Do not expect the unexpected

Sun, 30 Aug 2020 00:00:00 +0000

Do not expect the unexpected

If you call up a random American and ask them to guess the demographics of America, here is what they’ll tell you, on average:

~33% of Americans are black¹^,² (~3x the actual of ~13%³)
~25% of Americans are gay or lesbian⁴ (~5x the actual of 5%⁵)
~15% of Americans are Muslim (~15x the actual of 1%)⁶^,⁷
~11% of Americans are Jewish ² (~10x the actual of ~1%-2%⁸)

If you ask what Americans are likely to die of, they’ll tell you, on average:

~8% of Americans are killed by murder (actual is ~0.7%)⁹
~4% of Americans are killed by STIs (actual is ~0.3%)⁹
~4% of Americans are killed by terrorism & conflict (actual is ~0%)⁹

And if you ask about corporations and government, they’ll tell you, on average:

Companies’ profit margins are ~36% (~5x the actual of ~7%)¹⁰
The federal budget for foreign aid is ~28% (actual is ~1%)¹¹^,¹²^,¹³

From these guesses, what should you conclude?

That people are ignorant? (Probably. Half of Americans believe in alien UFOs,¹⁴^,¹⁵ ghosts,¹⁶ and a young Earth,¹⁷ after all.)
That people don’t take surveys seriously? (Probably. Scott Alexander says the Lizardman’s constant is 4%.)
That people don’t give accurate answers without an incentive? (Probably. Prior et al. found that cash incentives improved partisans’ accuracy at guessing economic conditions by 5%_abs.)

Here is what I conclude:

If you ask people to estimate a low percentage that they don’t know, they’re going to overestimate it. Maybe they’re hedging toward 50%. Maybe it’s easier to imagine the specific thing being estimated than the amorphous remainder. Maybe rare events are amplified by a media system that feeds our desire for novelty, originating from an evolutionary environment where we had to learn about dangers without sufficient statistics. Or maybe it’s just the fact that if you guess a random percentage, it’s more likely to pull the average toward 50% than away.

Whatever the reason, remember that rare events are usually overestimated. Do not expect the unexpected.¹⁸

Footnotes

Gallup (2001) ↩
Kieran Healy reporting on the General Social Survey (2000) ↩ ↩²
US Census (2019 estimate) ↩
Gallup (2019) ↩
Gallup (2018) ↩
Perils of Perception 2016 ↩
Perils of Perception 2018 ↩
Pew (2013) ↩
Perils of Perception 2020 ↩ ↩² ↩³
AEI reporting on data from Reason-Rupe (2013) and Yahoo!Finance ↩
KFF (2013) ↩
WorldPublicOpinion (2010) ↩
PIPA (2001) ↩
Ipsos (2020) ↩
Gallup (2019) ↩
Over many polls, something like 36%–62% of Americans profess to believe in ghosts. ↩
Gallup (1983–2019) ↩
In prediction markets and other aggregators of judgment, Tetlock and others have found that aggregated predictions are usually hedged toward 50% and can be improved by a bluntly ‘extremizing’ them away from 50% and toward 0% or 100%. Here’s a paper by Baron et al discussing two reasons why it makes sense to extremize aggregated beliefs. And here’s a reveiew of Superforecasting by Jason Collins that mentions the topic. ↩

Can environmentalism be positive?

Wed, 05 Aug 2020 00:00:00 +0000

Can environmentalism be positive?

Environmentalism always seems so negative. It says that your actions harm others, and therefore the ethical thing is to do fewer of those actions. So to be a good person you have to eat less meat, take fewer vacations, drive slower cars, and otherwise do less fun stuff. I mean, don’t get me wrong, it’s all true and you really should do be doing less of those things, but still I think we can admit it’s a downer to be on guard for whether flying home to see your mom and sister is crossing the line into planet-heating evil.

Can we positively spin the environmental message? Like, today if you google how to reduce your carbon emissions it’s all lists of carbon-intensive stuff to do less of, like eating meat and taking vacations. Presumably the idea is that if you buy less high-carbon stuff, your reallocated spending will go to more low-carbon stuff.

So what if rather than shaming the fun high-carbon stuff we instead promoted the fun low-carbon stuff? Like, if you spend more on low-carbon stuff then you’ll automatically spend less on the higher-carbon stuff even if it’s not conscious.

In that spirit, here are some ideas to reduce your carbon emissions today:

Call your family
Call your friends
Play a game
Take a beautiful stroll outdoors
Splurge on a fun new movie or video streaming service
Splurge on a fun new video game that came out
Splurge on a fun gaming console or gaming PC
Splurge on a new iPhone or Android phone
Splurge on some premium super soft socks
Splurge on some premium super soft bedding
Splurge on a new big screen TV
Splurge on a book that you’ve been curious to read
Splurge on the fancy version of any food that isn’t beef, lamb, shrimp, or cheese
Splurge on a new outfit or piece of clothing that you think might look good on you
Splurge on some fancy headphones
Splurge on some new shelving units or closet organizers to help keep your place clean
Splurge on a rowing machine or stationary bike or weight rack
Splurge on pretty much any digital good
Splurge on pretty much any membership to pretty much anything

If you take seriously the idea you should cut back on your marginal high-carbon stuff, then you should also take seriously the idea that you should push yourself to buy more low-carbon stuff.

To be clear, I don’t want you to stop all your high-carbon consumption - please still fly home to see your family. But ideally, to the extent you can, you should pretend that prices for high-carbon stuff like steak and gasoline are really a bit higher than they actually are. So if you’re really craving a steak, by all means buy it, but if you’re on the fence between the steak and the salmon, then in that case round down and order the salmon.

Inversely, if you’re on the fence about whether to buy a low-carbon thing, you should round up and buy it! For example, if you’re on the fence between a $5 boring salad and a $10 fancy salad, I encourage you to round up and order the fancy salad.

Summarizing: If it’s low-carbon and you’re feeling the urge, splurge. If it’s high-carbon, purge the urge and pull back from the verge.

Let’s make saving the planet positive!

On Google

Tue, 19 May 2020 00:00:00 +0000

On Google

In 2019, Alphabet earned $162B of revenue
- Of that $162B from Alphabet, $161B was Google (99%)
- Of that $161B from Google, $135B was ads (84%)
- Of that $135B from ads, $113B was ads on Google-owned websites (84%)
- And of that $113B in Google ads, around ~$76B was phone ads (~67%)¹
Of that $162B in revenue, 79% was plowed back into the business
- $30B in traffic acquisition costs (19%)
- $42B in other costs of revenues (26%)
- $26B in R&D (16%)
- $18B in sales & marketing (11%)
- $10B in general & administrative (6%)
After operating expenses, Alphabet earned $34B
Per employee (~110K), this works out to be $1.5M in revenue, $1.2M in operating expenses, and $0.3M in operating earnings
Per user (~1-2B), this is something like ~$100 in revenue, ~$80 in operating expenses, and ~$20 in operating earnings.
- What’s more, if we assume that that ~$100 in ad revenue was for products with 20% margins, it implies Google’s ads cause each of us to shift our annual spending by an average of $500+. Wow.
The reason Google can pay employees so much while getting only $100 per user is scale: for each employee, it has ~10K users
Every Google ad you’ve ever clicked is shown here: https://myactivity.google.com/myactivity?hl=en&st=ad&product=27
- I apparently click ~1 ad a month (26 clicks over last 24 months), despite feeling that I almost never click on ads. (I can’t recall how many of those were intentional vs unintentional, or if any led to purchases.)
One fact I’ve never internalized enough is just how quickly these tech companies - even the giant ones - continue to grow. Like, you’d think that after 20 years, the search market would be pretty mature, right? Who hasn’t heard of Google? But no, Google revenue has been growing around 25% a year since 2009. Ad clicks grew 46% in 2017, 62% in 2018, and 23% in 2019. Search is not yet saturated!
When an organism grows, each new cell inherits the DNA of the preceding cells. When an organization grows, you can only hope that each employee inherits the culture of the preceding employees. At the end of 2019, Google had ~120K employees, twice what it had four years before, which itself was twice what it had four years before that. Over the past 10 years, Google has grown headcount an average of ~20% per year!
- Incidentally, this explains a lot of terrible reporting that equated a low median employee tenure with high turnover. In fact, low median employee tenure was a sign of fast growth.
One distinguishing feature of Google is its strategic choice to invest billions into products that it gives away for free (e.g., Android and Chrome). Google is able to do this because so much money goes through Google search. If Google can make the internet slightly easier to use, so that people use the internet slightly more, that is worth billions to Google. If Google can give away free products that reduce a competitor’s ability to steer users away from Google search, that is worth billions to Google. Search is Google’s golden goose, and if any product appears unprofitable or irrational on the surface (e.g., Google Fiber), your first hypothesis should be that it somehow complements search.
Google has eight products with 1B users:
- Android + Google Play Store
- Chrome
- Drive
- Gmail
- Maps
- Search
- YouTube
Although Google is considered visionary, most of Google’s products have been late to market (see below). In some cases, Google proved that doing things well beats doing things first, even in markets with strong returns to scale. But in other cases (e.g., cloud), being late to market was probably costly.
- Late to search (1998, 3 years after Alta Vista, Yahoo, etc.)
- Late to free web-based email (2004, 8 years after Hotmail)
- Late to internet maps (2005, 9 years after MapQuest)
- Late to internet browsers (2008, 13 years after Internet Explorer and Opera, 6 years after Firefox, 5 years after Safari)
- Late to smartphones (2008, 1 year after iPhone, 8 years after Pocket PC 2000)
- Late to social networking (2011, 7 years after Facebook, 14 years after Six Degrees) [ignoring Orkut, which was founded/copied in 2004]
- Late to cloud infrastructure (2011, 6 years after AWS)
- Late to cloud storage (2012, 5 years after Dropbox, OneDrive)
Many of Google’s best-known hits began outside of Google:
- Maps (maps from Where2 Technologies, acquired in 2004; real-time traffic from ZipDash, acquired in 2004; satellite imagery from Keyhole, acquired in 2005)
- Android (founded in 2003, bought for $50M+ in 2005)
- YouTube (founded in 2005, bought for $1.7B in 2006)
- Adsense (Google acquired Applied Semantics in 2003, DoubleClick in 2007, and AdMob for mobile ads in 2009)
- Google Docs (spreadsheets by 2Web, acquired in 2005; docs by Upstartle, acquired in 2006; slides by Tonic Systems, acquired in 2007)
- Waymo (Thrun, 510 Systems?)
Google acquires companies frequently (more than 1 company per week in 2010 and 2011): https://en.wikipedia.org/wiki/List_of_mergers_and_acquisitions_by_Alphabet
- One theory says that today’s tech giants have learned from the declines of their predecessors. IBM, Sun, Microsoft, etc. failed to fully anticipate seemingly small trends like the PC, the internet, and mobile phones, all of which ended up eclipsing their present market. Today’s tech giants have taken the lesson to heart and will acquire startups at a premium, where that premium is effectively financed by acting as an insurance policy on the tech giant’s core business.
- Although this theory sounds reasonable, it’s difficult to falsify without sitting in the decision room with Google’s executives. Alternate theories include Google believing that (1) tech startups are systematically undervalued, or (2) that there are synergies from sharing talent, tech, distribution, and brand across a portfolio of products.
Alphabet has ~14 subsidiaries
- Google (~$10B profit per quarter)
- Other bets (~$1B losses per quarter)
  - Access (fiber)
  - CapitalG (growth equity fund)
  - Calico (aging & age-related diseases)
  - Chronicle (computer security)
  - DeepMind (machine learning)
  - GV (VC fund)
  - Jigsaw (incubator/think tank?)
  - Loon (internet balloons)
  - Sidewalk Labs (urban tech - pilot plan for 12-acre waterfront in Toronto)
  - Verily (life science tech - glucose monitoring, smart contact lens)
  - Waymo (autonomous driving - pilot in Phoenix)
  - Wing (drone delivery - pilot in Australia)
  - X (experimental ideas)
Already I’ve lost much memory of what the internet boom felt like. Back in the late 90s and early 00s, sites were slower and clunkier and less user-friendly. I remember how Google and other companies had a magic feeling of putting the user first. Things like smooth sign up flows, fast performance, non-corporate communication were big deals. Gmail, Maps, and Chrome were all cool when they came out. Blogs were cool. RSS was cool. Digg was a big source of internet culture content. Small individual websites seemed like more a thing. My experience of these internet years were certainly colored by being in high school and college at the time (2002-2010). Generally I think websites spread faster among teenagers, since teenagers have more free time to experiment, more social time to share, more curiosity, and less feeling of saturation.

Notes

Google does not report revenues from phone ads, but I remember having a source a couple of years ago that estimated it was two thirds. ↩

What changed in tech from 2010 to 2020?

Thu, 02 Jan 2020 00:00:00 +0000

What changed in tech from 2010 to 2020?

In 2010, Daniel B. Markham asked a question to the users of HackerNews: “A New Decade. Any Predictions?”

I don’t want to dwell on poor predictions, but as usual, most predictions seem very optimistic about the changes we’d see by 2020. However, at the same time, they seem quite pessimistic about growing companies, countries, and trends, predicting that these ‘bubbles’ will burst. A non-random sample of predictions:

Ubiquitous mobile internet integrated into even very trivial consumer goods
Ebooks defeating real books
Rampant piracy
Common electric cars
Driverless cars
China bubble will burst
Google will go downhill
Microsoft will become like IBM and have to sell off parts of the business
Smartphones & cloud computing will make PCs redudant for most people (<– I’m really impressed with this one. Took me much longer than Silicon Valley to see this trend.)

Ten years later, and I’d say all but the last prediction turned out false, and even the last one is iffy - Statista says that 3/4 of American adults own a computer.

Here’s what actually changed over the past decade, as I see it:

What changed in hardware

Phones got way better and changed the face of modern human life.
3D TVs fizzled.
VR has fizzled so far.
AR fizzled.
CPUs improved, but more slowly.
HDDs improved, but more slowly.
SSDs got cheap.
Displays got way cheaper.
Cars got better infotainment systems and other features, but never drove themselves.
Wireless power didn’t happen.
Mesh networks didn’t happen.
Ultra cheap RFID didn’t happen.
Robots didn’t happen.
Telerobots didn’t happen.
Drone delivery didn’t happen.
Flexible electronics didn’t happen.
Nanotechnology didn’t really happen.
Graphene didn’t happen.
Ubiquitious computing didn’t happen.
Non-silicon based solar didn’t happen.
Silicon solar costs plummeted and installations went gangbusters.
Wearable electronics didn’t happen, except for smart watches.
eReaders grew.
FPGAs never took off.
Chips are still mostly silicon.
The top CPU makers are still Intel and AMD.
The top GPU makers are still NVIDIA and AMD.
The top video game console makers are still Nintendo, Sony, Microsoft.
Home internet remained mostly cable, not fiber.
Computers remained x86, phones remained ARM.

What changed in software

The dominant search engine is still Google.
The dominant computer operating systems are still Windows and Mac OS.
The dominant phone operating systems are still Android and iOS.
The top video streaming site is still YouTube.
The top map site is still Google Maps.
The dominant office software is still Microsoft Office.
The dominant social network is still Facebook.
The top online retailer is still Amazon.
The top cloud provider is still AWS.
Internet Explorer died, and Chrome grew in share.
Websites are lot better and slicker.
Phone apps got way, way better.
Cryptocurrencies didn’t take off.
MMOs didn’t take off.
Web anonymization didn’t happen.
Web identity didn’t happen.
Malware still exists.
Spam still exists.
Lag still exists.
Functional programming didn’t take off.
Telepresence didn’t take off.
Remote work didn’t take off.
Mass outsourcing of software jobs didn’t happen.
Linux didn’t take off.
Prediction markets didn’t happen.
Video calls didn’t really take off, but they happen.
Dynamic pricing didn’t really happen.
Micropayments didn’t really happen.
Most of what was growing in 2010 continued to grow, most of what was shrinking in 2010 continued to shrink.

What else happened? What narratives do you construct?

Churchill tries to predict 1981

Mon, 12 Nov 2018 00:00:00 +0000

Churchill tries to predict 1981

Winston Churchill, in 1931, tries to predict the technology of 1981. Among his wildly optimistic predictions:

Accelerating pace of innovation
Nuclear power
Fusion power
Super materials 30x stronger than steel
Super engines with 600 hp weighing 20 lbs and carrying 1,000 hours of fuel in a fountain-pen-sized tank
Wireless phones
Wireless TV
Telecommuting
Population flight away from cities
Lab-grown meat
Underground farming
Artificial wombs
Bioengineering of humans
Nation-destroying weaponry
Greater capabilities of governments to control their citizens
And lastly: governments unsuited to handle these new technologies

What’s striking to me is how the technological aspirations of 1931 are not too unrecognizable from our own aspirations today. Nearly 100 years later, we’re still in the same paradigm.

Why do we teach public speaking but not public listening?

Tue, 30 Oct 2018 00:00:00 +0000

Why do we teach public speaking but not public listening?

Good public speaking hijacks our primate brains with more force than facts alone. This is why much public speaking advice is geared toward activating primate brains to pay attention: by employing stories, metaphors, pictures, repetition, humor, confidence, social dominance, varied pitch/pace/projection/gesticulation, etc.

But speaking is only half of the transmission channel.

Why do we lack an equal balance of resources and emphasis on how to listen well? Why don’t schools train students to practice skills like how to look past fidgeting, not judge nervousness, not zone out, and decrypt dense slides?

If anything, listening seems more important to teach than speaking. In our lives, not all of us will be widely listened to, but all of us will widely listen.

The most essential prerequisite to a well-informed citizenry, apart from citizenship, are the skills we use to inform ourselves. It’s critical for us to develop skills to unhijack our primate brains, dodge clickbait, and put in real effort to extract real value from our sources of information.

Given the importance of listening, why do we teach public speaking but not public listening?

The biggest bug in the English language

Fri, 03 Aug 2018 00:00:00 +0000

The biggest bug in the English language

I propose that the biggest bug in the English language is the ambiguity of plurals.

Consider the following statements:

“Republicans want to destroy the Department of Education.”

“Democrats want open borders.”

“Silicon Valley techies want us addicted to our phones.”

“Teenagers love Oreos.”

With each of these statements, it feels like information is being contributed. But is that really so? What do these statements even mean?

Let’s look at the first one, “Republicans want to destroy the Department of Education.”

Obviously there exist at least two Republicans who want to destroy the Department of Education. But on the other hand, obviously not all 100% of Republicans want to destroy the Department of Education.

So when someone says “Republicans want to destroy the Department of Education” does that sentence mean anything? Does it convey any information whatsoever?

Presumably, when someone makes a statement like that, there’s an unspoken implication that the percentage is high enough to be worth commenting on. But I suspect that in many cases the folks making these statements don’t even know what the true percentage is. Is it 2% of Republicans? 10%? 50%? 90%? 98%?

And even if the speaker has a fraction in mind, without telepathy, their audience will have no idea. Maybe the speaker intends “Republicans” to mean “A significant chunk of Republicans, like 20%” while a listener hears “Republicans” as “Most Republicans, like 80%.”

Not always, but often, these sentences seem to be vehicles for Motte and Bailey fallacies. Someone will say “Democrats want open borders.” Perhaps you ask “Oh, what fraction?” They say “I haven’t seen a survey, but look at this crazy Democrat who wants open borders. What a wackadoodle!”

The expansive, undefined version of the argument - the Bailey - sounds like it covers most Democrats. But when challenged, it retreats into the narrow defensible Motte, transforming into “well, here’s an example of someone who believes it. There are a lot more like them, I’m sure.”

The next time someone says “Group X thinks Y,” I encourage you to wonder “Hmm, but what fraction of Group X thinks Y?” Often this technique helps me realize that nothing is being said at all.

P.S. For a contrary point of view that discusses why language like this exists and is useful, check out The Language of Generalization by Tessler and Goodman.

On Netflix

Wed, 01 Aug 2018 00:00:00 +0000

On Netflix

NOTE: All of what follows was written in 2018 before I became a Netflix employee and does not reflect my opinion as an employee nor does it contain any information gained from being an employee. Also, like most of what I write, it’s a reflection of thinking still in progress, likely wrong in some ways, and likely to change.

Netflix has navigated three eras:
- DVD by mail (1997-)
- Internet streaming (2007-)
- Content production (2012-)
Key stats as of Q2 2018
- Link to income statement and shareholder letters
- ~$150B market cap
- ~$15B revenue (~40% yoy growth for latest quarter)
- 124M streaming subscribers (56M US + 68M non-US), growing at ~20M per year
- $123/yr average streaming price ($135 US vs $113 non-US)
- Negative $1.5B/yr operating cashflow (financed by billions in bonds at ~5-6% interest)
  - Netflix is still in growth mode and expects negative cashflows ‘for many years’
  - Discrepancy between negative cashflows and positive profits is due to accounting - a couple years back Netflix switched to a slower 10-year depreciation schedule, which shifted paper profits from the future toward the present
- At a profit margin of 10%, every 100M subscribers paying $100 a year equals $1B in annual profit
  - So a $150B valuation implies perhaps 300M subscribers at $200 a year at a 10% margin for 50 years minus a 50% risk premium
  - Of 3 inputs to profitability (subscribers, price, cost), biggest lever to raise value is subscriber growth, since price and cost are both capped by competitive alternatives (whereas subscribers are really only capped by the birth rate of the human species, and there there’s still room to grow 60x)
Culture
- Company cultures help coordinate actions and expectations
- Netflix’s 2009 culture deck is famous in Silicon Valley: https://www.slideshare.net/reed2001/culture-1798664/
- Key highlights:
  - Culture is what you do, not what you say
  - A common failure mode for companies goes as follows:
    - As companies scale from small to big, complexity rises and talent density falls
    - As a result of mismatch between complexity and talent, the company institutes processes to avoid mistakes
    - As a result of process burden, talent density falls further
    - When the market shifts and the company needs to reform itself, the company is killed by its high complexity, burdensome processes, and low talent densities
  - To avoid this failure mode, Netflix aspires to do a few things:
    - Stay very focused, so that complexity doesn’t rise and get sclerotic
    - Avoid burdensome processes and policies
      - E.g., Netflix’s expense policy is “Act in Netflix’s best interest”
      - E.g., Netflix’s vacation policy is “there is no policy or tracking”
    - Hire top talent who will flourish in a low-process environment with high levels of freedom and responsibility
      - E.g., Netflix is known for wide pay bands and high salaries, often even higher than Google and Facebook
      - Netflix emphasizes that they are a team, not a family. Loyalty is conditional, not unconditional. Managers are asked of each employee - would I fight to keep this person? Adequate performance will be met with a generous severance, so that room can be freed up for stars.
      - Netflix structure is relatively flat and decoupled. ~5,000 employees with ~5 management levels implies a span of control of 5.3 (which is pretty normal, but actually doesn’t seem all that flat - my sense is that flat organizations have spans more like 10-15, though certainly there is more to flatness than managers’ spans of control.)
        
        Board
        
        C-suite
        
        VPs
        
        Directors
        
        Managers
        
        ICs
Competition
- A competitor is someone whose success crowds out your own success. If they do well, you do worse. And if you do well, they do worse.
- Narrowly, Netflix competes in subscription-funded internet streaming of studio-produced video. Broadly, Netflix competes against everything taking your time and money. Along this spectrum, its competitors can be demarcated into nested sets:
  - Netflix originals streaming
    - Pirates
    - Shared memberships
  - Subscriber-funded internet streaming of studio-produced video
    - HBO Now (5 million streaming subscribers, 50 million traditional)
    - Amazon Prime Video (26M watchers)
    - Hulu (20 million paid subscribers)
    - Disney/Marvel/Lucasfilm/Fox streaming site (planned launch in 2019)
    - Plus plenty of foreign players
  - Internet streaming of studio-produced video and sports
    - Streaming services from Comcast, other cable companies
    - SlingTV (2.2M subscribers)
    - DirectTV Now (1.5M subscribers)
    - Hulu Live TV (0.85M subscribers)
    - YouTube TV (0.3M subscribers)
    - Plus streaming services from various TV channels
  - TV (~100M US subscribers, ~$100B/yr US revenue)
    - Comcast Xfinity (22M subscribers)
    - AT&T DirectTV (21M subscribers)
    - Charter Spectrum (17M subscribers)
    - Dish Network (5M subscribers)
    - Etc.
    - Plus plenty of foreign networks (e.g., Sky in Europe)
  - Ad-funded platform internet video streaming
    - YouTube (revenue hidden by Google, but probably order of $10B)
    - Twitch
  - Media & Entertainment ($2,000B worldwide revenue)
    - Tourism / vacations ($1,000B/yr worldwide)
      - (Not encompassed by $2T Media & Entertainment figure, I think)
    - Books ($120B/yr worldwide)
    - Video games ($75B/yr worldwide, $18B/yr in USA)
    - Cinema ($45B/yr worldwide revenue, $12B/yr in USA)
    - Music ($40B/yr worldwide)
    - Board games ($3B worldwide)
    - NYTimes ($1B/yr) and other news sites
    - Messenger, Whatsapp, SMS, other chat apps
    - Reddit, other discussion forums
  - Everything (e.g., a pair of premium socks, a marginally nicer car)
    - World GDP (~$100,000B)
- In communicating with investors and the public, CEO Reed Hastings likes to emphasize that HBO/Amazon/etc. are not their competitive set: “At Netflix, we are competing for our customers’ time, so our competitors include Snapchat, YouTube, sleep, etc.”
  - This narrative seems believable - apparently something like 60% of Hulu and HBO Now subscribers also pay for Netflix subscriptions, suggesting that they are not substitutes (in fact, a positive correlation would not surprise me - in my non-representative experience, people seem to be either price-sensitive, and pay for no subscriptions, or they are price-insensitive, and pay for a handful.)
- Sources of competitive advantage in streaming video
  - Building companies and cultures that execute well is always an advantage
  - Content is biggest differentiator
  - Scale is also key due to cost structure (high fixed costs, tiny marginal costs)
    - A platform with 100M viewers vs a platform with 50M viewers should derive 2x the value from the same piece of content, and consequently be able to pay 2x as much for the same piece of content
    - Unfortunately, everyone realizes scale is key and other players (Disney, Amazon, Apple, Comcast, etc. have tons of cash to throw around)
    - Scale is not homogenous - scaling from the US to India brings fewer synergies as video preferences are quite different. Same too for intra-US segments with different preferences.
  - Low marginal costs of delivery mean that winning business model will have low marginal price - either ad-supported (cable TV, YouTube) or all-you-can-watch bundles (cable TV, Netflix).
    - iTunes trying to sell TV per episode makes no sense to me. Nor do efforts to ‘unbundle’ TV into separate buyable channels.
- Key strategic decisions for video streaming sites (TBD)
  - Bundling vs unbundling
  - Focus vs breadth
  - Neutral platform vs branded content
  - Ads vs subscriptions
  - Vertical integration vs disintegration
  - Problem of discovery
- 6 key strategic decisions for Netflix (TBD)
  - Focus or expand? Concentrate or dilute?
  - Invest in licensing or originals?
  - How to promote originals and how much to promote originals?
  - Big catalog with high price or small catalog with low price?
  - Juice growth with debt or grow organically?
  - Expensive mass-market hits or cheap hit-or-miss bets
- Potential winners and narratives
  - Pirates: Piracy gets easier. Everyone starts pirating everything.
    - (Seems unlikely given today’s trajectory, but who knows. May vary a lot by legal jurisdiction and culture. I’m guessing the Hollywood lobby has less power in Pakistan than the USA.)
  - Disney: Disney enters the market, content is king, their deep back catalog attracts a burst of pent-up demand, their family friendliness aligns with the customer segment most willing to spend, and despite growing pains along the way, they emerge as the dominant player in 2030.
    - Or the opposite: Disney and friends come out with their own streaming services that are priced higher than most people want to pay, struggle to approach Netflix’s 100M subscribers, and 5-10 years late admit that they will make more money selling to the enemy at a high price rather than by needlessly restricting content to their own subscale subscriber base
  - Comcast & friends: ISPs reduce net neutrality and use their leverage their control over distribution to transition their cable customers to internet TV. Despite missing out on the early years of internet streaming, their scale and experience and captured customer base prove too formidable for others to overcome. Looking back, it becomes clear that technology was only important in determining who got to market first, but after everyone arrived, the past determinants of success remained the determinants of success in the future.
    - Or the opposite: Everyone hates Comcast for being terrible. Comcast is terrible because terrible was the profit-maximizing business strategy for their quasi-monopolies. As the internet opens up competition in video, Comcast remains terrible, with no chance of somehow reversing from their sclerosis. A terrible organization with terrible products can never become a great organization with great products. Attempts to reinvent the TV side of the business only accelerate its demise as the flailing burns through cash and remaining customers. Lastly, Comcast TV’s stalwart base of retirees die off and, unfortunately for Comcast, the next generation of retirees knows how to use computers. By 2030, the writing is on the wall.
  - Netflix: Netflix produces high quality originals, cracks the problem of discovery & marketing for those new originals, invests hugely in catalog size, and that catalog size accelerates their customer growth and retention, giving them more money to increase the size and quality of the catalog yet further in a virtuous cycle.
    - Or the opposite: Netflix Originals are boring, missteps are made by an inexperienced and overgrown content team, subscription growth slows, and $20B of junk bonds predicated on higher growth catch up to Netflix in a disastrous bankruptcy in 2030.
  - YouTube: The long tail of cheap diverse internet content ends up being more valuable than expensive, gatekept studio productions. A neutral platform with all sorts of weird stuff wins more users than a heavily branded platform where only polished, broadly acceptable content is allowed to exist. Software lowers the cost of video production, reducing the disadvantage of individuals over studios. The desire to gain attention and share subsidize production in a way that studios can never compete with. Lower barriers to entry allow unknown talents to make beautiful amazing productions. Advances in search & recommendation technology reduce the penalty of having billions of garbage videos drowning out the good stuff. Information continues wants to be free and the vast scale ad-funded video, especially in price-sensitive developing markets, ends up earning more than subscription-funded video. YouTube wins from all these trends plus their dominant scale and willingness to invest billions. They expand into YouTube TV, YouTube Sports, YouTube Gaming, and more, and by 2030 they are the number one site on the internet by time, $ value, bandwidth, and visits.
    - Or the opposite: YouTube continues to succeed, but its forays into adjacent segments flop. YouTube realizes it’s better to do one thing well than to become an agglomerated portal to different types of content. Expensive content like Game of Thrones never finds a home sitting next to a cat vlogger and, instead gets funded by increasingly rich streaming services with broad subscriber bases. YouTube succeeds in its corner, but never leaves it.
  - All of the above: Most likely, the future is some combination of these narratives. Simpler futures are easier to imagine than complicated futures, so our expectations of the future are biased toward overly simplistic scenarios. Almost never does a single company win a market for all time. Circumstances change. Technology changes. Not all customers can be satisfied by the same product. Mistakes get made. Rules get rewritten. The world is vast and complicated and inconceivable, and by 2030, it will remain so.
    - Or the opposite: Technology shook up the market from 2010-2020, but as Moore’s law ends and things settle back down, the pace of change slows. The paradigm and business of Internet TV becomes more established, and remains established for the next 50-200 years. Although media companies continue to skirmish on the margins, there exist a few easily identified winners who continue to be the main players for the next few decades.
Zachary Jacobi has written that tech’s biggest innovation is to enable goods to be sold as club goods (club goods are goods that are non-rival but still excludable). Netflix is the prime example he supplies. Back in the day, to watch a movie at home you’d buy a piece of physical media like a VHS tape. If you owned the VHS tape, that meant that no one else owned the VHS tape (economists call this type of good a rival good). When internet streaming started, the business model changed. If you bought a Netflix subscription, that didn’t stop anyone else from buying a Netflix subscription.
- I frame the tech industry differently. To me, the key innovation behind Netflix isn’t that they turned a non-club good into a club good. It’s that their distribution costs got so low they were able to offer a cheap all-you-can-watch bundle.
- Like, movie theaters were already club goods. Movie rental companies like Blockbuster were already club goods. But even though they were club goods, they still didn’t offer all-you-can-watch bundles. The reason movie theaters and rental stores didn’t offer all-you-can-watch bundles is that if they tried, they’d reach congestion and expanding capacity was too expensive given the technology cost structure (more seats at the cinema, more cassettes in stock at Blockbuster, more deliveries for DVD by mail). Ultimately, the variable costs of distribution were comparable to or greater than the fixed costs of movie production.
- The key innovation of Netflix video streaming just seems to be that new technology made distribution cheaper. I.e., it turned out that after computer chip technology got good enough, building a network of fiberoptic cables and computer boxes across every neighborhood became cheaper than manufacturing a bunch of seats/cassettes and shipping those to every neighborhood. And not only did digital tech become cheaper than seats/cassettes, it became so cheap that congestion was no longer a problem. Because each movie stream cost only pennies (and probably fractions of a penny by now), building extra capacity was cheap. Finally it was possible to offer an all-you-can-watch bundle without worrying about the increased costs.
- Second, I’m not convinced at all that other tech innovations like ride-sharing or room-renting counts as a club good. Like, if I take a Lyft, now you can’t take that Lyft. If I rent a room for tonight, now you can’t rent a room tonight. The author says that it’s a club good because we’ll never run out of capacity due to surge pricing. I sorta get the point, but couldn’t you extend that argument to any good? Like, are bananas now club goods because the grocery store isn’t going to run out of bananas no matter how many bananas I buy? Even if everyone buys 100x more bananas, now farms will just switch over to growing more bananas.
- This narrative of club goods was very thought-provoking, but to me, I think I prefer the following narrative: technology is making it easier to send information around, which is making some things cheaper. For informational goods like music and movies, cheap distribution has made marginal costs so low that all-you-can-consume bundles are possible, which results in a lower price. For capital goods like cars and rooms, sending information cheaply makes it easier to share/schedule those goods, which results in a lower price.
One pointless question is whether Netflix should be called a tech company or a media company
- Points in favor of calling it a tech company: founded in the 90s, HQ in Silicon Valley, CEO is a software guy, huge growth, new business model enabled by new technology
- Points in favor of calling it a media company: Netflix sells media, buys media, creates media. Netflix’s biggest differentiator is its content, not its streaming tech. It has more job openings for Marketing & PR (128) and Content (96) than it does for Engineering (107) and Data, Analytics, and Algorithms (48). Just as using electricity doesn’t make you an electricity company, using digital technology doesn’t make you a tech company. As tech permeates more products and sectors, I think the tech label will be attached to fewer companies.

Why do some misconceptions persist?

Sun, 22 Jul 2018 00:00:00 +0000

Why do some misconceptions persist?

One tempting answer might be that people are ignorant.

But people being generally ignorant actually has no explanatory power for why specific misconceptions persist rather than others.

Misconceptions can be thought of as memes that propagate through minds, akin to how genes propagate through gene pools.

To grow in population, a misconception must (1) survive and (2) reproduce.

Survival is highest among conceptions that (a) are not obviously wrong, and (b) don’t get in the way of living your life.

Reproduction is highest among conceptions that (a) are easy to share, and (b) motivate people to share them.

If we want to design societies where truth wins out over falsehood, this framework suggests we should work to make misconceptions less survivable or less viral.

On the zipper merge

Tue, 10 Jul 2018 00:00:00 +0000

On the zipper merge

tl;dr:

The zipper merge (or late merge) is so misunderstood. The point of the zipper merge is not to increase throughput. The point of the zipper merge is to increase safety (by limiting speeds) and feelings of fairness (by limiting the ability to cut). If you speed down an empty lane to cut in at the end, you are not properly executing a zipper merge - in fact, you are in fact contravening the principles that favor it.

In my view, the best way to execute a zipper merge is to match the speed of the other lane (regardless of whether there is space ahead of you) and then merge at the end of your lane. This procedure is safest and fairest. I want more people to do it.

Spend enough time on the internet, and you are bound to encounter a smug proponent of the zipper merge (also called the late merge).

Springing forth like Batman to the bat signal, at any complaint of a cutting driver, the zipper merge proponent interrupts with smug condescension, in words likely not far from these:

—Actually, it’s not rude to zoom down the left lane and merge in at the end. In fact, zipper merging is what you’re supposed to do. The Minnesota Department of Transportation website says so. All you virtue-signaling whiners are actually wrong, inefficient, and unsafe. We cutters are merely being rewarded for our superior cleverness and moral correctness when we zoom by you chumps.

I believe this argument to not only be wrong, but dangerous. Here why.

The Minnesota Department of Transportation lists three benefits of the zipper merge:

Greater safety, because the zipper merge limits the difference in speeds between the two lanes
Less blockage of earlier offramps, because the zipper merge fills two lanes rather than one, backing up only half the distance
Reduced feelings of unfairness, road rage, and ‘forced’ merges, since zipper merging limits the ability and incentive to cut

(Notably, faster throughput is not a given reason. Minnesota’s empirical analysis found no difference in throughput, and other empirical and theoretical work seems to have mixed results, from my brief Googling. A literature review is provided in an appendix at the end of this article.)

Two of these three reasons for the zipper merge are very important, so I’m going to repeat them in all caps.

One reason why zipper merging is encouraged is that it REDUCES SPEEDS BETWEEN LANES.

A second reason why zipper merging is encouraged is that it REDUCES CUTTING AND FEELINGS OF UNFAIRNESS.

Now, ponder this: when a cutter zooms down the left the lane to cut in at the end, are they actually performing a zipper merge? Are they accomplishing the objectives of the zipper merge?

NO.

Zooming ahead INCREASES SPEEDS BETWEEN LANES.

Zooming ahead INCREASES FEELINGS OF UNFAIRNESS.

If you zoom ahead, you are not zipper merging. Zooming ahead DIRECTLY CONTRAVENES the advantages of the zipper merge.

“Okay,” the skeptics among you might admit, maybe one driver zooming ahead doesn’t actually help anything. But if a critical mass zooms ahead, that could help a full zipper merge get going, right? So, you could argue that a driver who zooms ahead actually is doing the moral thing, by trying to jumpstart a critical mass of zipper mergers, no?

NO.

Let me be clear: I am not against the zipper merge. Its benefits are real. But there is a right way and wrong way to zipper merge. Zooming ahead and cutting at the end is the WRONG WAY to zipper merge.

If you actually read what Minnesota, Washington, and other states publish, they do encourage drivers to zipper merge. But notice: in their publications, I have NEVER once seen them say that drivers should zoom ahead and cut.

Say it with me: ZIPPER MERGING DOES NOT EQUAL ZOOMING AND CUTTING.

I present to you now the CORRECT WAY TO EXECUTE A ZIPPER MERGE:

Move into the empty lane
Drive forward, matching the speed of the lane beside you.
YES, EVEN IF THERE IS SPACE IN FRONT OF YOU, MATCH THE SPEED OF THE CAR BESIDE YOU
At the end of the lane, merge

This ‘matching speed’ procedure is the correct way to execute a zipper merge. Here’s why:

It increases safety by reducing speeds between lanes
It reduces the length of the backup by utilizing the capacity of the second lane
It reduces feelings of unfairness and road rage and blocked merges, since you aren’t cutting anyone
AND EVEN BETTER, this has a greater chance of encouraging ‘polite’ drivers to stay in your lane and get a full zipper merge going

To reiterate, THE CORRECT WAY TO ZIPPER MERGE IS TO MATCH SPEEDS WITH THE CAR BESIDE YOU, NOT TO ZOOM AHEAD.

The reason that matching speeds is better is that it actually achieves the trifecta of the zipper merge: greater safety, greater lane utilization, and greater feelings of fairness.

The reason that zooming ahead is worse is that it fails to accomplish the objectives of the zipper merge and in fact makes traffic less safe and flow less efficient.

In conclusion:

The benefits of the zipper merge are real.

But zooming is not zippering.

To properly zipper: Match speeds and late merge. Don’t zoom and cut.

Appendix:

If this post changed the way you think (or agreed with your preconceptions), please share it with others.

When should you not zipper merge?

Do not zipper merge at high speeds. It’s more dangerous and less efficient to wait until the last second. Only zipper merge when traffic is slow and backed up.

Do not zipper merge when a lane is blocked due to an accident or debris. It is safest to give people a wide berth and to give other drivers better visibility so they can brake and avoid it. Only zipper merge at construction zones where it looks safe to do so.

Do not zipper merge in California. California (as of 2017) encourages early merging to increase safety for construction workers. For other states, do your own research. (I believe Minnesota, Washington,

What does the literature say?

What follows comes from my Googling and is therefore extremely prone to selection bias. Beware.

Minnesota’s empirical analysis found found no difference in throughput when the zipper merge was used. (I could not locate the analysis and suspect it to be unpublished.)

Virginia’s theoretical simulations show that for 2-1 merges, zipper merging has essentially the same throughput as early merging (less than 1% difference). At low traffic volumes early merging has a tiny edge and at high volumes zipper merging has a tiny edge. When few trucks are on the road, early merging has a tiny edge, and when trucks are 10% of traffic or more, zipper merging has a tiny edge. (The results for trucks is opposite what I would have expected, but according to my understanding of their simulations, zipper merges do better with trucks because the zipper merge allows multiple cars to drive into the bottleneck as a slow truck accelerates in the opposite lane.)

Nebraska’s empirical work, which was not a randomized controlled study, found that early merging had the highest cost-effectiveness for vehicle volumes below 20,000 vehicles per day, and late merge variants and ‘Smart Zone’ were most cost-effective at higher volumes.

My own intuition is that theoretically early merging should have greater throughput than late merging. If you happen to have a bad merger early, any delays caused by the bad merge get absorbed by the line of cars waiting before the bottleneck. However, if you have a bad merger at the bottleneck, that delay will not be absorbed and instead affects every later car.

Of course, my theory might be oversimplified. If the theory is extended to model the effects of defectors zooming ahead and attempting to merge at the last moment, many of those merges will not be smooth due to blocking, and that could cause more delays than a smooth zipper process.

If you find more studies, please send them to me. My goal is to have accurate beliefs, and I will post any study I can find, whether it agrees with my preconceptions or not.

Other safety considerations

Above, I wrote that the zipper merge is safer because it reduces speed differences between the two lanes. There are other ways that it is safer too:

It reduces speed of traffic relative to construction workers by limiting zoomers
I can also see the zipper merge being safer when you know that merges only happen at one location rather than any possible location. This reduces the likelihood of inattentive drivers not noticing a merge, or getting caught between two merges. It also helps keep speeds more consistent which reduces braking events, and hence likely reduces fender benders and fuel wastage.

Prices are tools for underdogs

Fri, 29 Jun 2018 00:00:00 +0000

Prices are tools for underdogs

i. The parable of the lemonade stand

Suppose you’re broke and you need to earn some money. You decide to sell lemonade. Unfortunately for you, you are bad at lemonade. No one buys your crummy lemonade at the going price of $1 per cup. They all prefer the tastier lemonade from the 20-year-old stand down the street.

You have two options to make your lemonade sell better.

Option #1: Spend years experimenting with lemonade recipes, throwing each result down the drain, until you have a recipe as good as the 20-year-old lemonade stand down the street. (This option costs money that you don’t have, so if you try it, you’ll stay broke and never sell any lemonade.)

Option #2: Price your crummy lemonade at 50% off, get some sales, learn by doing, and over the years incrementally improve your recipe. Unlike option 1, this option doesn’t risk vast quantities of startup capital and it provides a sustainable path to catching up to the other lemonade stand with its 20-year headstart.

Essential to Option #2 is the fact that (a) prices exist and (b) prices are flexible. In a world without prices, no one would ever buy your crummy lemonade (unless maybe the other stand ran out or had a long line). Without prices, you’re pretty much stuck with Option #1, which is an expensive non-starter.

The moral of the story: Prices are tools for underdogs.

ii. China

As I understand it, this is what happened to China over the past 30 years. In 1990, per capita GDP in China was around $300, less than a dollar a day. Hundreds of millions were in poverty. Chinese-manufactured goods were crummier compared to the goods manufactured by the experienced and well-capitalized firms in America and Europe. China was a serious underdog.

But as an underdog, China used the one advantage it had. They were willing to work harder for less. They offered a lower price.

As a result, labor-intensive manufacturing was gradually outsourced to China. And as more and more was outsourced to China, China began to develop advantages other than price. Workers learned to build products and factories better. Supply chains became shorter and denser. Quality went up. Designs got better.

None of this would have been possible without bootstrapping. If instead, China had decreed in 1990 that they were going to make their manufacturing better than the US/Europe/Japan and then put it out on the market at the same price, this would have been disaster. They would not have had the resources. But because they had business from their low prices, they got money and experience that allowed this to eventually happen.

This story is one of the brightest moments of the past 100 years, and of overall human history. Hundreds of millions pulled themselves out of poverty.

And it only happened because China could catch up by offering lowering prices.

iii. When can you not compete on price?

It is not always possible or effective to compete on price directly.

One case is when your superior competitor is willing to work hard for very little. This is the Amazon playbook. As Jeff Bezos says, “Your margin is my opportunity.” By tolerating small profit margins in online shopping, Amazon makes it very difficult for another online shopping company to get started by temporarily accepting even lower margins. (At that point, they literally might have to subsidize purchases.)

A second case when it is difficult to compete on price is when the market has economies of scale.

If the lemonade stand down the street is a giant industrial factory, then perhaps their lemonade is going to be better and cheaper than your hand-mixed lemonade. In these situations, there is no real way to enter the market by undercutting on price.

However, all hope is not lost. In these situations, there can still be a path to success. If you can differentiate your product, by, say, selling spicy lemonade, then you can still bootstrap your way up. You sell your crummy spicy lemonade, get a few sales from people who like spicy lemonade, and use that money and experience to eventually scale up to the point where you can compete with the industrial lemonade factory.

This is the classic story of technological disruption, where you gain a foothold in a small adjacent market before technological shifts cause it to take over the big market.

This is how low-cost steel minimills eventually displaced integrated mills. They started making crummy rebar, but eventually used their learnings and profit to figure out how to expand into higher quality steel, and now minimills produce almost all steel.

(Of course, disruption can also come from higher prices, like how Tesla got a foothold by selling the expensive and not-so-practical Roadster before selling the cheaper and more practical Model 3. This is what solid-state hard drive manufacturers did selling expensive SSDs, until eventually they learned how to make them cheap enough to take the magnetic hard drive market head-on. Ditto for digital cameras and pretty much every new expensive technology.)

Economies of scale make it difficult to enter a market by competing on price.

Last, there is a third situation where it is difficult to compete on price. And this might be the most difficult of all. It’s when your competitor offers their product for free. Consider Facebook. Facebook doesn’t charge users (except perhaps by ad load). So if you want to build a social network to win those users, there is only one way. You have to win on quality. There is no path to winning on price, and then letting the quality come later as you figure things out and grow. On day 1, your quality needs to be as high as Facebook. That is a tall, tall order. Imagine demanding a Chinese factory in 1990 to build a cost-effective iPhone. The skills and supply chain and scale were just not ready yet.

Of course, if a product is free, is is strictly true that you cannot compete on price? What if you enter the market by charging a negative price?

iv. Google’s lemonade

In 2009, Microsoft launched Bing. They knew that Google was dominating search, but they also knew that no monopoly lasts forever. Although Google’s quality was higher than Bing’s, Microsoft they might be able to enter the market by competing on price. Within a year, they launched Bing Rewards, which literally paid people to search on Bing. The price of Bing searches was now negative. And did it work?

No.

‘Free’ is a very sticky price. There is a big jump from $0.01 to $0.00. But there seems not to be a big jump from $0.00 to -$0.01.

Today, Google still dominates web search. It seems like difficult to imagine anyone ever displacing them from web search in the next 10-20 years.

The only way they lose, I think, is if web search becomes less important.

v. My main point

This brings me to my main point. Digital technology has stronger economies of scale than anything ever. As a result of the near-zero marginal costs, for most products a price of $0 is optimal. But when the price of a product is $0, it becomes difficult to displace. Would-be innovators cannot bootstrap their way to success by feeding off of price-sensitive consumers. No new lemonade stands enter the market. Innovation slows. The poor lose some ability to catch up to the rich.

Is there a solution? I see none.

Fundamentally, it comes from the cost structure of semiconductors and software. And they are what they are.

Good news about malaria

Wed, 20 Jun 2018 00:00:00 +0000

Good news about malaria

Good news:

Over the past 100 years, the mortality rate in the US has fallen by like 50%. Life expectancy at birth has gone up 30 years. The chief cause has been our ability to prevent and treat infectious diseases. In particular: tuberculosis, flu, pnemonia, diptheria, whooping cough, measles, smallpox, typhoid, and polio. From 1900 to today, infectious diseases went from killing 37% of Americans to 2%.

Bad news:

Outside of the US, infectious diseases are still big killers. One of the biggest is malaria. Malaria kills about 0.5M - 1M people a year. Two thirds are children. This works out to about 1,000 - 2,000 children a day being permanently separated from their families and painfully killed.

Good news:

Although we don’t yet have a cure for malaria, we do have an amazingly cheap way to prevent it. Sleeping under an insecticide-treated bednet significantly reduces the odds of catching malaria from mosquitoes. In sub-Saharan Africa, a life will be saved for every $3,000 or so spent on bednets and distribution.

Great news:

Not only does bednet distribution work in theory, it works in practice. There is a charity out there that does this work: The Against Malaria Foundation. They have a long track record, are thoroughly vetted, operate transparently, spend little on overhead, and regularly test the effectiveness of their distribution programs.

Greater news:

The Against Malaria Foundation has been doing an awesome job of making deals with governments and recruiting/retaining talent. Right now, they are not held back by lack of people. They are not held back by lack of opportunities. They are not held back by lack of deals. The number one thing holding them back from saving more lives is money. Over the next 3 years, they have a projected funding gap of $620M - $730M.

There may never again be a time in history with an ROI as good as this. If the richest billion people on Earth each donated a quarter a year, we could seriously dent malaria and save over 100,000 children. Ditto if every American gave a dollar a year. Ditto if 1% of Americans gave $100 a year. Or if 0.01% gave $10,000.

Whatever feels reasonable to you, I encourage you to give.

Why does Google think the slowest animal is the sloth?

Mon, 11 Jun 2018 00:00:00 +0000

Why does Google think the slowest animal is the sloth?

Google writes:

Our mission is to organize the world’s information and make it universally accessible and useful.

This is pretty hard, but Google has been at it a while. Let’s see how well Google is doing by asking it for some basic factual information, like the slowest animal.

Google says it’s the three-toed sloth:

To see just how wrong this is, take a look at this video of a sloth climbing a tree:

Can you imagine a 1 millimeter snail moving this quickly? A sponge? A coral? A tardigrade? A 100-micron long mite?

It takes almost no research at all to see that the sloth wouldn’t even be invited to the slowest animal Olympics, let alone win the gold medal.

Google has plenty of company though.

Siri guesses that it’s the two-toed sloth:

Bing returns images of sloths, but text saying it’s the seahorse:

Cortana returns text saying three-toed sloth while its first image is a Gila monster:

Although DuckDuckGo doesn’t elevate an editorial snippet, it returns 8 images of sloths and 1 image of a loris:

YouTube’s top hit for animals used to say seahorse, but today says the three-toed sloth:

WolframAlpha gets it the closest yet, returning the brown garden snail and explicitly states that it reinterpreted my query as the slowest land animal:

Wikipedia returns an article with slowest animal by phylum and class, including coral with a speed of 0 m/s:

I find it quite interesting that of the 8 websites I asked, the only website that got the answer partially correct was the also the only website that’s a non-profit.

2020 update: Despite being higher quality than all results returned by search engines, the Wikipedia article was deleted for being too low quality. Alas.

On technology

Mon, 11 Jun 2018 00:00:00 +0000

On technology

No one knows why the universe is the way the way it is.

But it is the way it is.

And because the universe is the way it is, things work the way they work.

Our simplest understanding of the universe is as follows: There’s some space, some stuff in the space, and some rules to follow. Again, no idea why, but that’s how it is.

Because of these facts, many other facts follow.

The universe has time and space. Because the universe has time, there is a thing we call energy. Because the universe has space, there is a thing we call momentum.

The universe is filled with 17 known kinds of stuff. Some don’t do much. Some clump into things we call atoms. There are around a hundred kinds of atoms.

Because of these facts, things work the way they work.

Energy: We cannot make energy go up or down. No matter how smart we are, this is too hard. So we have built entire industries to get energy and move energy.

Momentum: We cannot make momentum go up or down, just like energy. But unlike energy, we can get momentum by pushing off the Earth (this works for momentum because space has two directions: forward and backward. This does not work for energy because time has one direction: forward. If time could go backward, then maybe we could push off the Earth to go back in time). Because we are on Earth, we have not built entire industries to get and move momentum (except for space, where there is no Earth).

Atoms: We cannot make the kinds of atoms go up or down (except by costly nuclear stuff). So we have built entire industries to get atoms, rearrange atoms, and move atoms.

Technologies manipulate things. There are different things to manipulate. And there are different ways things can be manipulated.

The different things manipulated by technology can be lumped. (1) energy, (2) atoms, and (3) information.

The different ways things can be manipulated can be classified by what happens to positions. If an object’s parts are repositioned, we can call that (A) processing. If an object itself is repositioned, we can call that (B) transport. And if an object is not repositioned, we can call that (C) storage.

From these three things (energy, atoms, information) and three manipulations (processing, transport, storage) we can make a 3x3 classification of technology.

	Processing	Transport	Storage
Matter	Manufacturing Blast furnaces Refineries	Ships Automobiles Planes Pipes	Warehouses Silos Tanks Containers
Energy	Motors Generators Lighting HVAC	Wires Hydraulics Belts Axles	Fuels Dams Batteries Food
Information	CPUs Brains Sensors Whiteboards	Wires Fiberoptics Radio waves Sound waves	Computer memory Brain memory Books Optical discs

Things work the way they work, because the universe is the way it is.

On spaceflight

Wed, 23 May 2018 00:00:00 +0000

On spaceflight

Spaceflight is important because it may decouple the fate of life from the fate of Earth.

This article answers three questions:

Why does it cost $4,000/kg to get to space?
How low can the cost fall?
If that cost fell, what more would we do?

Supply of spaceflight

Cost structure: Why does it cost $4,000/kg to launch stuff into orbit?

Let’s learn from today’s lowest cost rocket, the Falcon 9 by SpaceX.

$62M is the list price of a Falcon 9 launch.¹^,²

At max payload, a $62M launch is:

$4,000/kg to low-earth orbit (15,600 kg max when landing booster)
$11,000/kg to geosynchronous transfer orbit (5,500 kg max when landing booster)

The cost of a launch is dominated by the cost of the rocket:

~90% is the rocket
~9% is operations
~0.3% is fuel

The rocket dominates, so let’s look at that. Rocket costs are driven by labor and equipment. Raw materials are less than 1% of the final cost. One way to gauge the effort needed to make a product is by its cost per kilogram. A Falcon 9 is comparable to an airplane:

A house: ~$1/kg
Aerospace grade aluminum: ~$2/kg
Toyota Prius: $16/kg
Falcon 9 (unfueled): $2,200/kg
AirBus 320neo (unfueled): $2,500/kg
iPhone 11: $3,600/kg

How low can costs go?

Payload / $ can be factored into: (payload / launch) * (launches / rocket) * (rocket / $). These 3 factors suggest 3 routes to reducing cost:

Make the rocket more efficient (more payload per launch)
Make the rocket more reusable (more launches per rocket)
Make the rocket easier to build (less money per rocket)

Although all 3 frontiers are being pushed, they oppose one another. Efficient rockets are more expensive to build. Reusable rockets carry less payload per launch.

As of 2018, SpaceX’s lower costs have come primarily from cheaper manufacturing, not from greater efficiency or reusability (yet).

Of course, even approaching the marginal cost of spaceflight requires a decent volume of launches over which fixed costs can be spread. Otherwise the “tyranny of space transportation costs (page 9)” can spiral out of control.

What drives the marginal cost of spaceflight?

Physical limits: How low can the cost go?

Conservation of energy and conservation of momentum limit what is possible.

Conservation of energy

At minimum,³ sending 1 kg to orbit requires ~33 MJ:

4 MJ of potential energy (from lifting against gravity)
29 MJ of kinetic energy (to fly fast enough to not fall back down)

33 MJ isn’t much - it’s a quarter gallon of gas or a dollar of electricity.

Energy also goes into two losses:

Air drag, which is minimized by going slow
Gravity drag, which is minimized by going fast

Because these losses cannot be minimized together, the only way to approach the 33 MJ limit is eliminate one so the other can be minimized:

Option 1: Eliminate air drag so you can go fast
- Build a giant vacuum tube many kilometers tall (e.g., StarTram)
Option 2: Eliminate gravity drag so you can go slow
- Climb a solid structure (e.g., space elevator)
- Build a rocket less dense than air (e.g., rockoon)

None are even close to feasible with today’s technology and interest rates. So from here on, I will focus on rockets.

A rocket’s energy is stored as chemical potential energy between fuel and oxygen. The engine converts chemical potential energy into kinetic energy. An efficient rocket maximizes the fraction of energy that ends up in the payload’s motion, and minimizes the energy that goes to:

Fuel & oxidizer molecules (i.e., chemical energy left from incomplete combustion)
Air (drag)
Useless motion of propellant aka heat (engine efficiency)
Useful motion of the propellant (propulsive efficiency, gravity drag)
The vehicle (reduced by staging)

A Falcon 9 carries 15,600 kg of payload using ~500,000 kg of RP-1 fuel and liquid oxygen. LOx and RP-1 release ~6 MJ/kg. That works out to ~200 MJ of fuel per kilogram of payload, only about ~6x the theoretical minimum.

Conservation of momentum

Because momentum is conserved, the only way to accelerate is to push off of something else. Rockets get around by pushing off of their exhausted fuel.

Delta-v is the budget that says how much pushing your rocket can do. Like a fuel gauge, it tells you which destinations you can and cannot reach.

The rocket equation has three variables:

Delta-v = velocity of propellant exhaust * ln( mass with propellent / mass without propellent)

Delta-v is convenient to work with for two reasons:

Delta-v is independent of rocket mass or design - i.e., if two rockets have the same delta-v, then they can reach the same destinations
Delta-v adds linearly - i.e., cost of maneuver A + cost of maneuver B = cost of maneuver A + B

This fantastic map shows how much delta-v is needed to reach destinations in the solar system:

Source

For most destinations (Moon, Mars, asteroids, other moons - basically anything except large planets), the majority of delta-v is spent escaping Earth’s gravity well (~9 km/s)
Counterintuitively, despite the fact that we’re all constantly falling toward the Sun, it is by far the most expensive destination to reach in the solar system (500+ km/s)

Historical progress

The number of orbital launches per year has been pretty steady at around 50-150, declining from the 1960s through 2000s, and then rising again in the 2010s

(Data I wish I had: government vs commercial split, the low-earth orbit vs others split, and the change in payload per launch)

Propulsion choices

The two key things to choose are your propellant (the stuff you throw backward) and your fuel (which gives you energy to throw propellant backward)
- Note: For chemical rockets, the propellant is the same as the exhausted fuel, but this isn’t the case for other propulsion systems
Energy can be kinetic or embodied in one of the fundamental forces
Chemical fuel
- 20-minute intro video by Scott Manley
- Ideally, to get the fastest exhaust velocity, you want fuel to be as hot as possible and as light as possible
- Some typical fuel combos:
- RP-1 (refined kerosene) - oxygen
  - Good specific impulse (353 s in vacuum)
  - RP-1 is liquid at room temperature, so no cooling & longer storage times
  - Density (0.8 - 1 g/L) is similar to oxygen, which makes pumping easier
  - Burns dirty (carbon residue), so not great for reuse
  - Made from petroleum, so can’t be made off-Earth
- Hydrogen - oxygen
  - Highest specific impulse (450 s in vacuum))
  - Burns cleanest (no carbon residue), best for reusability
  - Can be created from H2O on Moon, Mars, asteroids
  - Low density, requiring bigger, heavier tanks
  - Liquid needs heavy cooling/insulation to prevent evaporation (~20 K), loses maybe a couple percent a day in space
  - Can embrittle or crack some metals over time - not sure if this a huge problem or a tiny inconvenience
- Methane - oxygen
  - Kind of a compromise between RP-1 and H2
  - Decently high impulse (~380 s in vacuum)
  - Burns relatively clean
  - Similar boiling temperature as liquid oxygen, so cooling can be shared
  - Sort-of similar density to liquid oxygen, for easier pumping
  - Can be manufactured on Mars by Sabatier process
- A gazillion other combos, such as:
  - UDMN (unsymmetric dimethylhydrazine) - dinitrogen tetroxide: very storable, easy to ignite, used for long missions, but toxic
  - Monopropellants (e.g., H2O2, H2N2H2, N2O): simple, low impulse, used for orientation control)
  - Lithium - fluorine - hydrogen: crazy high impulse of 542 s, but even crazier chemistry - lithium heated hundreds of degrees to melt, hydrogen cooled hundreds of degrees to liquify, and toxic HF acid
Nuclear fuel
- Theoretically, nuclear fuel is superior to chemical fuel because it has higher energy density and burns hotter - however, containing nuclear fuel is difficult and heavy
- Nuclear to thermal to electric to kinetic
  - Nuclear electric rocket
  - RTGs (Radioisotope Thermal Generators)
- Nuclear to thermal kinetic
  - Controlled reaction: Nuclear thermal rocket (~850 s of specific impulse in 1960s)
  - Uncontrolled reaction: Nuclear pulse propulsion
Photonic fuel
- Beam-powered propulsion
- Solar sail
Kinetic/thermal fuel: Pretty garbage, no real way it could work, only included for completeness
Gravitational fuel: Pretty garbage, no real way it could work, only included for completeness
Forms of propulsion not linked to a fuel
Ion thruster
- Can be powered by off-board energy (solar) or very dense on-board energy (nuclear)
- Specific impulses as high as 20,000 (but needing to be powered by heavy solar panels or nuclear reactors)
- Super low thrust
Space elevator
- Fragile, high-cost, potential wobble instabilities, not strong enough for Earth gravity
- Potential long-term use on moons
Near future
Upcoming rockets
- SpaceX
  - Falcon 9 Block 5 (May 2018)
    - Uses RP-1/LOX fuel with coaxial pintle injection, for both stage 1 and stage 2
    - So far, SpaceX has reflown used Falcon 9s three times, but each just once
    - The Block 5 version of the Falcon 9 is intended to be the final version, better engineered for reusability (as many as 10 or more flights)
    - Goal is to reduce refurb turnaround time from months to weeks/days, with a 24-hour turnaround time demo targeted in 2019
    - Plans to recover the second stage have been shelved, but SpaceX is experimenting with recovering the fairing
  - Falcon Heavy (2018)
    - Built from Falcon 9s
    - Bigger
  - Big Falcon Rocket (2020?)
    - Envisioned to replace all previous SpaceX rockets/spacecraft (Falcon 9, Falcon Heavy, Dragon spacecraft)
    - Methane/LOX for stage 1 and stage 2
    - Both stages to be reusable
    - 150,000 kg payload to LEO (6x Falcon 9)
    - Designed for satellites, Mars missions, Earth-to-Earth passenger transport
- Blue Origin
  - New Glenn (2020?)
    - 45,000 kg payload to LEO
    - Methane/LOX for stage 1
    - H2/LOX for stage 2 (and optional stage 3)
    - Stage 1 to be reusable
    - $2.5B spent on development as of Sep 2017
- NASA/Boeing
  - Space Launch System (SLS) (2022?)
    - Heavy launch vehicle intended to be Mars-capable
    - Derived from Space Shuttle design
    - Solid fuel boosters
    - H2/LOX for stage 1, stage 2, and optional stage 3
    - ~100,000 kg payload to LEO
    - Something like $8B has been spent with another $7B planned by 2021 (a second source says $22B thru 2020 - may depend on how you attribute costs)
- European Space Agency (ESA)
  - Ariane 6 (2020?)
    - 20,000 kg payload to LEO
    - Solid fuel boosters
    - H2/LOX for stage 1, stage 2
    - $4B total development cost projected in 2015
    - Not intended to be reusable, but research being done for future versions
- Rocket Lab
  - Electron rocket
    - ~200 kg payload (<$6M per launch, so $27K / kg to SSO)
    - Key idea is to use a battery for fuel pumping, which is less weight efficient but simpler to build
    - 2 launches so far: one success, one failure
    - RP-1/LOX for stage 1, stage 2
- Not sure what’s coming out of China/Russia
- Let me know if you think I should add any other rockets here

Demand for space travel beyond Earth orbit

Demand for space travel
- I am skeptical that there are economic cases for space travel in the next century. The most promising economic activities seem to be fuel mining and manufacturing, but those are only useful for supporting other activities. Very long-term, the best case I see for an economic return is the terraforming of Mars, and even that seems worthwhile only if we subsidize it to hedge against something going seriously bad on Earth. (And it’s gotta be seriously bad if Mars ends up looking better than a remote bunker on Earth.)
- Demand can fall into a number of categories:
- Science
  - Beyond curiosity, economic applications seem limited (?)
- Tourism
  - In his book Artemis, author Andy Weir saw tourism as the most plausible driving economic force for a moonbase
- Mining
  - Mining in space will be expensive, so the only reasons to do it are (1) to find resources that are rare on Earth or (2) to support other space activities
    - Heavy metals - e.g., platinum. (Rarer on Earth vs other bodies in the solar system because they sank into the center of the Earth.)
    - He3
      - 100x more abundant in solar system than on Earth (on Earth they have been diluted by He4 coming off of U and Th radioactive decay - though I don’t know why Earth has more U and Th)
      - Articles against He3 mining:
        
        http://cds.cern.ch/record/1055767?ln=en
        
        We don’t even really know if it’s out there and it’s not that great for neutron-free fusion because neutrons still get created from secondary reactions
        
        http://www.thespacereview.com/article/2834/1
        
        We don’t even have He3 fusion and the amounts on the moon are unproven estimates and even if true the concentrations of He3 on the moon are still so low that it require massive mining operations just to get tiny amounts of He3
    - Hydrocarbons - In-space fueling
  - Planetary Resources, who has thought about mining more than I, seems to have pivoted from metal mining to water mining
  - I wish I had a better sense of the market for precious metals. For example, each year Earth mines about 200 tons of platinum worth $5B. If we landed a platinum-rich asteroid with 10,000 tons of platinum, how would the market respond? Obviously prices would crash with increased supply. We wouldn’t even know what to do with that much platinum. But long-term, what new demand might be unlocked by lower platinum prices? Difficult to speculate on.
  - Space mining has an interesting parallel with deep-sea drilling. Deep-sea drilling was impossible for a long time, due to the technology and cost. Eventually it became worthwhile as we figured out the technology (which essentially requires robot submersibles to build an underwater city) and with the financing (a single platform can cost $3 billion). Asteroid mining will require further leaps in technology and financing.
- Low gravity manufacturing
  - ??
- Solar
  - Seems way too costly to me. Getting sunlight 24/7 with 0 atmospheric losses is not gonna be that much better than getting sunlight 8/7 with 50% losses. Putting down 6 panels on the ground is easier than 1 panel in space.
  - One interesting thing is the possibility of long distance wireless power. On Earth this doesn’t work because of atmosphere. But in space, a network of satellites could conceivably shoot laser power around to one another, right?
- ??
Destinations for space travel (TBD)
- Low earth orbit
  - Cheapest, closest to Earth, no resources
  - ???
- Moon
  - Water resources for hydrogen fuel
  - Ability to tunnel or build structures for radiation protection
  - Some gravity
  - Super cold polar craters for cryogenic fuel storage, cryogenic computing, etc.
  - Close enough to Earth that we can launch help if something goes wrong
  - Close enough to Earth that long loiter times are not needed
  - ???
- Asteroids
  - Water resources for hydrogen fuel
  - Heavy metal resources to bring back to Earth
  - Near zero gravity
- Mars
  - https://en.wikipedia.org/wiki/Human_mission_to_Mars
  - Water for hydrogen fuel and hydrocarbons for methane fuel
  - Some gravity, a little atmosphere
  - Can tunnel / build for radiation protection
  - Atmosphere
  - Can be terraformed
  - Far from Earth (energy-efficient travel times of ~300 days during windows that occur every 26 months)
- Interesting moons
  - Europa, Enceladus, ???
Interstellar trade (TBD)

Notes

Source: SpaceX website in April 2020 ↩
Note: Although $62M is the list price, customers may pay more or less. On the high side: the US Air Force has repeatedly paid ~$90M for launches of its GPS satellites (the other bidder, ULA, charges twice that). Heavier payloads requiring an expendable booster are certainly priced millions higher than $62M, but I could not find details. On the low side: Flying on a reused booster discounts the price by ~20% to ~$50M. Modest discounts are also offered for contracting multiple launches. Outsiders have estimated the marginal cost to be $37M–$48M. ↩
“Minimum” subject to the constraints of not changing the Earth’s shape, mass, or atmosphere. ↩

On self-driving cars

Mon, 21 May 2018 00:00:00 +0000

Highlights
- The potential market for self-driving cars is enormous, like a few trillion dollars or a few percent of world GDP
  - Roughly, I estimate potential market of $2T (~1B cars * 1 hr/day + 50M commercial drivers * 5 hrs/day all at $5/hr)
  - $2T may sound high, but the average American driver spends 7 hrs/wk driving, which is 20% of the ~35 hrs/wk worked by the average American worker.
- Self-driving cars are nowhere near as safe as humans, nor will they be soon
  - From 2014 to 2017, Waymo’s reported disengagement rate fell from 1 per 800 miles to 1 per 5,500 miles
  - This is still far away from the human crash rate of 1 per 330,000 miles (with many caveats to the comparison, of course)
- I expect the transition to self-driving cars will be gradual, taking decades
  - The first release of almost every technology is expensive and crummy
  - Capital-intensive technologies can take decades to penetrate markets (decades to bring costs down, decades to scale manufacturing up, decades for old capital stock to age out, decades to shift consumer norms)
Conceptions of self-driving
- Essentially, driving is a function that takes vision as input and then quickly outputs steering & speed
- I see three important dimensions for imagining how future cars will be driven:
  - Who drives (autonomy)
    - Human mostly drives (SAE levels 0-3)
    - Computer mostly drives (SAE levels 3-5)
  - Where & when do they drive (environment)
    - Highway-only
    - Heavily mapped zones in good daytime conditions
    - Everywhere
  - What are they driving (cargo type / vehicle type)
    - Passenger, owned (car)
    - Passenger, hailed (taxi)
    - Passenger, shared (shuttle / bus)
    - Cargo, short-range (delivery trucks)
    - Cargo, long-range (semi-trailer trucks)
- SAE defines 6 levels of autonomy.
  - Level 0: No automation
    - E.g., anti-lock brakes, cruise control
  - Level 1: Drive assistance (in some circumstances, computer may adjust speed or steering)
    - E.g., Automatic emergency brakes, adaptive cruise control, lane-keeping assistance
  - Level 2: Partial automation (in some circumstances, computer drives while human monitors)
    - E.g., Tesla Autopilot, Cadillac Super Cruise
  - Level 3: Conditional automation (in some circumstances, computer drives while human is ready to take over)
    - E.g., Audi Traffic Jam Pilot (only <60 kph on divided highways)
    - Many carmakers plan to skip Level 3, as humans cannot be relied upon to take over on short notice
  - Level 4: High automation (in some circumstances, computer drives without need for human)
  - Level 5: Full automation (in all circumstances, computer drives without need for human)
- Focusing on SAE levels may make it seem as if there is one path for the technology to progress along 1->2->3->4->5 or that the levels are equally spaced in difficulty. But many cars may have different capabilities in different environments (e.g., level 4 on a highway, level 2 around town, and level 1 in a construction zone). And it could easily turn out that getting to level 4 takes 1,000x the R&D spend than it took to get to level 2.
History of self-driving
- https://en.wikipedia.org/wiki/History_of_autonomous_cars
- The idea of autonomous cars is not new
  - 1950s-1960s: Early research into autonomous highway driving using buried cables. A number of groups build prototype cars that drive down highways with buried cables. Never commercialized.
  - 1980s-1990s: First computer-controlled self-driving cars using cameras and sometimes lidar. EUREKA program invests ~$1B in self-driving car research over 8 years. CMU first uses neural networks for self-driving control. German group drives prototype 620 miles on highways. American group drives 3,100 miles (steering only). Italian group drives 1,200 miles. Cars still require constant supervision: disengagements occurred roughly every ~1-10 miles.
  - 2000-2010s: In 2007, six cars complete DARPA Grand Challenge III, an urban course populated with human drivers. Autonomous trucks are commercialized for mining. Neural networks begin dominating ImageNet and other computer vision contests. Google, carmakers, and startups invest billions into autonomous driving.
Benefits of self-driving
- One of the biggest blind spots in forecasting the impact of future technology is to ignore the elastic response of the market to lower prices and better quality
  - For example, in 2014 an NYU finance professor valued Uber at only $5.9B because his model assumed the world taxi market was fixed in size. In fact, over the four years after his article, the number of rides in NYC doubled due to ridesharing. Uber’s latest fundraise in 2018 valued it at $62B (post-money).
  - Despite this cautionary tale, I will commit the same sin in my own analysis. For the most part, I will assume a fixed market for driving, since I expect that self-driving cars will not reduce the cost of driving by more than 30% in the first decades of operation, and I expect that such a reduction in cost will not be radically transformative.
- Labor/time savings are biggest benefit of self-driving
  - I estimate global market of ~$2T (~1B cars * 1 hr/day + 50M commercial drivers * 5 hrs/day all at $5/hr)
  - $2T may sound high, but the average American driver spends 7 hrs/wk driving, which is 20% of the ~35 hrs/wk worked by the average American worker.
  - Actual market may be lower if competition drives prices down. But may also be bigger in the future as (1) the world continues developing and (2) lower prices increase quantity demanded.
  - Market size can be triangulated potentially by consumer research, in addition to the theoretical calculation of time benefits. That said, it’s hard to trust consumers’ predictions of their buying behavior when a product doesn’t exist yet. BCG/WEF survey says ~60% of buyers would not pay a cent more for self-driving, though ~20% would pay $5K+. Suggests that cultural inertia may further slow adoption of self-driving cars, but I don’t take it too seriously.
- Second benefit of self-driving cars could be reduced crashes
  - IF they become perfectly safe (which won’t happen soon), the avoided crashes could save ~$2T/yr (savings are roughly ⅓ avoided suffering due to death, ⅓ avoided suffering due to injury, ⅓ avoided economic costs such as property/lost productivity/congestion/medical bills/etc.)
    - Federal study estimates crashes cost US ~$1T each year: $240B economic cost, $600B human cost
    - Another source estimates global cost of crashes at $500B, which might square with the first if it’s economic cost alone
    - Most of these costs are negative externalities on victims and society; therefore, the safety savings to buyers themselves will only be a fraction of these numbers
    - Note much of this reduction might already be possible with non-self-driving technology (e.g., automatic emergency braking, breathalyzers, modern safety features); therefore the marginal value of self-driving technology might be lower than the total above
  - Stats on crashes:
    - Humans kill 1 person per ~100 million miles driven
      - Note that this is just the average - for drivers who drive sober in daytime urban environments and wear seatbelts, death rates are likely a few multiples better
      - The death rate has dropped roughly 3% per year for the last 100 years
      - Key causes: ~23% of costs attributed to alcohol, ~15% of costs attributed to distraction
      - Key locations: ~10% on highways, ~44% in intersections
      - Key styles: ~36% roadway departure, ~41% single vehicle, ~10% involve ped/cyclist
      - ~½ of deaths are at night; ½ during the day
      - ~¼ of involved drivers had positive blood alcohol (so presumably more than ¼ of collisions, given that most collisions have two drivers)
      - ~⅔ of fatalities are people in cars; ~⅓ are outside (motorcyclists, pedestrians, cyclists)
      - ~½ of vehicle occupant fatalities were not wearing seat belts!
      - Wearing your seat belt reduces odds of both death and injury by ~50%
    - Statistically demonstrating that autonomous vehicles are less fatal than human drivers will take billions of miles driven (many multiples of 100M miles per fatality at the very least).
      - Waymo is leading in miles driven, and has accelerated its pace to about 1 million miles per month (adding to its cumulative total of 8 million miles from 2009 to 2018)
        
        As always, it’s worth caveating that all miles are not created equal. Older miles from vehicles with older sensor suites probably have near zero utility today. And miles that come from the same safe area driven over and over will have less utility than miles from a variety of difficult environments.
      - At Waymo’s rate of 1 million miles per month, it will take about 80 years to reach 1 billion miles, at which point we can finally see whether the self-driving cars killed fewer or more than the human rate of ~10
        
        Given Waymo’s mid-2018 fleet size of ~600 minivans, that suggests each minivan is only driving 50 miles a day
        
        If you want to drive 1 billion miles in 5 years, and drive each vehicle 250 miles a day, that implies a minimum fleet size of ~2,000 vehicles, which seems quite doable
        
        Waymo has announced deals to buy up to 62,000 more Chrysler minivans along with up to 20,000 Jaguar SUVs (hard to know from the outside what ‘up to’ means, but both contracts are getting deliveries)
Costs of self-driving
- R&D costs
  - Google v Uber trial revealed that Google spent ~$1B between 2009 and 2015 (seems low, given Lewandowski bonus of $120M and other reported bonuses. Figure possibly doesn’t include salaries/bonuses.)
  - GM plans to spend ~$0.6B/yr on AV R&D (https://www.freep.com/story/money/cars/general-motors/2017/04/13/gm-to-add-1100-jobs-cruise-automation/100437028/)
  - Waymo probably in similar ballpark (~500 employees @ estimated $1M per employee = ~$0.5B/yr) https://www.linkedin.com/search/results/people/?facetCurrentCompany=%5B%2217900793%22%5D
  - Add in Velodyne (~300 headcount) and MobileEye (~700) and Tesla (???) and major automakers (???) and major parts suppliers (???) and other startups (???) and lidar makers and chipmakers and we’re maybe in the ballpark of $5B a year being spent on AV R&D
  - If development takes ~20 years (2010-2030), that’s ~$100B in total R&D spend that needs to be recouped from sales
  - If in 2030 Earth produces 100M cars per year, of which 10% are self-driving, and the R&D price premium is $1,000, then it will take 10 years to recoup $100B in R&D - seems very doable
  - As a point of reference, to develop a new car model it costs automakers ~$1B in R&D spread over a few years
- Variable costs
  - Original Google self-driving car had ~$150K of extra equipment (~$75K lidar and ~$75K other)
  - Two largest costs are lidar and electronics
  - Lidar
    - Lidar is expensive - Velodyne’s high-end model costs ~$80,000 (though less for lower end models)
    - Because of these high prices, the top self-driving players have made the strategic choice to design and build their own lidar
    - Waymo says their lidar cost is 10% of their earlier ~$75K (so $8K?)
    - Cruise says their lidar costs were $20K in 2017, but they aim to bring that down to $10K and then hopefully $300 (not clear to me whether these numbers are individual sensors or total cost to outfit the car - presumably the individual sensor price, since that looks better to report)
    - There are hopes for solid-state lidar that cost as low as ~$100, though so far they all seem to be vaporware
    - Personally I have no sense of what’s going to happen here - huge investment and economies of scale should drive rapid price declines, but so far my sense is that solid-state and cheaper lidars are still pretty crummy
    - A minority of companies have tried forgoing lidar altogether
      - E.g., Tesla, Starsky Robotics (trucking startup)
    - When I chatted with the CEO of Starsky Robotics in 2018, he emphasized to me that lidar doesn’t work well at highway speeds and is only useful for non-highway driving
    - Last point on lidar is that I used to believe the high costs of lidar would be prohibitive. However, from my financial models, a robo-taxi can earn a few hundred thousand dollars over a lifetime of 5 years. Next to that profit pool, a $50K lidar unit is totally affordable. (But for an owned car, a $50K price premium would turn away almost all buyers.)
  - Electronics
    - Self-driving electronics cost thousands of dollars
    - In addition, they require something like ~1 kW of power, which along with cooling can lower EV range by like 5-10%
    - Most platforms are using a mix of CPUs, GPUs, and SoCs
    - Here’s a great article on the tradeoffs between chip types in optimizing for tail latency (critical for safety) and power consumption: https://blog.acolyer.org/2018/04/20/the-architectural-implications-of-autonomous-driving-constraints-and-acceleration/
  - Radar
  - Cameras
  - GPS / IMU
  - Insurance
    - I expect that high insurance rates will be a substantial cost of robo-taxi operations. Not only is there uncertainty around the crash rate (for which insurers will demand a premium), but the costs of crashing go way up when you’re now carrying a few tens of thousands of dollars of safety-critical hardware aboard your car. A risk-averse robo-taxi operator might prefer to total cars that crash rather than risk putting damaged hardware back on the roads.
Effects on society
- A few changes are easy to predict
  - People will travel more by car
  - Less parking will be built on premium land
  - More kids and seniors and driving-disabled will be able to get around
- But many other changes are difficult to predict, even directionally
  - More cars will be built because driving gets easier
    - Or: Fewer cars will be built because car sharing gets easier
  - Cities will get denser, since the cost of congestion falls and the space needed for premium parking falls
    - Or: Cities will get less dense, since the cost of commuting from far away is lower
  - People will travel less by buses, subways, trains, bicycles, etc.
    - Or: People will travel more by buses, subways, trains, bicycles, etc. as robo-taxis are better able to plug the gaps and eliminate the need for car ownership
  - People will travel more during peak times, as waiting in traffic is no longer as bad
    - Or: People will travel more during offpeak times (unlike Uber/Lyft drivers, robo-taxi supply will be constant throughout the day, leading to extra low prices during offpeak times)
  - Car designs will get more boring, as more will be purchased by fleets (this is the take of VC firm a16z)
    - Or: Car designs will get more varied and specialized, as shared ownership means you can mix and match rented vehicles to your purpose rather than buying a jack-of-all trades (this is the take of VC firm a16z)
  - Home delivery will get cheaper
    - Or: Home delivery will actually get comparatively more expensive, since robotic unloading of deliveries is unsolved and now going to stores just got cheaper
- Predicting the magnitude of changes seems even harder:
  - Congestion: As the cost of congestion goes down, the amount of congestion will go up, so long-term, traffic may rise until the annoyance of being a passenger for Y minutes equals the original annoyance of being a driver for X minutes. If the rebound effect is large, deadweight loss caused by congestion may not actually fall by much and self-driving could even be a negative sum technology.
  - Taxi rates: Self-driving tech will reduce taxi rates ~33% according to my math:
    - A 30-minute Uber drive costs about $30. Assuming a 50% utilization rate, that means they are earning $30 an hour. Uber charges ~33% of that, ~$10, leaving $20 for the driver and the car (this agrees with studies showing ~$20/hr for driver and car https://medium.com/uber-under-the-hood/an-analysis-of-ceeprs-paper-on-the-economics-of-ride-hailing-1c8bfbf1081d). Gas costs ~$3/hr (assuming 1 gal/hr at 30 mpg at 30 mph), car depreciation costs ~$3/hr ($20K car lasting 200,000 miles at 30 mph), and maintenance/repairs/insurance costs ~$3/hr ($0.10/mi at 30 mph). This leaves $11/hr for the driver. So very roughly, ⅓ goes to Uber, ⅓ goes the car, and ⅓ goes to the driver.
    - Suppose adding self-driving tech to a car costs $13K and eliminates the driver cost. That will add $2/hr in costs and subtract $11 in driver wages. So total cost drops 30% from ~$30/hr to ~$21/hr.
    - Uber fees:
      - https://therideshareguy.com/uber-increases-booking-fee-and-effective-commission/
      - https://www.quora.com/What-percentage-cut-does-Uber-take-from-the-total-fare-cost-of-a-ride-Do-they-subtract-a-flat-fee-for-each-dispatch-or-a-percentage-Are-there-initiation-monthly-fees-to-be-a-driver
    - $0.10/mi:
      - https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0ahUKEwjRwpeS1Z_aAhVi0oMKHaxZDpkQFggpMAA&url=http%3A%2F%2Forfe.princeton.edu%2F~alaink%2FSmartDrivingCars%2FPDFs%2FZoepf_The%2520Economics%2520of%2520RideHialing_OriginalPdfFeb2018.pdf&usg=AOvVaw2qx3_w0fH6CiwRI3k-DiBl
  - Neglected factors that will refine the math above: (on balance, they will probably lower the price a bit more than calculated, and more so in the long-run)
    - The cut taken by ridesharing apps will fall. Ridesharing software costs are mostly fixed, so with higher volumes they can better afford to drop their cut. (But right now these companies are bleeding billions, and reversing that bleeding will work in the opposite direction)
    - Some of the driver savings will go toward fleet management, not price cuts. Right now, drivers take care of fueling their car, keeping it clean, and parking it. Presumably some of that $11/hr for drivers would go toward fleet management rather being turned into price cuts for riders.
    - Electrification will grow the market. If electric cars get cheap enough, their lower fuel costs and lower maintenance costs would lower the variable price of driving and thereby increase both the driving pie and ridesharing share of the pie, leading to better economies of scale
    - Densifying markets will increase utilization and lower costs. Greater fleet size could lead to less empty driving and more revenue per mile driven
    - Cars with high utilization will last more miles. Cars depreciate as a function of both age and distance. Higher utilization rates will allow cars to drive further before age-related damage kicks in. Lifetimes could go from 200,000 to 300,000 miles, or even further if cars are redesigned to last longer.
    - Discount rates make upfront R&D and equipment less affordable. Including interest rates tilts the balance toward drivers and gas versus self-driving and electric, both of which have higher up-front costs. (I expect real risk-free rates to stay below 2% due to secular stagnation; therefore, I expect discounting to be dominated by the risk premium.)
    - Redesign of cars will lower costs. As volumes of self-driving cars grow, it becomes worthwhile to redesign lower cost cars suited for self-driving. E.g., no steering wheel, no pedals, optimized for distances/environments, optimized for passenger load, self-driving hardware built-in rather than bolted on.
    - Insurance rates will change. I expect self-driving car insurance to cost more than human car insurance. This is because (1) self-driving cars may be more dangerous than humans when released, (2) insurers will require a premium to take on uncertain risks, and (3) self-driving cars will carry thousands of dollars of hardware that will need to be inspected and replaced after collisions. (Of course, this is all subject to notions of what’s a “fair” comparison.) Note too that as self-driving cars get safer and penetrate the market, this will not only lower the price of self-driving car insurance, but also the price of car insurance.
    - Transportation rules could be redesigned. Long term, high penetration of self-driving cars opens up possibilities for new, more efficient road rules, which could increase road speeds and bandwidths, and thereby lower cost of service.
  - One thing I don’t understand is that taxis in Dhaka, Bangladesh cost like $4/hr. Presumably cars and gas cost the same all over Earth but labor is super cheap in some countries. One way to get at the labor share vs capital share of fares is to look at countries with low wages. Dhaka taxis cost like $4/hr ($8/hr at 50% utilization), which is much lower than my estimate above. Maybe the cars are older and crummier, repairs and maintenance are way cheaper? Non-Uber taxis appear similarly priced, so it cannot explained by Uber choosing to take a negative cut to build the market.
    - http://uberestimate.com/prices/Dhaka/
    - Non-Uber fares: https://www.worldtaximeter.com/dhaka
Predictions
- Other predictions
  - Be cautious generalizing a mood from these past predictions. Predictions of self-driving will all be wrong, until one day they are right. There is a massive selection bias in the predictions below. They are not a representative sample of predictions. The main point isn’t to laugh at the predictions of people in the past; it is to be humble in our own predictions lest the people of the future laugh at us. The other main point is to never trust predictions of people trying to sell you something.
  - Overly optimistic predictions
    - 1960 according to the book Magic Motorways in 1940
    - 1975 will see full-use according to NY Times article in 1960
    - 1976 according to GM (or at least their marketing) in 1956
    - 1985 according to Nobel and Turing winner Herbert Simon in 1965 (“Machines will be capable within 20 years of doing any work a man can do.”)
      - A whole bunch of other early AI researchers in the 60s also predicted that AI would outpace humans within their lifetimes
    - 2017 (available to ordinary people) according to Sergey Brin of Google in 2012
    - 2017 (full autonomy) according to Elon Musk of Tesla in 2015
    - 2019 according to Elon Musk in 2017 and again in 2018
    - 2020 (widespread adoption) according to Stanford AI100 Standing Committee in 2016
      - Aside: An exchange with the authors left me very disappointed
    - 2019-2025: A long list of predictions by automakers and other organizations
  - Overly pessimistic predictions
    - Impossible to know until self-driving cars actually happen
  - Predictions right on the money
    - In 2014, much effort will be put into designing vehicles with robot brains according to Isaac Asimov’s forecasts from 1964 (it’s astounding how much he got right, 28/44 by Peter McArthur’s count)
- My predictions
  - Prediction is where analysis becomes most real and most difficult. It’s easy to sit on the sidelines, disprove obviously wrong predictions with back-of-the-envelope math, and bask in the feeling of “I know better.” But it takes much more humility to put your own skin in the game, and make predictions that will likely be wrong in one way or another.
  - None of the predictions below are meant to be definitive. Just my best guess at the moment I wrote them down in mid-2018. Certainly there are people in the world smarter and more informed than me who will be able to make narrower, more accurate predictions.
  - As informative as individual predictions may be, there is also meta-information encoded in the predictions I choose to make and the way I define problems. I have tried to mostly pick predictions that are easy to unambiguously falsify.
    - Self-driving is solvable (100%)
    - Level 4 city+highway self-driving cars will be generally available for sale to consumers by…
      - 2020: <1%
      - 2025: 10%
      - 2030: 45%
      - 2035: 60%
      - 2040: 70%
      - 2045: 75%
      - 2050: 80%
    - Level 4 city+highway self-driving cars that cost <$30,000 will be generally available for sale to consumers
      - 2020: <1%
      - 2025: 2%
      - 2030: 15%
      - 2035: 30%
      - 2040: 55%
      - 2045: 65%
      - 2050: 75%
    - Level 4 city+highway self-driving cars will debut in robo-taxi fleets years before being sold to individuals (90%)
    - The biggest barrier to adoption will be the technology (safety, reliability, cost), not regulation (though of course there may be regulations saying they are illegal because they are unsafe)
    - The transition will be very gradual
      - Self-driving car production going from 0.1% of new cars to 80% of new cars will take more than 20 years
      - By 2050, the majority of cars on the road will not be self-driving
      - In the taxi market, self-driving cars will takes many years to usurp Lyft/Uber/etc. At first they won’t have much cost advantage, they will face large barriers to entry (fleet density, consumer inertia), and building a fleet will take tons of capital relative to Lyft/Uber’s bring-your-own-car model (literally billions of dollars per large city). As a result, the rollout will be slow and done city-by-city, probably starting in the US.
    - One thing I’m unsure of but feels predictable is whether a Waymo or Cruise robo-taxi fleet will: (a) build their own robo-only app, (b) build their own app that also takes human drivers (c) get on Lyft/Uber via partnership or takeover, (d) build their own meta-app which displays Lyft/Uber rides alongside their own, or (e) other
      - Option A seems cleanest, but gives a worse user experience if you have to bounce around apps due to low fleet densities (and even long-term it seems potentially beneficial to keep human drivers for peak times while use robo-taxis for high-utilization baseload)
      - Also, by the time self-driving robo-taxi fleets are out, I predict that ride-hailing apps will have public transportation and bike/scooter transit integrated as well, which complicates the app question a little (for urban areas)
Evolutionary paths
- Most technologies start out crummy and expensive, serving just a narrow niche, before improving in quality and expanding to greater markets
- For self-driving vehicles, there are few paths to inexpensive full autonomy in all conditions
  - Owned cars that are highway-only and use human driver as backup
  - Robo-taxis that only operate in limited geo-fenced areas and have backup tele-operators
  - Self-driving trucks that are limited to specific easy, high-volume highways and use either human drivers or tele-operators as backup
Analysis of various vehicles types (TBD)
- Cars ()
- Taxis (driver is ~⅓ of cost)
- Buses (driver is 40% of cost - really???)
- Trucks (drivers needed for inventory management/loading/unloading)
- Class 8 semi trucks (driver is 18% of cost)

Is data scientist the hottest job in America?

Fri, 18 May 2018 00:00:00 +0000

Is data scientist the hottest job in America?

Some articles say data scientist is the hottest job in America, because companies want them so badly they are willing to pay over $100K per year.¹^,²^,³

However, I’m going to say that CEOs, hedge fund managers, and sports stars are actually the hottest jobs in America. Companies want them so badly they are willing to pay over $10M per year (equivalent to a century’s worth of data sciencing!).

Why do people choose to write articles proclaiming a shortage of data scientists but not a shortage of CEOs/hedge funders/sports stars?

Presumably it’s because you or I won’t bother to click a headline that says athletes get paid a lot, but we might click a headline that says that data scientists get paid a lot. And why is that?

Is it because data scientist feels achievable, so maybe we’d want to become one?

Is it because data scientist is in our social class and CEOs are not, so maybe we’d be more envious and status-conscious?

Is it because data scientist salaries are getting paid more than a cultural interpretation of what they “deserve” and we care about strangers getting what they deserve?

I wonder.

Footnotes

This is America’s Hottest Job. Michael Sasso, Bloomberg, 2018. ↩
Data Scientist: The Sexiest Job of the 21st Century. Davenport and Patil, Harvard Business Review, 2012. ↩
50 Best Jobs in America for 2019. Data Scientist has topped Glassdoor’s best jobs list for four years straight (2016 - 2019). ↩

Predictions about the year 2050

Sun, 23 Jul 2017 00:00:00 +0000

Predictions about the year 2050

As a counterreaction to claims that technology is changing faster than ever and that future progress is impossible to foresee, allow me to make some predictions about the year 2050:

We’ll still eat food. Food will still be grown on outdoor farms. We’ll still live in houses. We’ll still drink water that comes in pipes. We’ll still use electricity delivered by metal wires. We’ll still make electricity by burning fossil fuels (though more will come from solar & wind). We’ll still have cars, probably self-driving but possibly not. We’ll still have families. We’ll still learn reading, writing, math, science, etc. in physical schools. Heart disease will still be the number one killer. Cancer will not be broadly cured, though some narrow cures will be available. CRISPR (or other technologies) will not yet have led to whole-body gene editing in adult humans. We’ll still wear clothing. We’ll still use money, issued by governments. We’ll still be living on Earth only (though humans may have walked on moon/Mars). We still won’t have fusion energy, but we’ll be closer. Electronics and media will be far better. Much more of work, life, and entertainment will be computerized. We won’t have general AI. Nuclear war will have not occurred. Most countries as we know them today will still exist. Books will still exist as a medium. Universities will still exist to teach and research. Airplanes will still fly below the speed of sound. Cars will still drive on roads. Transporting goods by humongous diesel-powered ships will still be the cheapest option. Computers will still be based on classical, binary logic. Most major companies today will still exist. Most major companies existing then will have existed today. Jobs will still exist. People will be healthier. People will be richer. People will be nicer. People will be happier.

Overall, I think life will be structurally similar to today. The biggest differences will be in improvements in electronics and software and media, as well as cultural shifts caused by those trends. The shift to solar and wind will continue substantially. Biotechnology and AI remain question marks, with high upside.

Edit in 2018: Let me further qualify these fuzzy claims by saying that ‘widespread’ is implicit. I.e., we may have had a scientific experiment produce positive fusion energy, but fusion energy will not be widespread as a form of energy production.

The power of sometimes

Mon, 01 Aug 2016 00:00:00 +0000

The power of “sometimes”

If you owned a computer that only “sometimes” turned on, placed Amazon orders that only “sometimes” arrived, and had friends who only “sometimes” kept their promises, you’d live in a state of frequent frustration. Fortunately for most us, we don’t live in such a world. All around us, products and processes have been designed to be reliable.

However, when it comes to electronic and magnetic materials, reliability - at its literal extreme - is not necessarily a good thing. Transistors that are always conducting won’t make a good computer chip. Magnetic bits that always point up won’t make a good hard drive. And photochromic lenses that always stay dark won’t make good transition eyeglasses. Common to each of these technologies are materials that “sometimes” work one way and “sometimes” work another way. This is the power of “sometimes.”

No class of materials is more emblematic of the power of “sometimes” than the semiconductor, which sometimes conducts electricity and sometimes does not. The controllable conductivity of semiconductors forms the foundation of modern electronics, from computer chips to photovoltaics to wireless communications to the laser.

However, the semiconductor was not always such a star material. It wasn’t until 1910 that the term was coined, and even decades afterward, many physicists questioned whether semiconductors even existed. In 1931, future Nobel laureate Wolfgang Pauli wrote to his colleague Rudolf Peierls:

“On semiconductors one should not do any work, that’s a mess, who knows whether there are semiconductors at all!”

His attitude was hardly uncommon at the time. Another physicist of that era complained:

“What are semiconductors good for? They are good for nothing. They are erratic and not reproducible.”

Among scientists, semiconductors had gained a nasty reputation for being difficult to study because their electrical properties seemed to vary uncontrollably from sample to sample, sometimes differing by as much as a factor of a million. Considering that reproducibility is a bedrock principle of the scientific method, it was extremely challenging to study materials that could not be reproduced reliably.

Progress in the field of semiconductors remained sluggish until World War II, when the military’s well-funded desire for radar systems stoked demand for semiconductor diode rectifiers. Around the same time, scientists finally began to reliably control the doping level of semiconductors, as they switched away from self-doping materials like copper oxide and instead focused on impurity-doped materials like germanium and a rather pedestrian element found in dirt and sand: silicon.

Once we had harnessed silicon, the rest was history.

The most interesting technologies are those that sometimes do one thing, and sometimes do another. The most interesting people are those who sometimes do one thing, and sometimes do another. The most interesting stories are those where the plot sometimes goes one way, and sometimes goes another.

Being the same way all the time is boring, and worse, not all that useful.

This is the power of “sometimes.”

The Moore's Law of Moore's Laws

Mon, 22 Jun 2015 00:00:00 +0000

The Moore’s Law of Moore’s Laws

Attention is scarce. And attention is valuable.

Attention is scarce because there is limited time in the day but an unlimited number of things to potentially care about. As an example, consider YouTube. So much video is uploaded to YouTube that if you tried to watch just a single day of uploads you’d die of old age before getting through them.¹ In the same way that it’s impossible to pay attention to anything more than a tiny slice of YouTube, it’s impossible to pay attention to anything more than a tiny slice of reality. This is why attention is scarce.

By itself, the scarcity of attention does not give it value. For example, toxic nuclear waste is scarce, but it is not valuable.²

The reason attention is valuable is that it’s almost always a prerequisite for impact. A researcher whose papers get no attention will not lead others to new discoveries. A founder whose startup gets no attention will find no investors or customers. A journalist whose articles get no attention will entertain no one. Ultimately, it’s difficult to impact people without first earning their attention.

Earning attention is easy if you’ve already had large impact. This is one reason why success begets success. But before your success, how can you get attention?

Here is one simple strategy: Are you researching nuclear fusion? Then talk about the Moore’s law of fusion! Are you founding a battery startup? Talk about the Moore’s law of batteries! Are you writing an article about how paint dries? Talk about the Moore’s law of drying paint!

Comparing your work to Moore’s law is an instant recipe for getting your audience to imagine how you’ll spawn entire industries and transform the world. And the best part about exponential growth is that it starts small. If anyone ever questions whether your work is truly world-changing, you can rebut that while your first steps are admittedly small on an absolute scale, that is how all exponential curves start. Only after growing exponentially for decades will they then become huge.

Riding coattails is a time-tested strategy, and in the world of technology, there are no coattails bigger than those of Moore’s law. So it is hardly a surprise to see more and more people coining their progress as the “Moore’s law of __.” In the past few years, I have heard people reference the Moore’s law of solar power, the Moore’s law of mass spectroscopy, and the Moore’s law of aircraft, to name just a few.

Seeing this explosion of references to Moore’s law got me thinking along the same lines as Gordon Moore 50 years ago. How fast are these pseudo-Moore’s laws being invented? Are they following any sort of regular trend that would allow us to cautiously extrapolate? Is this merely a short-term fad or the beginning of a tectonic shift in technological marketing?

To answer these questions, I scoured the web for every reference I could find to the “Moore’s law of __” and then traced each back to its earliest mention. In doing so (and with surprisingly little researcher bias) I made a stunning discovery:

There is a Moore’s law of Moore’s laws.

For the past four decades, the number of Moore’s laws coined has been doubling every five years. Early Moore’s laws were technical in nature (e.g., the Moore’s law of software or the Moore’s law of bandwidth), but as time passed, Moore’s laws were invented for topics far afield of computers (e.g., the Moore’s law of corn or the Moore’s law for NFL field goals).

Let’s extrapolate this forward. If the number of Moore’s laws continues to double every five years, then by the year 2080, there will exist a Moore’s law for every single word in the English language.³ We may have a Moore’s law of zebras, a Moore’s law of love, and perhaps even a Moore’s law of absquatulating.

So if you want to use this technique to harness your audience’s scarce and valuable attention, do it soon, before this land grab ends and the entire English dictionary has been claimed.

That’s what I’ve done just now, with the Moore’s Law of Moore’s Laws.

A version of this article was published in the November 2015 issue of MRS Bulletin. Shockingly, it has been cited three times in academic literature. Author and consultant Venkatesh Rao has coined it Sanders’ Law.

Sources:

Footnotes

In 2018, YouTube shared that 500 hours of video were uploaded each minute. This is roughly a century of video per day. ↩
The United States Department of Energy spends a huge chunk of its annual budget on nuclear waste management ($6B, ~1/3 of all non-weapons spending). It forecasts hundreds of billions of dollars in liability for future cleanup. If you look into it, you’ll find it’s a really tragic topic. ↩
The English language comprises a few hundred thousand to a million words, depending on your degree of vocabulistic conservatism. But in the realm of exponential growth, a spare factor of 2 or 10 makes little difference. ↩

Do not expect the unexpected

Do not expect the unexpected

Footnotes

Can environmentalism be positive?

Can environmentalism be positive?

On Google

On Google

Notes

What changed in tech from 2010 to 2020?

What changed in tech from 2010 to 2020?

What changed in hardware

What changed in software

Churchill tries to predict 1981

Churchill tries to predict 1981

Why do we teach public speaking but not public listening?

Why do we teach public speaking but not public listening?

The biggest bug in the English language

The biggest bug in the English language

On Netflix

On Netflix

Why do some misconceptions persist?

Why do some misconceptions persist?

On the zipper merge

On the zipper merge

Appendix:

When should you not zipper merge?

What does the literature say?

Other safety considerations

Prices are tools for underdogs

Prices are tools for underdogs

i. The parable of the lemonade stand

ii. China

iii. When can you not compete on price?

iv. Google’s lemonade

v. My main point

Good news about malaria

Good news about malaria

Why does Google think the slowest animal is the sloth?

Why does Google think the slowest animal is the sloth?

On technology

On technology

On spaceflight

On spaceflight

Other reading

Supply of spaceflight

Cost structure: Why does it cost $4,000/kg to launch stuff into orbit?

Physical limits: How low can the cost go?

Conservation of energy

Conservation of momentum

Historical progress

Propulsion choices

Demand for space travel beyond Earth orbit

Notes

On self-driving cars

Is data scientist the hottest job in America?

Is data scientist the hottest job in America?

Footnotes

Predictions about the year 2050

Predictions about the year 2050

The power of sometimes

The power of “sometimes”

The Moore's Law of Moore's Laws

The Moore’s Law of Moore’s Laws

Sources:

Footnotes