies where they were most ignored. They are blatant attempts at value-washing, appearing to be value driven when you are not.

Try these authenticity tests to your company values. Sniff out the hubris and bullshit.

Test 1: Bad acronyms - values created to fit word

If your values set is an acronym, they’re almost certainly inauthentic. The net result of fuzzy HR thinking is so often the ‘bad acronym’. Chances are that someone has shoehorned some abstract nouns into a word that sounds vaguely positive, completely losing sight of the original intention. Are they telling you that their values ‘just happened’ to fall into that acronym? Actually, what happens is that at least some of the values emerge from the acronym. That's bullshit.

How about this from a Cheshire voluntary group: FLUID - Freedom 2 Love Ur Identity. Or another real example of a crap acronym: VALUE - this HR person actually went online as she could only think of Value Added….. and wanted others to fill the acronym out! They did, and she was delighted with, Value Added Local, User friendly Experience. What a load of puff.

When values are created to fit a word you are engaging in an infantile exercise that treats employees like children. Even worse is the use of middle letters, rendering the acronym, as an aide memoire, completely useless. Here’s a real example. It’s a cracker. PEOPLE: Positive Spirit and Fun, HonEsty and Integrity, Opportunities Based on Merit, Putting the Team first, Lasting value for Clients and People, Excellence through Professionalism. One overlong, impossible to remember acronym with eleven nouns, and I love the way they have to use the ‘E’ in the middle of HonEsty to make it work. This, by the way, is from an HR consultancy.

It’s not that I hate acronyms. They’re great as memorable cues. For example, I rather like ABC (Airways, Breathing, Circulation) in first aid. I also have a soft spot for funny acronyms, such as ALITALIA (Airplane Lands In Turin And Luggage In Ancona), BAAPS (British Association of Aesthetic Plastic Surgeon) unbelievably a real organisation, and DIMWIT (Don't Interrupt Me While I'm Talking).… it’s just that I’m a fully paid up member of the AAAA, the Association Against Acronym Abuse. And let's just quietly forget Microsoft's 'Critical Update Notification Tool'.

Test 2: Alliteration test

You hear alliterative value lists all the time - 'Imagination, Integrity, Innovation' (two organisations I know have this one set, clearly having cribbed it from the internet, or lists of 'C' words such as creativity, curiosity and collaborative. These are far too conveniently alliterative for my liking. The world is not intrinsically alliterative and if your list of values all start with the same letter - it's forced nonsense.

Test 3: Negative test

Lists of values are often so obvious that they are hardly worth mentioning. Sure, you can say we all need to be 'Customer friendly' and so on. But who would say that being Customer unfriendly was ever on the cards? The ‘negation’ test is a useful filter. Ask whether any normal human being would deny having the stated opposite or negative value. If the answer is invariably NO, as it’s not a value but a basic, common sense belief. Human nature is a complex thing and people are too different to be corralled into value sets. Beware of BIG words like integrity, imagination, creativity, innovation…… if your values are abstract platitudes – no one will care.

Test 4: Are they really values?

A value is something that determines a moral decision. Yet many organizational ‘values’ are not values at all. ‘Imagination’, for example, is not a moral value, neither I would argue is 'creativity'. I’m not sure that ‘Leadership’ is an intrinsic value, in the sense that Pol Pot was a leader. So, for this test, look at each value in turn and ask whether it really is a value or activity, competence or some other thing?

Test 5: Diversity problem

There’s something odd about having diversity as a value within a non-diverse, fixed value set. Empirically, people have different sets of values. We know this from large-scale studies, such as the World Values' Survey, going since 1989, in over 100 countries. An organisation is likely to have a mix of nationalities and cultures; religious, secular, liberal, conservative, individualistic, communal. Imposing a single set of values from above may not fit with this diversity of cultures and values. If diversity of values matters, the imposition of a set of fixed values makes little sense.

To practice diversity is to live with a diversity of values. At the Skill Summit, some companies seemed to imply that if you didn't fit in with their imposed values, they'd try to get you out. Really? When values become reasons to sack people, you've got to worry. Even the phrase 'Don't be a dick' worries me. Companies often have dicks in the workplace. So what? Lots of very competent and talented people are 'dicks'. Elon Musk is a dick. Steve Jobs was a dick. Gates was a dick. Get over it. We're all different.

Test 6: Sniff test

It is usually quite easy to expose the hypocrisy of corporate values, namely the

hypocrisy of an organisation that exhorts ‘values’ by looking at its a) tax affairs b) senior staff salaries, c) senior staff bonuses d) customer list e) behaviours. If the company plays the tax avoidance game using offshore tax arrangements, or transfer pricing – that’s almost every large tech company, Google, Apple, Amazon, Starbucks etc. etc. then add hypocrisy to their values. If the CEO earns a ridiculous amount of money but doesn’t pay a living wage to the people at the bottom, the value of their values is nil. To be more precise, if your company pays the CEO way more than x10 the salary of the lowest member of staff – question the values. If, as a bank or other organisation, you’ve mis-sold, ripped people off and generally fiddled the markets, ripped off suppliers, don’t pay on time - don’t even mention values.

Read Nagel's Equality and Partiality. It doesn't take long to work out that stated public values are often different from personal values. The same with organisations. You get the idea. Subject your organisation to a sniff test. Take the values and really ask – of the people who have told you that they matter – whether they’re applied at the top of the organisation and in its financial dealings.

Conclusion

In truth, everyone knows that values are actually marketing exercises, used by organisations as slogans. They have little to do with actual behaviour in organisations. They infantilise people, reduce them to ciphers. Ask the person in the street if large organisations have served society well in terms of values? Banks? Supermarket chains? Tax dodging tech companies? Tax dodging retailers? Football organisations like FIFA? Sports organisations? Political parties? Energy companies? No.

We have a crisis of trust in institutions because people parrot values which they don’t then practice. The ‘values’ obsession is just another example of overreach by HR. It keeps them occupied and gives everyone the sense that moral purpose has been served. It may even mask the reality of controlling behaviour. When I hear people discuss values, or see ‘values’ training, I hear moralising. Lots of back-slapping and ‘aren’t we great’ type platitudes. We’re all different. It is the workplace not a moral crusade.

Forget the buzzwords, the brutal truth about company values is that a select group at the top come up with 'values' and we all have to march in step to those values, even though, as most of us know, the further up an organisation you go, the more rarified values become.

Groucho Marx said "I have values and if you don't like them.... I have other values" and if asked whether I change my values if they are not the same as my employer, I have argued with many people for many years that employers and HR have no right to do this. The answer is NO. It's an excuse for one group to impose their personal views on another and is causing untold damage in organisations.

People have values, organisations don’t.

20 ways to use NEW Google Video analysis

2024-11-17T13:20:00.013+00:00

Google’s Gemini-Exp-1114 has launched and is getting rave reviews. It beats OpenAIs GPT-4o model hands down.

VIDEO

The BIG new feature is the analysis of video. You can upload video for analysis from your Google drive and off it goes.

Examples include analysis of:

STUDENT PERFORMANCE

Presentation performance
Eye tracking in teaching surgery etc
Coaching on sports performance
Feedback on musical performance
Teacher/lecturer/trainer performance
Compare present with past performance

STUDENT LEARNING

Lecture summarisation
Image, graph, diagram analysis into notes
Generate practical tutorials from tasks/procedures
Vocational tutorials
Pick out key clips
Create branched scenarios from video

TASK ANALYSIS

Behavioural tasks analysis from real world actions
Workflow efficiency - redundant steps in task execution
Performance analysis of meetings
Highlighting health & safety in real world environments

ASSESSMENTS

Assessments created from a lecture
Create video-based assessments
Assessment of student presentation of assignments
Identify specific errors made when doing something

You get the idea.

BREAKTHROUGH

It may lift us out of the current world of teaching and learning, incredibly limited by its focus on ‘text’. We are drowning in a sea of text, yet we know it does little in teaching an assessing actual skills and performance.

MODEL PERFORMANCE

Gemini-Exp-1114 has knocked OpenAI off the top of the Chatbot Arena leaderboard, the de facto goto place for benchmarking. It scored particularly well in Maths, Creative Writing, Longer Query, Instruction Following, Multi-turn and Hard Prompts. This will also help with the video analysis.

LONGER QUERY and Instruction following is a query, prompt or instruction that includes a detailed and extended description of what the user wants the AI to do or generate, allowing you to get the model to follows specific steps, formats, or guidelines in its response.

MULTI-TURN is dialogue with the AI that spans multiple exchanges (turns), where each response builds on the previous ones, remembering prior questions and answers to provide more coherent responses.

HARD PROMPTS are prompts that are intentionally complex, ambiguous, or challenging for the AI to interpret or execute correctly. It is here you expose the model's limitations, so useful in evaluating its boundaries.

CONCLUSION

The competitive race is producing faster better models but more importantly, a new layer of multimodal functionality is making AI far more usable. All of the above show that the new models are improving practical usefulness. AI is moving away from just a model release to something that gives better practical results, especially in education and learning.

Is AI generated video as GOOD as real video in learning?

2024-11-14T08:31:00.005+00:00

There are many surprising things we can learn from research into video and learning. I have given many talks on the subject showing research on video and memory (the transience effect), does learning at x1.5 or x2 affect learning? Do segmentation, length, perspective, picture quality, audio and so on affect learning? Here are 15 THINGS that may shock you from the research… some will surprise you!

But is AI generated video as good as real video in learning?

Leiker et al (2023) in Generative AI for learning looked at this hypothesis.

The study took 83 adult learnersn randomly assigning them into 2 groups:
1. Traditionally produced instructor video
2. Video with realistic AI generated character
Pre and post learning assessment and survey data were used to determine what was learnt and how learners perceived the two types of video.

NO SIGNIFICANT DIFFERENCES
No significant differences were found in either learning or how the videos were perceived. They suggest that AI-generated synthetic, talking head learning videos (limited) are a viable substitute for videos.

This doesn't surprise me. I’ve been creating avatars of myself at increasing levels of fidelity in appearance, movement, lip-synch & voice, speaking many languages from Chinese to Zulu. This involved going into a studio for video capture and separate audio studio for voice capture. A range of services are available from Synthesia to Heygen. These avatars can be used as employees in management training, patients in healthcare training and customers in retail training.

SIGNIFICANT DIFFERENCES
Any form of human interaction can use this technique for training; in instructional videos, trigger videos, branched scenario videos and videos with additional AI generated learning experiences and assessment. In fact, the use of AI can lead to significant UPLIFTS in learning outcomes. In one trial with a client, before GenAI appeared, in 2020, AI enhanced learning resulted in a 61% increase in assessed learning.

INTERACTIVE CHARACTERS
We now have avatars that one can converse with using AI chatbot technology taking it to another level through scenarios and simulations, using real dialogue. We can expect tons of these to appear in computer games (OpenAI have dealings with GTA). But it is in training that they have huge potential. It has been impossible to create high fidelity simulations for soft skills in the past. I created a lot using fixed video clips in interviewing skills, conflict, language training and so on. They took a lot of time to design write and produce. These are about to get a lot quicker and cheaper.

CONCLUSION
The use of AI generated video is already here and will continue to evolve. We are not yet at the level of full drama but the direction of travel is clear.

GenAI synthesises and does not copy – huge case won by OpenAI

2024-11-08T13:53:00.006+00:00

This big news.

In what is seen as a critical test case, SDNY Judge Colleen McMahon has dismissed the idea that training a LLM is copying. The ruling, (without prejudice) did not provide judgements on what I'm about to say, merely stated the arguments and provides the explanatory detail, which I think is sound.

Generative AI ‘synthesises’, it does not copy. This is central. It’s a bit like our brains, we see, hear and read stuff but memory isn’t copying, it’s a process of synthesis and recall is reconstructive. If you believe in the computational theory of mind, as I do, this makes sense (many don't).

What is even more interesting is the conclusion that the datasets are so large than no one piece is likely to be plagiarised. That, I think is the correct conclusion. It would take 170,000 years for us to read the GPT4 dataset, reading 8 hours a day. Any one piece is quantifiably minuscule.

On the idea that regurgitated data has appeared. It would appear that this problem has been solved (almost), with provenance identified by some systems, such as GPT o1. In other words, don't worry, it was an early artefact of largely early systems.

I was always sure that these cases would result in this type of ruling, as the basic law of copyright depends on copying, and that is not what is happening here. All freshly minted content is based on past content to a degree and here it is not just a matter of degree (it’s minuscule) but also the methods used. Complex case but right rationale.

I think we're seeing many of the ethical objections to AI fade somewhat. There are still issues but we're moving past the rhetorical phase of straw men and outrage, into detailed analysis and examination. This is an important Rubicon to have crossed. Many so called 'ethical' issues are just issues that need to be worked through, rather than waved as flags of opposition. We are seeing the resolution of these issues. Time to move on.

AI school opens - learners are not good or bad but fast and slow

2024-11-05T13:10:00.006+00:00

What was surprising about this initiative was the strong reaction of outrage and dismissal. It is only 20 people at GCSE level, in a fascinating experiment but you’d think it was Armageddon. We have seen a rise in home schooling and school absences post-Covid. Not all are happy with the current schooling for their children, especially those with special educational needs. Why wouldn’t we want some experimentation in this area and AI is an obvious area to look.

Learners are not good or bad but fast and slow

The pedagogy is sound for some, perhaps not all. Rather than a one-size-fits-all direct instruction, each learner goes at their own pace. Sitting in rows in a classroom, rows in a lecture - that's the model this challenges. The myth is that the traditional model did what many claim it does.

Bloom researched this 50 years ago. Not his famous 2 Sigma paper, which over-egged the effect, but the idea of time to competence. He is best known for his ‘taxonomy’, but he never did draw a pyramid and his taxonomy was radically altered by subsequent researchers, as it was too primitive, rigid and far from representative of how people learn. His more important work in self-paced learning, led him to believe, in ‘Human Characteristics and School Learning’ that learners could master knowledge and skills given enough time. It is not that learners are good or bad but fast and slow. This recognises that the pace of learning varies among individuals rather than being a measure of inherent ability

The artificial constraint of time in timed periods of learning, timetables and fixed-point final exams, is a destructive filter on most. The solution was to loosen up on time to democratise learning to suit the many not the few. Learning is a process not a timed event. Learning, free from the tyranny of time. allows learners to proceed at their own pace.

Bloom proposed three things could make mastery learning fulfil its potential:

1. Entry diagnosis and adaption (50%) - diagnose, predict and recommend

2. Address maturation (25%) - personalise and adapt

3. Instructional methods (25%) - match to different types of learning experiences and time

That is what they are doing here. Lesson plans focus on learners rather than the traditional teacher-centric model. Assessing prior strengths and weaknesses, personalising to focus more on weaknesses and less on things known or mastered. It’s adaptive, personalised learning. The idea that everyone should learn at the exactly same pace, within the same timescale is slightly ridiculous, ruled by the need for timetabling a one to many, classroom model.

Learning coaches

There are three learning coaches, that’s one per 7 pupils, quite a good staff/pupil ratio compared to almost all schools. They are trained to oversee and encourage, rather than teach directly. That’s fine, as the direct instruction is done online.

By outsourcing subject matter expertise to the technology – AI has a degree in every subject, speaks many languages, can be adjusted to any level. It is this access to any subject that is so compelling. I have written about the realisation of a Universal teacher before. It is getting ever nearer.

It is also available 24/7, anyplace, the advantages over a strictly timetabled school are obvious. Holidays can also be taken at any time. These are simply practical advantages.

On top of this are the opportunities to make learning e accessible through adjusting the level of the language and opportunities for T2S and S2T, along with help on dyslexia and other disabilities, at a level way above normal school environments.

Criticisms

One criticism is that this will not developing emotional intelligence, as if single-age groups, sitting 30 or more in a small room encourages this more than smaller groups. They have learning coaches and are still speaking and interacting with each other. Do we say that working remotely from home has the same effect? Yet that has been normalised. At least these students are together in one place.

There is this idea that the only way to develop critical thinking is sitting in a row in a classroom or lecture theatre. Critical thinking is not some isolated skill taught on its own, it needs domain knowledge and this is what this approach encourages. AI can already critique a claim, debate with you and critique your own work. It will also unpack its own reasoning.

There is also plenty of opportunity for creating safe spaces for discussion and debate. Debate and discussion can be fostered formally and informally in this environment. There is even the possibility of debating online adversaries. The learning coaches deal with behaviour, public speaking and debate.

Costs

At an eye watering £27.000 a year, it’s a rich person’s game. With 20 start-up pupils, that’s over half a million revenue straight off the bat. But the cost to the state per pupil is £8500 in Scotland and £7200 elsewhere in the UK. One can see economies of scale emerge quickly if it works. But before spitting out the withering criticism, let’s see if it works.

Conclusion

For the first time in the history of our species we have technology that performs some of the tasks of teaching. We have reached a pivot point where this can be tried and tested. My feeling is that we’ll see a lot more of this, as parents and general teachers can delegate a lot of the exposition and teaching of the subject to the technology. We may just see a breakthrough that transforms education.

5 big surprises in Wharton report on GenAI

2024-11-04T00:31:00.012+00:00

Enjoyed this report as it was so goddamn honest, contradicting everything the 'consultants' and 'ethics' folk were recommending for the last year and more!

1. Gen AI Strategy is led internally—NOT by consultants

This bucks the trend. The strategy work is not led by the usual suspects or suspect advisors, many of whom have no real experience of building anything in AI. The bandwagon got bogged down by the sheer weight of hucksters. This technology gives so much agency to individuals within organisations, from to anyone producing text of any kind to coders, that the sisters and brothers are doing it for themselves. You wonder whether consultancy itself is under real threat from AI?

2. NO stringent policies on use in organisations
Interesting. Seems like a contradiction – massive rise in use but little sign of policies being used. I suspect that people have seen through the platitudes that are so often seen in these documents and statements of the blinding obvious, over-egging and exaggerating the ethical dangers.

3. Most employees do NOT face heavy restrictions in accessing Gen AI at their companies
The scepticism, regulatory effort, fear-mongering, even doomsters last year seem to have given way to a more level-headed recognition that this is a technology with real promise, so let's allow folk to use it, as we know they already do! My guess is that this AI on the SLY stuff happened so quickly that organisations just couldn't and didn't know how to respond. Like a lot of tech - it just happens.

4. Companies are adapting by expanding teams and adding Chief AI Officer (CAIO) roles
I wasn’t sure about this, as the first I heard about was in this report! I suspect this is a US thing or exaggerated in the sense of just having someone in the organisation who has emerged as the knowledgeable project manager. Can see it happening though.

5. LESS negativity and scepticism
More decision-makers feel ‘pleased’, ‘excited’, and ‘optimistic’, and less ‘amazed’, ‘curious’ and ‘sceptical’. Negative perceptions are softening, as decision-makers see more promise in Gen AI's ability to enhance jobs without replacing employees. This makes sense. A neat study showed that the scepticism tended to come from those who hadn’t used GenAI in anger. Now that adoption has surged, nearly doubling across functional areas in one year, the practical experimentation has shifted sentiment.

We seem to be going through the usual motions of a technological shift, where we get a period of fierce fear and resistance that gives way to the acceptance that it is all right really. The nay-sayers needed get it out of their systems, before use surges and realism prevails.

https://ai.wharton.upenn.edu/focus-areas/human-technology-interaction/2024-ai-adoption-report/

Should L&D be renamed the ‘PERFORMANCE Department’?

2024-10-30T16:38:00.004+00:00

4 hours agoL&D is a wagon without a horse, with no pulling power or sense of direction. Could a change of focus provide the direction, momentum and horsepower that has been lacking?

Could we rename Learning & Development, as just ‘Performance’, like Marketing, Legal and Finance Departments? Learning puts too much focus on courses. ‘Development’ is too vague. We must link what we do with measurable outcomes linked to proven performance, productivity and progress.

‘Performance’ puts the individual into the flow of the organisation. we need to foster a work environment where employees feel valued, motivated, and invested in – through their performance, not through a diet of ‘courses’.

A wider perspective seeing learning as performance would link what we do and what needs to be done. The organisation would see results and more sophisticated learning and work solutions related to strategy would emerge.

Linking learning with performance go hand in glove, allowing us to rise above just course delivery into more considered informal and incremental learning, also more performance support, linking inputs with outputs.

Focusing on performance can be challenging and needs a clear causal link between learning initiatives and productivity gains but it can be done.

Of course, we don’t just want ‘faster horses’ as Henry Ford said, we need new different forms of locomotion. The technology of the age is digital and AI. Employees can be more autonomous, self-driven through bottom-up agency.

AI has shown there is an immense thirst among employees for going ahead and doing things faster and better. We use this technology to both learn and get things done, seeing no difference between the two. Learning and doing things faster and better become fused.

In writing something with evidence and import, many use AI to do fast research, write concisely, even critique the proposition. They see AI as a means to an end, the end being solid output.

AI gives us agency to improve ourselves with guidance from the organisation. It fuses learning and productivity, easy to use, fast and motivating. It should be in our domain, our responsibility.

It would also force us to be more data savvy. Forget happy sheets and bums on seats (measures wrong end of learner), let’s get serious with actual evaluative data. AI gives us a data scientist in our pockets – let’s use it.

Of course, AI as just one catalyst for action, there are many others, such as performance support, the wider world of career development, developing strategic skills for future needs and challenges.

The name matters less than the direction of travel. Strategic alignment with organisational goals should be our new goal. It puts an end to those endless discussions about the future of L&D by asking ‘to what problem are we a solution’? We all know it needs to be more performance focused, more relevant, more integrated with the organisation.