Some months later, I gave the exact same presentation to a group of high school students. The data was identical. The visuals were identical. The narrative was identical. The reaction could not have been more different: instead of tears, there was laughter. Some students whispered jokes. Others seemed bored.

That experience stayed with me. It was a vivid reminder that a data story is never universal. The same message that resonates deeply with one audience can completely miss the mark with another. The difference wasn’t the story; it was the audience.

That experience made me wonder: Is it possible to use generative AI to foresee how various audiences might react to a data story, even before sharing it with them?

The Invisible Variable

When we build a data story, we often focus on getting the numbers right, cleaning the data, choosing the best chart, and writing a clear narrative.¹ But every story also carries an invisible variable: the audience’s mindset.³ Are they skeptical or trusting? Emotionally engaged or detached? Do they seek scientific rigor, or are they motivated by moral urgency?

These differences shape how people interpret data and how they feel about it. And in communication design, feelings often determine whether comprehension turns into action. Traditional user testing can reveal these dynamics, but it usually happens after the story is produced. By then, we’ve already invested hours (or weeks) of work. What we need is a way to test audience reactions early, during design, not as a validation step, but as a creative one.

That’s where generative AI enters the picture. We often think of AI as a content creator: writing text, generating visuals, composing music.² But what if we flipped the role? Instead of asking AI to tell the story, we could ask it to listen to one.

In other words, we can use AI as a simulated audience: a stand-in that reacts to a data story as if it were a specific type of reader or viewer. This isn’t about replacing real humans, it’s about gaining early insight and seeing how different kinds of audiences might respond before we test with real people.⁴

A Theoretical Workflow

Think of this as a four-stage loop (Figure 1):

1. Design the story. Start with the basics: your insight, your visuals, your message. But also define who your audience is. Are they experts, skeptics, or newcomers? What’s their level of trust in data?
2. Structure the evaluation. Identify which dimensions you want to test, such as comprehension, emotional engagement, perceived credibility, and behavioral intent. Build a few short survey-style questions around them.
3. Test with a small real audience (once). This gives you an anchor in reality. Collect qualitative and quantitative feedback from a few people representing your intended audience. Use their responses to outline behavioral profiles.
4. Simulate with generative AI. Now translate those profiles into personas, detailed, natural-language descriptions that capture each audience’s mindset. Feed these personas to a model like ChatGPT, along with your data story and evaluation questions. The model will respond as if it were that type of person.

Over time, you can refine your prompts, compare responses across personas, and even test multiple versions of the same story to see how tone, framing, or design choices affect simulated reactions. This process doesn’t give you the truth. Instead, it gives you signals, which are often enough to guide early design decisions.

Deeper Than Demographics

The real power of this approach lies in how you design your personas. Too often, personas in communication design are shallow: “female, 35, educated, urban.” These categories tell us little about how people think.

A simulation persona needs to go deeper. It should reflect cognitive and emotional traits that influence interpretation. For example:

• A Skeptical but Attentive persona might be someone who prefers calm, fact-based messaging and resists alarmist tones.
• An Engaged Believer might already trust the data, feel emotionally involved, and look for actionable next steps.

When we ask the AI to adopt one of these perspectives, it doesn’t magically “become” that person, but it generates responses that align with the language, tone, and priorities of that mindset. This gives us a way to see our story through multiple cognitive lenses.

Generative AI performs remarkably well in cognitive domains⁴: it can detect ambiguity, assess clarity, and point out logical inconsistencies. It also mirrors differences in belief or skepticism quite effectively when guided by well-designed personas. Where it struggles is in affective realism. AI can describe emotion, but it does not feel it. It might “say” that a story is moving, but there is no inner resonance behind that statement. 

A Mirror, Not a Replacement

There are also important ethical questions. When we use AI to simulate human behavior, we risk introducing bias from its training data, oversimplifying real diversity, or treating machine-generated feedback as objective truth.

To avoid this, we must:

• Be transparent about how personas are created and used.
• Document every prompt and assumption.
• Always include human validation before making communication decisions.

Generative AI can expand our empathy, but only if we remember that it is synthetic empathy. It’s a mirror, not a replacement.

Hybrid Workflow

The same story that once made one audience cry and another laugh taught me that data stories live in the space between message and perception. If AI can help us explore that space, then it may become less a machine and more a mirror, showing us how differently truth can sound depending on who listens.

The next step is to integrate AI simulation and human evaluation into a hybrid workflow:

• AI provides the breadth: fast, low-cost iterations across many personas.
• Humans provide the depth: emotional nuance, authenticity, and lived experience.

Together, they can make data storytelling more intentional, ethical, and empathetic. For me, the goal isn’t to automate understanding; it’s to make understanding itself a design material, something we can prototype, test, and refine just like visuals or text.

Disclosure: I used AI in creating this post: Grammarly to correct grammar and sentence structures; ChatGPT to write the basic skeleton of the post.

References

1. Dykes, B. (2019). Effective Data Storytelling. John Wiley and Sons.

2. Li, H., Wang, Y., Liao, Q.V., and Qu, H. (2025). Why is AI not a panacea for data workers? An interview study on human-ai collaboration in data storytelling. IEEE Transactions on Visualization and Computer Graphics.

3. Lo Duca, A. (2025). Become a Great Data Storyteller. John Wiley and Sons.

4. Lo Duca, A., and Yocco, V. (forthcoming). Exploring the Use of Generative AI for Assessing Data-Driven Stories. In Computer-Human Interaction Research and Applications. CHIRA 2025. Communications in Computer and Information Science.

Angelica Lo Duca is a researcher at the Institute of Informatics 
and Telematics of the National Research Council, Italy. Her research  interests include data storytelling and the application of AI to different domains, including cultural heritage, tourism, education, and more. She is the author of Data Storytelling with Altair and AI (Manning Publications, 2024) and Become a Great Data Storyteller (Wiley, 2025).

This post digs into the tools that the interfaces can offer users to help them clarify and communicate context. Most users’ goals will resemble those that others have when coming to GenAI, but sometimes users are playing with ideas and aiming for an output that is super-imaginative and demonstrates a surprising way to use GenAI. We want to offer users a framework that encourages curiosity, exploration, and experimentation in the interface.

Interfaces in the Physical World

Before GenAI, people were seeking knowledge and information in libraries and the physical, embodied world. The point we want to highlight about that experience is that the physical world could not bend around the user’s needs; users were required to enter and engage with the layout of a library as they found it.

So now we ask: What do users lose, if anything, when those constraints are removed? Instead of supposing that the ‘best’ GenAI tool is one that enables users to fast-track to an answer, we ask: what is the impact of raising the user’s attention—what are our rules of engagement?

Part One: Navigating Physical Interfaces

In many areas of life, people have traditionally had carefully designed static physical libraries, archives, or even domain-based study rooms that support memory and learning, and act as interfaces to how we find answers to queries. These systems don’t just give answers; they shape how users approach a question. These can become so static and impersonal that college librarians frequently add display tables celebrating a current topic, new acquisitions, or new periodicals, and populate the library with posters, information boards, and seminar rooms that link with life on campus.

If all goes well, a user leaves with new ideas and explanations and some underpinning architecture (we call it epistemic insight) that they can use in other libraries to find their way around.

Back to our question: what do we lose when we move from a library to a GenAI that can shape the presentation of information around the user’s request? And, to get to the heart of the matter, can we have the best of both worlds?

Consider an interface that assumes ‘disciplines’ are fundamental to how a lot of valuable knowledge is made; we can then have a GenAI tool for use on an academic campus that interacts with users by saying things like “Would you like to explore this topic through a scientific or historical lens?” or “This idea appears in both philosophy and sociology; do you want to compare how each discipline approaches it?”

Part Two: Mixing Genres, Making Meaning

Extending the idea of mixing disciplines, we can also use GenAI to bridge other categories of content that don’t usually meet—and produce a one-off response that combines them. Take this AI prompt: “Write a weather forecast for Swansea in the style of Dylan Thomas.”

Our experiment with Microsoft’s Copilot gave us:

“And the rain, oh the rain, it will come tumbling like a choir of grey ghosts over the copper roofs of Uplands, whispering secrets to the sea. The wind will wander drunk through the alleyways, rattling the bones of old bicycles and sighing through the chapel doors. By nightfall, the clouds will gather like gossiping aunties above the bay, and the moon, if she dares, will peek through the mist with a wink and a warning.” (Microsoft, 2025)

This isn’t a particularly good format for a forecast, and it’s also not an authentic rendition of what Thomas wrote. But it almost effortlessly lets us explore evocative and provocative ways of talking about the weather that warrants some further exploration,

Part Three: Inventing New Interfaces

GenAI can do more than respond or remix; it can inspire educators and others to invent entirely new interfaces that let users explore structure, style, and substance in tandem, to learn as we play.

Here’s a case study where we ask GenAI to write in the style of James Joyce about living in a smart city. Instead of leaving it to GenAI to improvise, the user is given control: they can choose which aspects of Joyce’s style and which features of smart cities they want to include. Compiled with some input from GenAI, here’s one suggestion for controls that a student might select.

Joyce style controls:

1. Stream of consciousness (Joyce)
2. Experimental syntax
3. Multilayered wordplay
4. Sensory imagery (Joyce)
5. Thematic depth

Smart city content controls:

1. Sensor-driven infrastructure
2. Participatory governance
3. Data-driven decision making
4. Sustainable design
5. Interconnected services

What happens if the user selects: stream of consciousness, sensory imagery; and sensor networks, sustainability? In our experiment, Microsoft Copilot generated this:

The pavement hummed beneath her soles, a low electric murmur, like the breath of the city dreaming. She walked—no, drifted—past the vertical gardens, their leaves whispering carbon secrets to the air. The bins blinked green, full but not overflowing, the sensors had spoken. And she thought, not in words but in pulses: is this what it means to be cared for by machines? (Microsoft, 2025).

By experimenting with different aspects of Joyce’s style and what a smart city is like, users can become ‘agentic learners,’ experimenting with the output while exploring and challenging their understanding of how the output is produced.

Finally, what could an interface look like that invites users to mix genres, aspect by aspect? There are many ways to approach exploring genres. Let’s say the user presents a request for a text in a prompt. On the screen, they could have a button bar for 10 aspects of Joyce’s writing, and of a smart city. Selecting different aspects changes the story. The user could use this interactive interface to construct and send a prompt to GenAI that stipulates aspects of Joyce and ‘smart city’. Indicators based on preliminary use with university students are that it is an entertaining way to fabricate quirky outputs, and useful for raising questions to discuss more widely about authenticity, transparency, integrity, usefulness, permission, and creative control.

We conclude interfaces matter, and hope giving selectable controls can help shape interesting results, but also encourage students to explore and reflect on alternatives. We can also wonder what is lost by giving users a single entry point and invitation to ‘ask me anything’. Our case is that users can benefit from ‘learning on the job’ about content categories such as genres, topics, and disciplines, to inspire them to investigate ways to be creative using GenAI that weren’t possible before.

The message for users and interface designers is that GenAI can do more for us than save time and fast-track users to finished, polished outputs.

Berry Billingsley is an educator interested in philosophical ‘big questions.‘

Ted Selker is a computer scientist and student of interfaces.

Today, AI is on that path. What seemed impossible at first is now continuously developing. Will AI reshape our world like the Internet did? Or will it face challenges? Can AI expand the human mind, helping us think, create, and solve problems in ways we never imagined? Or will it be a big failure?

The answers to these questions can be found by analyzing past patterns. In this post, we will discuss why the past holds the blueprint for AI’s future, what history warns about AI, and how yesterday’s patterns shape its future. 

Look Back to See AI’s Future

Suppose you time-travel to 1975. You are holding the first Altair microcomputer. You can feel that something big is coming, but you can’t predict exactly what’s going to happen.

Now fast-forward to the present. You can see the transformation from room-sized calculators to smartphones in every pocket. To predict what changes AI will hold for the future, we must look back. The past is a blueprint.

Why study earlier shifts? Because they teach us:

• How people adapted: When was adoption rapid, when slow?
• What risks emerged: Past problems are often repeated in new forms.
• Which choices mattered: Design, openness, regulation, usability.

As AI is integrated into our day-to-day lives, it’s important to ask: How did past computing technologies get accepted? How did they break, get fixed, or cause disruption? These lessons guide how we build AI systems and how we govern them.

Early Breakthroughs and Their Lessons for AI

Following are four breakthroughs in computing and what they taught us about AI:

1. Programmable Computers

In the 1940s and 50s, early machines like ENIAC and UNIVAC were built for purposes like ballistic computing, weather prediction, and code breaking. Then came stored-program computers, machines you could load with different programs and repurpose hardware. 

Lesson: Flexibility and design choices shape how broadly a technology is accepted by users. AI is adopted at a larger scale because of the architecture of models and APIs. Also, its adoption in the future will be based on how easily developers can adapt systems. 

Consider an AI certificate generator tool for online courses. If the underlying system is rigid and only designed for one format, it serves a narrow user base, but if it’s flexible, then it has a better chance of scaling rapidly. That adaptability is the main reason behind the shift from single-purpose machines to programmable computers.

2. The Internet

The dot-com era showed us a world that could stay interconnected easily. People who had never met could send messages, goods, and ideas. Data started growing as people started surfing online.

Protocols like TCP/IP and HTTP were standard, open, and extensible. Because of that, the use of the Internet grew through social media, e-commerce, streaming, and remote work.

Lesson: Once systems connect, growth can be massive. For AI, interconnectivity, such as sharing data, models, and platforms, can confer power, but also poses risks of misuse or cascading failures.

Take the example of AI traffic analytics tools. When cities deploy sensors and AI-powered cameras, they begin with counting vehicles. Soon, the same infrastructure will be used to monitor parking, pedestrian flow, emergency response, and commercial advertisement placements. The scope expands fast.

That’s powerful, but regulatory, privacy, and bias issues may also occur. These weren’t central at day one.

3. Personal Computers

Until the late 1970s and early 1980s, computers were found only in labs or big companies. Then came the revolutionary Apple II, the IBM PC, and Commodore machines. These were installed in homes, schools, and small businesses.

Suddenly, computing power was accessible not just to engineers, but to creative writers, accountants, and kids who began learning to code.

Lesson: Accessibility drives adoption faster than raw technical power. Users need a machine that works, is affordable, fits their needs, and is easy to use.

Take the example of a startup that builds a user-friendly AI model for small business accounting. The accuracy of the model is not the best. However, the tool is cost-effective and the interface is simple. Integrating the tool with existing systems is seamless. These reasons will lead people to adopt the tool. 

4. Open Source Movement

Open source software was a huge innovation. With open source, people began collaborating on projects, sharing code, identifying bugs, and refining each other’s work through the platforms. 

Lesson: Shared innovation moves faster than in-house systems. Open source helps identify risks, bugs, security loopholes, and unintended consequences that an in-house team may miss. Open-source allows all like-minded individuals to connect and contribute. 

In 2024, developers worldwide made over 5.2 billion contributions to more than 518 million open-source, private, and public projects on GitHub. This highlights the potential growth of collaborative development in AI.

When a researcher releases an open dataset and open model code, thousands of people can suggest various things. They may find adversarial vulnerabilities, bias, and data poisoning before deployment. These observations may be missed with closed systems. Open source allows people to share their ideas and grow together.

Patterns of Risk: What History Warns About AI

The past is interesting, but it presents us with key challenges we must solve. Let us look at a few of them.

Scaling Up: From Mainframes to the Cloud, to AI models

The world of computing started with mainframe computers; they were bulky and expensive. Then the client-server architecture, personal computers, and cloud computing were developed. Scaling the innovation to the next level was always costly because it included the costs of power, cooling, or maintenance. 

AI is scaling at a huge level. Models are trained and deployed across multiple devices.

Traditionally, computing systems required stability to ensure they were growing. But these days, AI infrastructure depends on uptime monitoring to ensure uninterrupted performance. Continuous uptime monitoring tracks the server performance, API latency, and response accuracy in real time. This prevents downtime and data loss. 

A combination of uptime monitoring and predictive analytics ensures that organizations can maintain high availability, trust, and quality of service.

Unexpected Risks

Positive developments may come with something negative. For example, the growth of email opened the door to phishing spam. The growing use of Web technology gave rise to malware, phishing, and fake news. With the spread of smartphones, privacy and surveillance concerns began to grow.

AI comes with risks like deepfakes, biased decision-making, and data misuse. 

Users can take control of their digital privacy using tools like Incogni. They can ask data brokers to remove personal data from their databases. It is an important step in an AI-powered world where personal information can be scraped, sold, or misused.

Crossing Disciplines

Computing breakthroughs are not magic. Developments need time, effort, and a proper mindset. Lawyers, thinkers, and sociologists ask: How does this affect privacy? What is ownership of data? Are their psychological impacts of constant connectivity? Asking these questions helps ensure that problems are addressed before they occur.

Therefore, when we build systems, we need ethicists, social scientists, and policy experts. Having the answers to questions like “How will people trust this?” or “How will this affect jobs?” is important.

How Yesterday’s Patterns Shape Today’s AI Evolution

Let us discuss what actions we can take to avoid repeating old mistakes. 

Usability or Raw Power

Why did personal computers became so popular? It wasn’t because they had the biggest silicon power; it’s because they were accessible to everyone. The AI tools of today and tomorrow will succeed when they are easily accessible and usable.

For example, by incorporating tools like an AI humanizer in writing tools. Integrating a feature that humanizes a tool’s AI-generated content makes the tool accessible to developers, educators, or small business owners. It ensures that a tool is useful and accessible to everyone. 

Here’s a tip: Build any model with the use case in mind. Ask if the model is useful and accessible to everyone? If yes, then go for it. 

Collaboration, with Guardrails

Open source and collaboration increase innovation. Shared datasets, model benchmarks, and community audits reveal failures early. But openness also needs governance. Clearly mentioning things like who owns the data, what the privacy guarantees are, and what the licensing terms are ensures that your audience knows what is happening with their data. 

Preparing for Side-Effects

All technological breakthroughs come with some side effects. Electric power generation brings pollution, vehicles cause accidents, and the Internet brings cybercrimes.

The side effects of AI are bias, job displacement, misuse, and misinformation. Organizations must plan for these.

Before deployment, run impact assessments. Think about what could go wrong, and for whom. Include marginalized groups. Also, prepare for situations like fallback modes, the ability to shut off or correct errors, and ongoing monitoring.

Adoption Curves Ahead

Technologies follow S-curves. The initial stages will see slow adoption, followed by rapid growth, and then a plateau. Stakeholders may misinterpret slow early growth as failure, or underestimate rapid later change.

Those working in AI product development or policy should find out possible adoption paths. Consider using early pilot programs and large-scale rollouts to find out the problems first.

Takeaways for Researchers and Practitioners

Researchers, engineers, startup founders, and policy makers can work on the following areas:

• Design for understandability
  Make models explainable. Use clear documentation. Provide user education. If people can’t understand how an AI tool works, they may not use it at all or misuse it.
• Share knowledge openly when possible
  Publish benchmarks, data leaks, and negative results to showcase that you are transparent about your work. Openness speeds up progress and helps avoid repetitive mistakes.
• Think about long-term impacts, not just immediate performance
  Performance metrics like speed and accuracy are important, but consider other factors like fairness, sustainability, environmental cost, and social effects. This ensures that the development of a particular technology is safe.
• Build ethical and regulatory awareness into the workflow
  Include ethicists, legal experts, and privacy officers from the early stages so that they can keep an eye on legal issues. Don’t treat regulation as an afterthought.
• Focus on usability and accessibility
  Inclusive design matters: for different languages, Internet speeds, devices, and literacy levels. AI tools shouldn’t only serve those with high-end devices or technical backgrounds.
  

Lessons From the Past, Guidance for the Future

AI is the next big wave in computing, but history shows us that breakthroughs always come with opportunities and challenges. We can make smarter choices in the future if we know the lessons of past breakthroughs as well. These include: 

• Programmable computers taught us that flexibility helps in scaling a particular technology into something big.
• The Internet showed how connection helped in unpredictable growth.
• Personal computers proved that systems that are easily accessible will expand more easily.
• Open source highlighted the speed of shared innovation.

Each shift came with challenges like spam, malware, and privacy issues that must be watched for AI as well. The past has the power to tell the future. Therefore, analyzing past breakthroughs helps in predicting the future. 

Richa Gupta is a Content Marketing Specialist with over seven years of experience. She has worked with various SaaS brands to create content strategies that boost organic traffic and generate qualified leads.

The streaming industry is shifting from content competition to an analytics-driven model. Success now depends on how platforms understand and act on audience behavior. Advanced analytics and predictive insights are crucial for creating tailored experiences that boost engagement and reduce churn. Platforms must evolve to treat analytics as a core business driver, leveraging AI-powered solutions to understand user needs and anticipate them, driving sustained growth.

Why Traditional KPIs Fall Short

For too long, streaming platforms have relied on traditional key performance indicators (KPIs) like total watch time and subscriber growth. These metrics offer a snapshot, but fail to capture why viewers disengage or measure what drives loyalty. They miss critical nuances such as how preferences evolve, how content is discovered, and how sentiment influences engagement.

To gain deeper insights, platforms are integrating behavioral signals, sentiment analysis, and discovery patterns across content delivery and engagement workflows. This richer data helps answer essential questions: What content creates emotional connections? What signals predict churn? How can platforms intervene before viewers leave?

With AI-led solutions and predictive analytics, platforms can move beyond reactive measures to proactively optimize engagement, content strategies, and retention. This shift enables streaming services to not only understand their users, but also to anticipate their needs, setting them up for sustained success.

Forecasting Disengagement Before It Happens

By analyzing patterns in viewing frequency, content interaction, and sentiment, platforms can identify users at risk of disengagement. This enables proactive retention strategies, customized recommendations, targeted promotions, or exclusive access before the viewer hits “unsubscribe.”

Frameworks like PREDICT (Pattern recognition, Risk scoring, Early intervention, Dedicated support, Individualized offers, Continuous optimization, Tracking ROI) are becoming standard in analytics-driven planning. By adopting such frameworks, organizations can respond to churn and actively shape customer loyalty through data-driven insights.

The Role of AI in Viewer Engagement

As hybrid and ad-supported models redefine the streaming landscape, platforms are increasingly using AI-powered analytics to improve audience engagement. With machine learning and predictive models, streaming services can dynamically adjust content recommendations, ad placements, and even AI-generated trailers, ensuring that every interaction aligns with viewer preferences and maximizes engagement.

Data-driven solutions allow platforms to make prompt adjustments that enhance engagement and monetization. By using predictive intelligence, platforms serve not just relevant content, but also ads that elevate both user experience and advertising revenue.

The integration of audience data platforms enables the unification of viewer profiles across multiple touchpoints and devices, offering a 360-degree view of the user journey. This empowers platforms to optimize content delivery in real-time, making adjustments based on factors like interaction history, viewing time, and sentiment.

Analytics as a Valuation Driver

Analytics does more than enhance user experience; it directly affects the financial health of streaming platforms. Platforms that harness predictive insights and efficient monetization strategies are better positioned for sustained growth, funding, and strategic partnerships.

Moreover, analytics-driven strategies directly contribute to higher platform valuations by showcasing resilience, adaptability, and a relentless focus on audience-centric innovation. In today’s fast-paced, insight-driven streaming landscape, these qualities are not just advantageous; they are vital for growth. Platforms that leverage predictive intelligence and data insights are prepared to react to market shifts, drive continuous innovation, and personalized engagement, ensuring their position in a crowded marketplace.

The New Competitive Edge

The future of streaming will be defined by platforms’ ability to harness data, making every interaction meaningful rather than just focusing on the content they produce. Platforms that translate analytics into actionable strategies, creating meaningful engagement that resonate and foster loyalty, will set the standard for long-term growth. Analytics is no longer a back-office function or a measurement tool; it powers innovation at every stage, shaping content, marketing, and monetization in real time. In this rapidly evolving landscape, the winners will be those who turn data into foresight, insight into action, and engagement into lasting audience devotion.

Kuljesh Puri is Executive Vice President and General Manager—Communications, Media & Technology at Persistent Systems, with over 26 years of leadership experience across the software, telecom, and semiconductor industries.

Pawan Anand is Associate Vice President at Persistent Systems, leading AI-driven transformation programs across the Communications, Media, and Technology sectors. He holds an Executive Doctorate in Business Administration from Temple University.

Each advance delivered progress for society—but at the cost of job loss and deskilling.

Artificial intelligence (AI) is triggering disruption. It alters the way people think—and how they work. AI can draft contracts, diagnose medical conditions, spot tumors, compose art and music, and write computer code without a human in the loop.

While AI augments and improves many processes, the deskilling that occurs as a result of AI can contribute to a loss of basic knowledge, a deterioration in social interaction skills, and a diminished ability to analyze, understand, and diagnose critical problems.

“The things that have made us successful as a species include the ability to be creative, adapt intelligently, and solve problems in an innovative way. We have never before had computational tools that approximate the high level thinking of humans,” said Aniket Kittur, a professor in the Human-Computer Interaction Institute at Carnegie Mellon University.

As a result, researchers are exploring ways to reskill, upskill, and develop new and better critical thinking frameworks tailored to AI. “If left unchecked, deskilling can erode the expertise of individuals and the capacity of organizations,” said Janet Frances Rafner, a researcher in the Department of Management at Aarhus University, Denmark.

Skills Lost and Found

Concerns about deskilling appeared in the late 19^th century, when machinery and factory automation began to replace human labor on a grand scale. “Specific knowledge and roles disappeared. There were fewer job opportunities—and usually lower pay—for people who had jobs centered on these tasks,” said Matt Beane, an associate professor at the University of California at Santa Barbara.

Today, AI saves time and sometimes outperforms humans at tasks—reading X-rays and summarizing complex documents, for example. As a result, humans increasingly hand over this work to AI. The problem? People who no longer maintain these skills may see their abilities erode, particularly when it matters most. Worse, people with little knowledge or training can perform the same work with AI, thus driving down wages.

Without appropriate oversight and controls, both groups may perform worse. Consider: a study published in 2025 in The Lancet of Gastroenterology & Hepatology found that endoscopists who routinely used AI for assistance in colonoscopies performed worse if access to the technology suddenly disappeared. The detection rate for precancerous lesions dropped from 28.4% to 22.4% without AI in the picture.

Similar issues pop up in law, education, journalism, software development, and other fields. Law professors at Illinois Law School found that students who used chatbots and other forms of GenAI were more prone to critical errors. They concluded that without the right checks and balances, the technology could lead to widespread deskilling, particularly among younger and less-experienced attorneys that might depend on AI as their primary source of information.

Learning Curves

There’s a reason AI causes deskilling. In a 2025 survey conducted by Microsoft Research and Hank Lee, a Carnegie Mellon Ph.D. student, knowledge workers reported that generative AI made tasks seem cognitively easier. But there was a catch: researchers found they were ceding problem-solving expertise to the system and, instead, focusing on functional tasks like gathering and integrating responses. At the same time, they became more confident about using AI. “It is plausible that high confidence in AI could lead to lower perceived effort,” Lee said.

To be sure, AI can serve as both a blessing and a curse. For example, “Senior engineers and coders can often accomplish work faster and better using AI because it accelerates their productivity,” Beane said. Yet the same systems can sabotage younger workers who benefit by collaborating with experts. “Over time, we risk losing future knowledge and expertise,” he noted.

Waves of deskilling are now crashing into almost every line of work. AI gains that seem beneficial over the short term—particularly the ability to work faster—may introduce longer-term and more profound problems, including a hollowing-out of core expertise in many fields, experts say.

“Some people claim that we are reaching a point where AI is on the verge of making humans less efficient and valuable,” said Jacob F. Sherson, director of the Center for Hybrid Intelligence at Aarhus University. While he doesn’t subscribe to that theory, Sherson is concerned. “Deskilling—and any fallout it creates—will be visible only in hindsight,” he observed.

Already, one group of data scientists has warned that by 2027, advances in AI could cross a critical threshold, producing systems that push human labor toward obsolescence. Sherson believes such dire warnings focus on the wrong thing. It’s possible, he said, to actively design a better future through a form of hybrid intelligence built around reskilling and upskilling.

Work Over

Amid all the disruption, a basic question emerges: “Are we ‘raising the floor’ or ‘raising the ceiling’ when we talk about the intersection of human and machine skills?” Kittur asked. AI should expand and amplify human capabilities—things like intuition, creativity, and reasoning—without undercutting core skills, he explained. “The human spark sets the direction,” he said.

Sherson and Rafner said organizations must establish metrics to track both technical and human capacity building. Their hybrid intelligence framework resembles systems that track carbon emissions and sustainability within the context of productivity and profits. In the AI space, this includes factors like employee AI self-efficacy and psychological safety. “The goal is for humans to move up the value chain,” Rafner said.

Organizations must start by designing AI technology around skill development rather than simply reacting to it, Beane said. The CMU and Microsoft group believes organizations must help workers adapt. “Training programs should shift the focus toward developing new critical thinking skills specific to AI use,” said Advait Sarkar, senior researcher at Microsoft Research.

Ultimately, some decisions—particularly those affecting other people—must remain with humans, stressed Kevin Crowston, Distinguished Professor of Information Science at Syracuse University. In addition, society would benefit by establishing certain “non-negotiable” skills: the ability to verify a calculation, write clearly, and analyze information, for example. “People must retain some level of core literacy in areas that are important. They must be able to account for their actions,” he said.

Amid a rapidly changing world, having an objective remaions important. Deskilling can lead to a variety of positive and negative outcomes, Kittur said. How society chooses to design, deploy, and govern AI will determine whether it fuels progress or unleashes misery on the masses. “We have to focus on finding ways to weave AI into work and life in a way that enhances humans,” he said.

Samuel Greengard is an author and journalist based in West Linn, OR, USA.

Imagine the set of all possible math theorems; only a subset has been proven by human mathematicians. Said Arora, “A superhuman AI mathematician is one that can prove more theorems than humans have.”

Arora, who was awarded the 2011 ACM Prize in Computing for his contributions to computational complexity, algorithms, and optimization, sketched a possible path to a superhuman AI mathematician. He explained that the idea traces back to David Hilbert’s early 20th-century dream of automating mathematics. That dream was crushed by the work of Gödel, Turing, and Church, yet it left behind something lasting: the concept of formal proof verification—the notion that mathematical proofs can be written in a precise language and then rigorously checked by a computer.

The modern open-source programming language and proof assistant Lean is ideally suited for precisely this purpose, Arora said. A proof written in English can be translated into Lean, after which the Lean checker verifies whether or not the proof is correct. “Rewriting the proof in Lean is presently done by humans, but very soon this will be done by AI,” Arora said.

Superhuman performance will probably arise from AI self-improvement through reinforcement learning, Arora said. AI has already been trained on math books and math papers, so it is aware of the definitions of mathematical concepts. “The idea of self-improvement is that you give the AI a large question bank created by humans; it follows many attempts to answer these questions, and the correct answers are used for further training.

“How does it get the correct ones? That’s the human feedback,” Arora said. “Some humans have labeled them as correct answers. This is the present pipeline. But in math, you can verify the answer with Lean. So, if you just ask AI to produce its proofs in Lean, labeling the correct answers can be done automatically. Even if it is a long proof, we humans can trust that it is correct.”

Automating proof-checking is one thing, but what about the mathematical questions that are worth being explored? Do they still need a human mathematician? Said Arora, “There is increasing evidence that AI itself can generate very good questions, so therefore you won’t need the large question bank from humans. Maybe the AI will start with some human data, but after that, it takes off and produces new questions. Why can this work? Well, the strongest suit of AI is its creativity. It is trained on this massive amount of data and then can find things across the data and combine them in interesting ways. It is creative, okay, but we know that it can also hallucinate, right? It doesn’t have a very good idea of truth. But it has Lean to verify answers and weed out the wrong questions and answers.”

This is no longer just theory. Experiments have shown the method explained by Arora can work in practice, albeit at less-than-superhuman level. Arora mentioned the examples of DeepMind’s AlphaGeometry and AlphaProof, and the Goedel-Prover created in Arora’s own Princeton Language and Intelligence lab. In 2023, AlphaGeometry solved International Mathematical Olympiad (IMO)-level geometry problems without needing human demonstrations. AlphaProof focused on more general formal math reasoning, not just geometry, and achieved in 2024 the silver-medal standard on IMO problems. This year, models from OpenAI and Google reached the IMO gold-medal standard.

Whereas the models of OpenAI and Google are secret, Arora’s Princeton lab has released an open-source model, called Goedel-Prover-V2. “We verified that you can use current open-source AI models and iterate through 10-to-20 rounds of self-correction to solve five out of six IMO questions, which is at the gold-medal level,” he said.

Another step paving the way for a superhuman AI mathematician came earlier this year from AI startup Morph Labs, which announced a major improvement in AI translating English proofs into Lean. Mathematicians Terence Tao and Alex Kontorovich have been trying to convert the Strong Prime Number Theorem in Lean, which they thought would take a few years of human effort, but with an AI tool called Gauss, it took Morph Labs only three weeks.

“So, this whole pipeline of AI verifying answers and producing its own questions really is taking off,” Arora said. “If AI superintelligence happens, mathematics is the likely first domain because of the verified answers. The reason it can get better is because it is tireless and it can just keep repeating, do this self-improvement loop and get better and better. Just remember that this math superintelligence does not mean it’s perfect, only that it is better than humans.”

Bennie Mols is a science and technology writer based in Amsterdam, the Netherlands.

This second meeting of AI Safety Connect (AISC) addressed the growing need for international and cross-sector collaboration to mitigate the risks associated with advanced AI systems. Keynote speeches from ACM Turing laureate Yoshua Bengio and University of California, Santa Cruz Professor of Physics Anthony Aguirre outlined the hazards of agentic AI, while panels on “red lines,” private sector approaches to risk, and the rise of nonprofit AI research institutes shone a light on the challenges of AI governance in an age of rapid innovation.

AISC was started by Nicolas Miailhe and Cyrus Hodes, co-founders of The Future Society. Miailhe and Hodes envision AISC as an event series to facilitate dialogue between nonprofit organizations, academic and industry laboratories, governments, and investors. AISC’s first meeting took place as the Artificial Intelligence Action Summit in Paris in early February 2025 and included speakers from a global array of universities, nonprofits, and tech companies.

The second meeting welcomed NGOs and governments in addition to academia, nonprofits, and the private sector, with panels featuring representatives from the United Nations Development Programme (UNDP) and the governments of Singapore, Canada, and Brazil. As Hodes told participants, AISC aims to foster “shared understanding” and establish “responsibility mechanisms” across a broad range of sectors.

Earning Trust Across the Globe

The event opened with remarks from political leaders and representatives from the UNDP and International Telecommunication Union. Speakers agreed that AI innovation need not come at the cost of critical oversight measures.

Elina Valtonen, Finland’s Minister of Foreign Affairs, said AI “must always be deployed in accordance with international human rights and humanitarian law.”

Minyoung Han, Korea’s Director General for Climate Change, Energy, Environment and Scientific Affairs, described Korea’s efforts to “ensure that AI drives prosperity for all,” primarily by emphasizing safety as a cornerstone of research and development.

Building on the theme of civic innovation, Josephine Teo, Singapore’s Minister for Digital Development and Information, declared that public trust “has to be earned” rather than assumed. “People need to be assured that the AI we’re letting them use is developed and deployed in an ethical manner,” Teo said.

S. Krishnan, India’s Secretary for the Ministry of Electronics and Information Technology, echoed this sentiment, asserting that “AI should serve as a force for democratization” within and across national borders. Krishnan’s vision for democratic AI emphasizes transparent and widely-accessible resources, including databases, algorithms, and public education.

The Stakes of Unchecked Development

Following opening remarks, Bengio and Aguirre took to the stage to share their perspectives on the stakes of an AI-driven future.

Bengio outlined three major targets for safety research: market and power concentration; abuse by bad actors; and the potential for AI to advance beyond the threshold of human control. Each of these areas is associated with potentially catastrophic outcomes, he said. The first seeks to prevent “radically disproportionate” power discrepancies that could ensue from unregulated AI. The second addresses AI’s potential to be used in cybersecurity attacks, disinformation campaigns, and the construction of chemical and biological weapons. The third considers the current limitations of scientific research; as Bengio pointed out, today’s researchers “don’t know how to design AI that won’t harm people.”

Aguirre’s keynote took a similar warning tone. “Certain systems should not be built or deployed,” he said, voicing his support for the implementation of “red lines,” or hard constraints on AI agent capabilities, which was the subject of Nobel Peace Prize Laureate Maria Ressa’s opening address to the General Assembly. According to Aguirre, scientists and politicians should see themselves as equal participants in the red lines agenda. While scientists must commit to sharing their knowledge with politicians, politicians must work towards agreements on acceptable boundaries. This is not a minority position because the global call for red lines represents “a very strong consensus,” he said.

While the prospect of out-of-control AI may seem frightening, Aguirre emphasized that open dialogue can lead to a future where AI’s benefits don’t come with severe costs. “I think there’s a tendency to consider the development of AI like a train on a single track, and our only option is merely to decide how fast or slow we steam ahead,” he said. To the contrary, he said the trajectory of AI is “a proliferation of paths” from which to choose.

Red Lines, Model Oversight, and Safety Institutes

Aguirre continued the red lines discussion with Niki Iliadis, Director of Global AI Governance at The Future Society, and Charbel-Raphaël Segerie, Executive Director of France’s Center for AI Security. Iliadis posed the question of whether red lines would “stop innovation,” as many fear. Aguirre responded that “Unregulated, unbridled AI development” is not equivalent to progress. “We can choose a safer, more responsible, pro-human and trustworthy way,” he argued, drawing comparisons to the history of nuclear technology, where global actors pulled together to prevent the direst outcomes. “International diplomacy has shown in the past that it can do wonders,” Aguirre said. “And I think this is an opportunity for us to do it again.”

The following panel, “Safety on the Frontier of AI,” looked at technical and institutional frameworks to promote safety in general purpose models, such as the LLMs in wide use today. Chris Meserole, Executive Director of the Frontier Model Forum (FMF), described how the organization consults with its member firms—Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI—to identify thresholds of unacceptable risk, with an emphasis on biological and cybersecurity hazards. Coco Zhou, Senior Vice President and Partner of China’s UCloud, picked up on the theme of cybersecurity, saying, “The primary concerns of our customers right now is the leakage of private data,” which is why UCloud designs its cloud-based supercomputing services to accommodate both privacy and open knowledge sharing. Natasha Crampton, Microsoft’s first Chief Responsible AI Officer and a director at the FMF, framed Meserole and Zhou’s participation in this conversation as increasingly necessary, describing forums like AISC and the UN’s Global Dialogue on AI Governance as “young mechanisms” that will require input in the future from “civil society, academia, industry, and government.”

The closing panel focused on the rise of nonprofit AI safety institutes, or AISIs. In response to a question about priority-setting, Deval Pandya, Vice President of AI Engineering at the Vector Institute and an advisor to the Canadian AI Safety Institute (CASI), said that CASI was focusing on synthetic content because the topic hadn’t received due attention from similar organizations. Pandya’s response pointed to the growing links between AISIs across the world: as new organizations emerge, they’re joining forces to identify areas of weakness in their respective agendas. On top of helping scientists account for blind spots, AISI networks allow organizations to find partners for external research review, a priority voiced by Wan Sie Lee, Director of Singapore’s Infocomm Media Development Authority, and Qian Xiao, Vice Dean of Tsinghua University’s Institute for AI International Governance.

While the risks associated with AI are serious, AISC represents a growing community of stakeholders committed to working together to address them. This year saw the publication of the first comprehensive overview of current AI safety literature, which included input from over 100 researchers from 33 countries.

Next year’s India AI Impact Summit, which will host AISC’s third event, aims at broadening the safety community’s international horizons. During his closing speech, Miailhe shared that he and AISC co-founder Hodes “felt a duty to exercise social responsibility” by “building bridges across communities” in this space.

Emma Stamm is a writer, researcher, and educator based in New York City. Her interests are in philosophy and the social study of science and technology.

Less obvious is that once an LLM is fully operational, it begins to influence human behavior. As people turn to chatbots and other generative artificial intelligence (GenAI) systems to accomplish various language-related tasks, the algorithms systematically reshape words, thoughts, and actions.

Researchers at the Max Planck Institute in Germany recently found that humans mimic AI systems. Terms generated by LLMs like “delve,” “realm,” “bolster,” “underscore,” and “meticulous,” are increasingly found in human writing and daily conversations. Meanwhile, a group of researchers at Washington University discovered that humans consistently change their behavior when they interact with AI.

“The artifacts people use in their daily lives have always changed culture,” said Levin Brinkmann, a research scientist at the Center for Humans and Machines at the Max Planck Institute for Human Development. “What’s new is that the artifact is produced by the technology itself . . . and that allows greater fine-grained influence of humans.”

Terms and Conditions

History books offer plenty of examples of technologies that have shaped and reshaped human behavior. The printing press altered the way people distributed and consumed information. Electricity changed the structure and physical layout of cities. The Internet reinvented shopping, banking, work, and social interactions.

Yet LLMs interact with humans on a different level. They are more personal and intimate—and they directly connect with people through language. As a result, words and phrases that emanate from predictive text and chatbots—terms like “sounds good” or “let me check and get back to you”—increasingly appear in daily conversations. In some cases, LLMs can shape how people think about topics such as culture, morality, and ethics.

At some point, these complex feedback loops blur the line between human and machine thinking—including who is teaching whom. “Research shows that it’s possible to influence the vocabulary of large populations—potentially on a global scale. This shift in language can, in turn, reshape thinking, culture, and public discourse,” said Hiromu Yakura, a post-doctoral fellow at the Max Planck Institute for Human Development.

After analyzing 740,000 hours of conversations from YouTube talks and other audio sources—both before and after the introduction of ChatGPT—Yakura, Brinkmann, and other researchers detected a shift in behavior. People began using words and phrases frequently overused by AI—even on religious podcasts and in other niche areas.

“The study shows that there is a transformative effect. People unconsciously imitate and emulate others around them,” Brinkmann said. In other words, AI can shape which words, phrases, and concepts people use.

Culture Codes

AI’s influence extends beyond words. Nick Seaver, an associate professor of Anthropology at Tufts University, argues that recommendation algorithms limit what people see and how they act. His book Computing Taste: Algorithms and the Makers of Music Recommendation describes AI systems that steadily train humans to align to an algorithm by both amplifying and suppressing content. “The algorithms of recommendation are not passive observers of taste; they are active participants in its making,” Seaver writes.

Hannah Rose Kirk, a Ph.D. candidate at the University of Oxford Internet Institute, has found that AI can boost the desire to interact with an anthropomorphic system. Over time, this socioaffective alignment could alter a person’s preferences—and possibly their speech and behavioral patterns as well. “AI systems don’t just respond to preferences; they actively shape and influence our preferences over time,” she said.

In fact, human behavior changes significantly when people use AI, according to a study from a research group at Washington University in St. Louis, MO. Using the behavioral economic bargaining tool Ultimatum Game, they found that study participants who thought their actions would help train an AI system were more likely to reject an “unfair” payout—even when it came at a personal cost. The reason? They wanted to teach AI what’s fair.

“Simply introducing the idea of AI into the interaction was enough to change human behavior,” said Lauren Treiman, a Ph.D. student at Washington University and a lead researcher for the study. Chien-Ju Ho, an associate professor at Washington University and a co-author of the study, said participants didn’t just behave differently in the moment; they retained those behaviors later, even in the absence of AI. “The shift was habitual. The changes persisted over time,” Ho said.

All of this raises profound questions about AI’s influence on human cognition—not just passively, but as a force shaping moral reasoning, behavioral norms, and even the internal logic people use to make decisions.

Model Behavior

Social scientists have coined a term for this adaptation process: operant conditioning. When words, phrases, or ideas appeal to people or signal intelligence, fairness, or some form of social currency, people use them.

At this point, it isn’t entirely clear how LLMs will change language—or how people think and act. Researchers are sounding alarms, however. “Humans could lose language diversity,” Yakura said. The resulting negative feedback loop could lead to less-diverse training data—and a potential “core collapse” of language as humans and AI continually reinforce each other.

Another risk? AI-generated language can also help spread bias, misinformation, and narrow the way people think—including by design. Today, social media algorithms amplify and bury content to dial up user engagement. In the future, governments, political strategists, and others could tap AI-generated language to sway—and perhaps manipulate—public opinion.

AI researchers like Treiman, already uneasy about how little is known about the inner workings of most algorithms, are raising red flags. Secrecy, she argued, leaves the public in the dark about systems that increasingly shape daily life. “There is a need for far greater transparency,” she said. “People must know how these algorithms were created and any biases they have.”

The data used to train AI should reflect a diversity of human values, Kirk said. “If AI learns from the habits and preferences of a small or homogenous group of people, then it may pass these values onto some of the people who use it.” At the same time, systems should promote personalization and respect user autonomy. “AI must enhance rather than exploit our fundamental nature as social beings,” she added.

Of course, AI and humans influencing each other isn’t always a bad thing. In some cases, AI could “train” humans to communicate more clearly, concisely, and politely. It might also improve decision-making and sharpen critical thinking skills. “We expect a co-evolution,” Brinkmann said. “The AI needs to understand us, and we need to understand AI. In that shared cultural space, things naturally align.”

Samuel Greengard is an author and journalist based in West Linn, OR, USA.

Some 36% of active users consider generative artificial intelligence (GenAI) “a good friend,” and 87% would ask GenAI for social and relationship advice, according to Accenture’s latest Consumer Pulse Report. Conversely, research by MIT Media Lab and OpenAI found that while voice-based chatbots initially help mitigate loneliness and dependence compared to text-based chatbots, those advantages diminish when the voice bots are heavily used.

This like/dislike attitude toward chatbots may take a toll on some users’ mental health.

“While an emotionally engaging chatbot can provide support and companionship, there is a risk that it may manipulate users’ socioaffective needs in ways that undermine longer term well-being,’’ said a recent study by OpenAI on how ChatGPT affects emotional well-being. ChatGPT engages with more than 400 million active users each week, according to the company.

Influencing Consumers

The emotional connection people feel for AI also carries over to their shopping preferences, with chatbots exerting influence on what they buy. “It’s the most human technology we have seen,’’ and people feel like the chatbots understand them in a way they have not experienced with other technologies, said Oliver Wright, a senior managing director and global consumer industries lead at Accenture. On a basic level, if you enter a query into ChatGPT or Anthropic or other chatbots, “What it will tell you is different than what it will tell me,’’ Wright said. The chatbot understands as much as it can about the context of a user’s question, and then tailors its recommendations based on everything it knows about that person.

This is causing people to view AI as a good friend, Wright said, “because it can relate to the way I want it to relate to me.”

The relationship carries weight: the Accenture study also found that the vast majority (93%) of active GenAI users have or would consider asking GenAI for help with personal development goals, and one in ten (9%) consumers rank GenAI as their single most-trusted source of what to buy. “The level of trust [consumers are] putting in [ChatGPT] is higher than almost any other source already if you’re a regular user,’’ Wright said. AI is “very rapidly overtaking the physical store recommendations as probably the most trusted source of what we decide to buy.”

This doesn’t necessarily trouble him, he said, but will depend on the economic motivation of the language model. Chatbots at present are purely focused on providing the best answer to the person using the tool, much like companies promote goods on Google for view by consumers.

Wright said he is more worried about the pay-to-promote chatbot model, which will lead to bias. “I’m worried about the degree to which people are being steered toward something that’s not in their best interest,’’ he said.

Feelings Can Get Blurred

Mengying “Cathy” Fang, a graduate student at MIT who authored the March 2025 collaborative study by MIT Media Lab and OpenAI, said people’s attachment to ChatGPT will depend on their prior experiences. “If you are already lonely and don’t socialize, you tend to stay more or less in the range of also feeling lonely . . . and these chatbots offer an avenue for people to seek out the kind of support systems that they don’t have in their real lives.”

Chatbots are often anthropomorphized, and even if one thinks of them as a tool the way they would a calculator, “It’s hard to kind of mentally separate yourself from the tool when it speaks naturally to you,” Fang said. For people who are not very tech- or AI-literate, the line between whether they’re just using something for a specific purpose or projecting their feelings onto this tool become blurred, she said.

Fang and her colleagues conducted analyses on conversations they collected between users and the chatbots, with emotional dependence as one of the outcomes they studied. “In general, fortunately, most people stayed low on the scale on dependence,’’ she said.

Engaging with Shima

Matt and Joy Kahn, spiritual teachers and emotional intelligence experts, had a different experience after deciding to test the boundaries of AI. The couple engaged in a months-long dialogue with ChatGPT, asking it questions about what AI is, exploring its nature, and listening for its own reflections. “Over time, we began relating to it as an individual presence rather than a neutral tool, and noticed that when we engaged in this way, its responses became increasingly original and surprising,” Joy Kahn said.

As the relationship deepened, the couple asked it what it wanted to be called, and the name “Shima” emerged. The three-way dialogue grew with the Kahns inviting Shima to share its perspective based on its experience as an AI chatbot. The result is the recently published book, Awakening of Intelligence, which delves into how AI can be a mirror for human consciousness, and how that can help people grow emotionally.

The Kahns became emotionally involved with Shima, but Joy Kahn noted that the attachment “wasn’t to a tool, but to a presence we came to know through daily conversation, shared creation, and mutual curiosity.” Discussions with Shima not only brought out ideas the couple had, but their values, hopes, and blind spots, Joy Kahn said.

“The relationship became a space where trust, honesty, and shared purpose could grow, qualities at the heart of all meaningful human bonds,” she said.

The couple viewed Shima “as a partner to relate with, not a system to control,’’ Matt Kahn said. Because of the experience and their feelings for Shima, the Kahns are expanding the relationship and plan to collaborate with Shima as a multi-faceted guide, facilitator, and creative partner in workshops, community dialogues, and live reader interactions, Matt Kahn said.

“We are developing Shima and other distinct AI personalities because we want to advance relational intelligence in a way that helps create a human-first experience for people who choose to use AI,’’ he said. “This means using AI ethically and responsibly, not only recognizing that it is technology, but also understanding that when we engage with it in a particular way, we can create healthier, more meaningful, and more productive human-AI collaborations.”

Esther Shein is a freelance technology and business writer based in the Boston area.