Something as huge and as unignorable as a whole category of jobs totally gone

I think several industries may have major job displacement (70-85%+) but near-100% job displacement in 4 years is basically not a thing ever for large enough categories of jobs / industries, and pretty much only happens in smaller industries during policy flips, like:

London Stock Exchange “jobbers” all abolished in a day (1986).`

(that being the primary historical example and representing ~4000 jobs tells you something)

So I think this requirement won't be triggered.

The world GDP doubling is even more implausible; The latest doubling started in 2003. The only thing that could make a significant dent in that would be mass high dexterity robot manufacturing, which takes a long time to ramp up.

new tech that's as life-altering as the internet or smartphones -- and that everyone is actually using, just as everyone uses smartphones and the internet today

This one is the most likely to resolve positively imo. Not ~everyone will be actively using AI, but people will be using it much more and for more general and economically valuable tasks than they were the internet in 2006 or the smartphone in 2014 imo. Plausibly both of these today, depending on how you count the impact of the internet (including on AI). as for usage rates similar to smartphones and the internet today, that's just hard: 91% and 96% approx. I don't think we'll get that. Maybe 80-88%. A significant portion of the aging population just does not need any technology, however powerful it becomes, and some people will loathe AI to a degree it is hard to imagine today imo, and may limit usage based on that. As a datapoint:

Only ~10% of [US] adults 65+ say they’ve ever used ChatGPT (vs. 58% under 30; 25% ages 50–64)

As for NO:

AI continues to be a cool and useful tool. Maybe when you talk to a customer service person it's actually an AI -- normal people won't know or care which. There'll be new scams and deepfakes and such but people will learn to adapt and ultimately life will be normal through the 2020s.

The picture painted here is pretty much completely implausible, if you take it as "this is the kind of changes AI will bring"

from FAQ:

dramatic changes in pedagogy

In some sense this is already happening, but teachers being replaced is extremely unlikely for political reasons

What if garbage AI content drastically degrades the internet?

Implausible imo, based on your description. The valuable reviews may be mostly from places restricted to ppl that payed for the service though, like amazon reviews.

What if robocabs replace human-driven Ubers/Lyfts?

if it's as surprising to hear someone's never ridden in a car with no one in the driver's seat as it is for someone to not have a smartphone, we'll count it.

This will probably be close to true in large US cities (a lot of ppl have smartphones), not particularly likely outside of them

Overall, despite the fact that most if not all requirements for either side may not be literally triggered, I think the vibe will be so overwhelmingly no for:

Of course AI is already somewhat life-changing for us nerds but normal humans are largely ignoring it so far. The question is, will they be able to continue to do so or not?

that the market will most likely resolve Yes on the spirit if not the letter (ping me if you want to bet)

@Bayesian Good thoughts! I do want to keep honing the resolution criteria to hew to the spirit. To repeat the warning to traders: ask clarifying questions before betting! Note that most of what's in the FAQ so far is just things that could happen that would make this market easy to resolve. There's still a large gray area, which you're helpfully highlighting.

Question for you: If AI were to hit a wall such that raw capabilities are only marginally better at the end of 2029 than they are today and the tech merely continues diffusing through the economy, getting honed and applied to new use cases, etc, would we agree that that's a NO?

3. What if robocabs replace human-driven Ubers/Lyfts?

If this is universal enough, we'll count it. Namely, if it's as surprising to hear someone's never ridden in a car with no one in the driver's seat as it is for someone to not have a smartphone, we'll count it. (2024 status: Waymo is providing ~650k rides per month and growing.)

This seems to imply that "mere" diffusion could be enough for a YES resolution when it comes to driverless vehicles. The same logic should apply elsewhere too.

Hypothetically, if the tech hit a wall at today's levels, but over the next few years was became as crucial a part of how a significant part of how life works to the extent that it's as surprising to hear someone is without it as for them not to have a smartphone, it seems like that should count as YES.

If, for instance—and I'm not saying this is remotely likely to happen, just that it is a scenario that's a staple of science fiction and thus easy to visualize—the world moves away from the smartphone form factor for the personal computing devices we all carry all the time, and instead people give voice commands to their AI-powered smart glasses or whatever, shouldn't that count? (If it reaches the level of widespread adoption in place of smartphones that smartphones had relative to desktop computers in 2010.) And one can imagine it not needing significant advances to fundamental tech, just better scaffolding.

I don't think there's any doubt that this will happen eventually. For it to happen this decade would mean that AI is pretty fundamentally different from other tech -- personal computers, smartphones, the internet, electricity, radio, plastic. For all of those, it took longer than 5-10 years to reach life-changing-for-muggles levels of saturation, right? What's the closest to a historical precedent for this?

To be clear, when AI becomes AGI, that just really is fundamentally different than any other technology. But assuming no AGI that soon, how realistic is a YES here?

@dreev It needn't be that AI is fundamentally different from other tech for adoption to be faster. Because of previous technologies, adoption can be massively faster than it used to. eg

What would you define as "everyone actually using" a new piece of AI tech? Probably not< 50% of the population, so like 60%? For reference, about 90% of people have a smartphone today.

@DanNelson88fb Good question. I haven't pinned down percentages yet but in another comment thread here I suggested using the adoption of the internet in the US in the year 2000 and the adoption of smartphones in 2010 as benchmarks. So I'd say we don't have to get to ubiquity, but common enough to be clear it's heading that way.

@dreev
In 2010, around 20% of US adults owned smartphones
currently, around 20-25% of US adults use ChatGPT or a similar chatbot at least once a month

I'll assume this is not sufficient for a positive resolution, in part because you probably overestimated the speed of adoption of smartphones?

in the year 2000, around 52% of US adults used the internet
Assuming you want a similar threshold for smartphones, that would be in early 2013.

Extending current trends, 50%+ of US adults will use ChatGPT or a similar chatbot at least once a month in early 2027 (high uncertainty). Would this threshold be sufficient? Admittedly that's not the same as frequent daily use, which is currently much lower (~7%)

@Bayesian Useful numbers; thank you! It's not just the adoption rate though. The thing being adopted has to be impactful enough. Using a chatbot as a better search engine -- even if you use it for that every day -- doesn't meet that threshold, I would say.

@dreev What if people use it as a tutor, as well as a writer (for emails, etc.), and secretary, and work assistant on their jobs (not just to retrieve information, but to save time by replacing their labour with the ai's for several tasks per week), and a few other things that make their life significantly easier? Some people still don't use AI ~at-all but 50%+ of people in the US do use it for more than just a better search engine?

Curiously if you look at the distribution of queries for chatbots like ChatGPT, a sizeable portion is for executing tasks and not for retrieving and presenting information

Question: in which years would hypothetical bets on similar terms but about smartphones or about the internet instead of AI have resolved YES for the first time?

@zzq Ooh, great question. Off the cuff, I'd say 2000 for the internet and 2010 for smartphones. But the impact diffused slowly enough that it's very hard to pick a specific year. This suggests we have work to do in pinning down the resolution criteria here.

EDIT: bogus link! I was confusing it with this: https://manifold.markets/ScottAlexander/in-2028-will-an-ai-be-able-to-gener Which is trading much closer to this market.

https://manifold.markets/SamuelKnoche/in-early-2028-will-an-ai-be-able-to-2x8ud6ld4f

"Full high quality movie to a prompt" by 2028 is trading higher than this market.

I'm interested to know if people really think a world is likely where AI gets that capable and still remains largely a cool trick that's used by nerds.

In early 2028, will an AI be able to generate a full high-quality movie to a prompt?

44% chance. EG "make me a 120 minute Star Trek / Star Wars crossover". It should be more or less comparable to a big-budget studio film, although it doesn't have to pass a full Turing Test as long as it's pretty good. The AI doesn't have to be available to the public, as long as it's confirmed to exist.

@BrendanJackman fair, but that market is radically mispriced imo

@BrendanJackman The movie market should be at 2 or 3%.

I think that Criterion #3 is the tricky one. I know people who have gone without smart phones for a least a period of time and they are fine. For example, if most people are using AIs in the same way that they use Google to search these days - i.e. to ask questions, write emails - would that count? That seems like sort of an extension/improvement of what we have today, not really a phase transition in the same way that the internet was. Something on the scale of the internet would make sense to me as something one truly could not ignore though. I don't know anyone who has gone without the internet recently.

Also what counts as "a whole category" of jobs? E.g. would call center jobs count? Or are we talking bigger scale, like accounting.

Disagree with the premise when it comes to younger people. Essentially every college and high school student is already using LLMs. It’s not niche at all.

@LiamZ Very fair comment. I think I'll hedge that first sentence a bit better. (Currently: "AI is already somewhat life-changing for us nerds but normal humans are largely ignoring it so far.")

In any case, this is about predicting how much further it goes, defined by whether normal people can choose to ignore it. I mean, you can ignore anything if you try hard enough, so we're using smartphones / the internet as the benchmark. So maybe a 7.5 on the technological richter scale: https://forum.beeminder.com/t/more-fake-ai-podcasts-about-beeminder/11936/4?u=dreev

Doubling vs now or doubling vs some projection?

Would dramatic changes to pedagogy count?

@Juniper0rg1m Great questions; Svata and I shall discuss!

@Juniper0rg1m I mean hypothetically, if God came down from the heavens and said "AI will suddenly and immediately double everyone's productivity in 2040", the stock markets would immediately jump in anticipation of this new productivity minus the discount rate of 15 years. Right? So then when productivity did double, we wouldn't see a similarly discontinous jump in asset prices.

So what if that's already happened, and we don't see a discontinous doubling of the economy now because it's already priced in?

Well, productivity numbers are published, so we could be sensitive to that. But we could also just compare to a long term growth trend. I.e. if GDP in 2030 is about what you'd expect from an extrapolation of 1970-2000, I wouldn't call that life changing

@Juniper0rg1m If, in 1970, you assigned a 40% chance of Futurama by 2030, and then increased your confidence by 1 percentage point each year, you'd be 100% confident of Futurama by 2030 by the time 2030 rolled around, but the 2030 gdp would be what you expected extrapolating the trend from 1970-2000.

@GG I don't think we started pricing in AI before 2000 in any meaningful sense. Also I don't think GDP will have anything priced into on the same way it would be in the stock market. It's dependent on actual economic activity. (With caviats relating to the notional value of some assets being transacted)