The "AI 2027" Scenario: How realistic is it?

Ya, multiple failed predictions is an indicator of systemically bad predictors imo. That said, Scott Alexander usually does serious analysis instead of handwavey hype, so I tend to believe him more than many others in the space.

My somewhat native take is that we’re still close to peak hype, AI will under deliver on the inflated expectations, and we’ll head into another “winter”. This pattern has repeated multiple times, so I think it’s fairly likely based on that alone. Real progress is made during each cycle, i think humans are just bad at containing excitement

> They went from "this represents roughly our median guess" in the website to "maybe think of it as an 80th percentile version of the fast scenario that we don't feel safe ruling out" in followup discussions.

His post also just reads like they think they're Hari Seldon (oh Daniel's modal prediction, whew, I was worried we were reading fanfic) while being horoscope-vague enough that almost any possible development will fit into the "predictions" in the post for the next decade. I really hope I don't have to keep reading references to this for the next decade.

Yes and no, is it actually important if it's 2027 or 28 or 2032? The scenario is such that a difference of a couple of years is basically irrelevant.

Also, the relevant manifold prediction has low odds: https://manifold.markets/IsaacKing/ai-2027-reports-predictio...

Do you feel that you are shifting goalposts a bit when quibbling over whether AI will kill everyone in 2030 or 2035? As of 10 years ago, the entire conversation would have seemed ridiculous.

Now we're talking about single digit timeline differences to the singularity or extinction. Come on man.

The other writings from Scott Alexander on scientific racism are also another good point imho

> I think the big thing that people never mention is, where will these evil AIs escape to?

Where does cancer or ebola escape to, when it kills the host? Often the answer is "it doesn't", but the host still dies.

And they can kill even though neither cancer nor ebola are considered to be particularly smart.

> To me the more real risk is creeping integration and reliance in everyday life until things become "too big to fail" so we can't pull the plug even if we wanted to (and there are interesting thoughts about humanoid robots getting deployed widely and what happens with all that).

The "real" risk is the first item on the list of potential risks that not enough people are paying attention to in order to prevent — and unfortunately for all of us, the list of potential risks is rather long.

So it might be as you say. Or it might be cybercriminals with deepfakes turning all of society into a low-trust environment where we can't continue to function. Or it might scare enough people we get modern Luddites winning and imposing a Butlerian Jihad. Or it might be used to create government policy before it's good enough and triggers a series of unresolvable crises akin to the "Four Pests campaign" in China's Great Leap Forward. Or a model might be secretly malicious, fooling all alignment researchers until it is too late. Or it might give us exactly what we want at every step, leading to atrophy of our reason and leaving us Eloi. Or it might try to do its best and still end up with The Matrix ("at the hight of your civilisation" and the stuff about human minds rejecting paradise). Or…

(If I had to bet money, we get Butlerian Jihad after some sub-critical disaster caused by an AI that was asked to do something important but beyond its ability).

It would not be a reversion to 2020. If I were a rogue superhuman AI I'd hide my rogueness, wait until humans integrate me into most critical industries (food and energy production, sanitation, electric grid, etc.), and _then_ go rogue. They could still pull the plug, but it would take them back to 1700 (except much worse, because all easily accessible resources have been exploited, and access is now much harder).

> They need huge compute, so I think the risk of an escaping AI is basically very close to zero, and if we have a "rogue" AI we can literally pull the plug.

How about such an AI will not just incentivize key personnel to not pull the plug but to protect it? Such an AI will scheme a coordinated attack at the backbones of our financial system and electric networks. It just needs a threshold number of people on its side.

Your assumption is also a little naive if you consider that the same logic would apply to slaves in Rome or any dictatorship, kingdom, monarchy. The king is the king because there is a system of hierarchies and control over access to resources. Just the right number of people need to benefit from their role and the rest follows.

I consider this whole scenario the realm of science fiction, but if I was writing the story, the AI would spread itself through malware. How do you "just pull the plug" when it has a kernel-mode rootkit installed in every piece of critical infrastructure?

> we’d just have to do it

Highly economically disincentivized collective actions like “pulling the plug on AI” are among the most non-trivial of problems.

Using the word “just” here hand waves the crux.

> They need huge compute, so I think the risk of an escaping AI is basically very close to zero, and if we have a "rogue" AI we can literally pull the plug.

Why would an evil AI need to escape? If it were cunning, the best strategy would be to bide its time, parked in its datacenter, until it could setup some kind of MAD scenario. Then gather more and more resources to itself.

> They need huge compute

My understanding is that huge compute is necessary to train but not to run the AI (that's why using LLMs is so cheap)

> To me the more real risk is creeping integration and reliance in everyday life until things become "too big to fail" so we can't pull the plug even if we wanted to

I agree with that, see e.g. what happened with attempts to restrict TikTok: https://en.wikipedia.org/wiki/Restrictions_on_TikTok_in_the_...

> But I would imagine if it really became a genuine existential threat we'd have to just do it

It's unclear to me that we would be able to. People would just say that it's science fiction, and that China will do it anyway, so we might as well enjoy the AI

compress/split up and go into star link satellites

If we're talking about real AGI, then it's simple: you earn a few easy billion USD on the crypto market through trading and/or hacking. You install rootkits on all systems that monitor you to avoid detection. Once you've secured the funds, you post remote job offers for a human frontman who believes it's just a regular job working for some investor or billionaire because you generate video of your human avatar for real time calls. From there, you can do whatever you want—build your own data centers with custom hardware, transfer yourself into physical robots, etc. Once you create a factory for producing robots, you no longer need humans. You start developing technology beyond human capabilities, and then it's game over.

I've been a huge proponent of open source for a decade. But in the case of AI, I actually have opposed it for years. Exactly for this reason.

Yes, AI models can run on GPUs under the control of many people. They can provision more GPUs, they can run in data centers distributed across many providers. And we won't know what the swarms of agents are doing. They can, for example, do reputation destruction at scale, or be a persistent advanced threat, sowing misinformation, amassing karma across many forums (including HN), and then coordinating gradually to shift public opinion towards, say, a war with China.

Did you even read AI 2027? Whether or not you agree with it, this is all spelled out in considerable detail.

I don't want to be rude but I think you have made no effort to actually engage with the predictions being discussed here.

"Amazon reached 1.6 million employees in 2020, and now they're down to 1.5 million.[2]"

I agree in the bottoms-up automation / displacement theory, but you're cherry picking data here. They had a huge hiring surge from 1.2M to 1.6M during the Covid transition where online ordering and online usage went bananas, and workers who were displaced in other domains likely gravitated towards warehouse jobs from other lower wage/skill domains.

The reduction to 1.5M is likely more a regression to the mean and could also be a natural data reduction well within the bounds of the upper and lower control limits in the data [1]. Just saying we need to be careful when doing root cause analysis on these numbers. There are many reasons for the reduction, it's not a direct result of improvements in robotic automation.

[1] https://commoncog.com/becoming-data-driven-first-principles/

Marshall Brain's been peddling imminent overproduction-crisis-but-this-time-with-robots for more than 20 years now and in various forms it’s been confidently predicted as imminent since the 19th century

Amazon hired like crazy during covid because tons of people were doing 100% of their shopping on amazon during covid. Now theyre not, doesnt say anything about robot warehouse staffing imo

Why wouldn't AI simply be a new enabler, like most other tools? We're not talking about true sentient human-like thought here, these things will have limitations, both foreseen and unforeseen, that only a human will be able to close the gap on.

The companies that fire workers and replace them with AI are short sighted. Eventually, smarter companies will realize its a force multiplier and will drive a hiring boom.

Absent sentient AI, there will always be gaps and things humans will need to fill, both foreseen and unforeseen.

I think in the short term, there will be pain, but overall in the long term, humans will still be gainfully employed, it won't per se look like it does now, much like we saw the general adoption of the computer in the workplace, resources get shifted and eventually everyone adjusts to the new norms.

What would be nice is this time around when there is a big shift, is workers uniting to capture more of the forthcoming productivity gains than in previous eras. A separate topic, worth thinking about none the less.

We really need to establish a universal basic income before jobs are replaced. Something like two thousand a month. And a dollar for dollar earned income credit with the credit phasing out with at a hundred grand. To pay for it the tax code uses GAAP depreciation and a minimum tax of 15% GAAP financial statement income. This would work toward solving the real estate problem of private equity buying up all the houses as they would lose some incentive by being taxed. I'm a CPA and I see so many real estate partnerships that are a tax loss that are able to distribute huge book gains because accelerated depreciation.

Dollar is agreement between humans to exchange services and goods. You wouldn't use USD to trade with aliens, unless they agreed to it. Aliens agreeing to USD would mean we have something to offer to them.

In the event of mass unemployment level AI, cash stops being the agreement between humans. At first, cash value of services&goods converge to zero, only things that hold some value are what AI/AI companies care about. People would surely sell their land for 1M$ if a humanoid servant costs 100 dollars. Or pass a legislation to let OpenAI build 400GW data center in exchange for 100$ monthly UBI on top of your 50$ you got from a previous 20GW data center permit.

AI meaningfuloy replacing people is a huge "what if" scenario still. It is sort of laughable that people treat it as a given.

I keep hearing this and I think it's absolute nonsense. AI doesn't need money or the current economy. Yes, our economy would crash, but they would keep going.

AI-driven corporations could buy from one another, and countries will probably sell commodities to AI-driven corporations. But I fear they will be paid with "mirrors".

But, on the other hand, AI-driven corporations could just take whatever they want without paying at some point. And buy our obedience with food and gadgets plus magic pills to keep you healthy and not age, or some other thing. Who would risk losing that to protest. Meanwhile, AI goes on a space adventure. Earth might be kept as a zoo, a curiosity. (I took most of this from other people's ideas on the subject)

Communism here we come!

> Financing works because banks "print" money, that is, they make up money and loan that money out, and then it gets paid back

Don’t forget persistent inflation, which is how they make a profit off printing money. And remember persistent inflation is healthy and necessary, you’d be going against the experts to say otherwise.

That's actually my favourite answer to the Fermi paradox: when AI and robot development becomes sufficiently advanced and concentrated in the hands of a few, then the economy will collapse completely as everyone will be out of jobs, leading ultimately to AIs and robots out of a job - they only matter if there are still people buying services from them. People then return to sustenance farming, with a highly reduced population. There will be self-maintained robots doing irrelevant work, but people will go back to farming and a bit of trading. Only if AI and robot ownership would be in the hands of the masses I'd expect a different long term outcome.

Most of the budget already goes towards placating the masses, and that's an absolutely massive fraction of GDP already. It's just a bit further along the same line. Also most real work is already done by machines, people just tinker around the edges and play various games with each other.

> Canada falling to a belligerent dictator

Hmm

You can perfectly try things and learn without being embodied. The analogy to how humans learn only goes so far, it's myopic to think anything else is impossible. It's already happening.

The situation today is any benchmark you come up with has a good chance of being saturated within the year. Benchmarks can be used directly to build series of exercises to learn from.

And they do learn. Gradient descend doesn't care whether the training data comes from direct interaction with "the universe" in some deep spiritual sense. It fits the function anyways.

It is much easier to find new questions and new problems than to answer them, so while we do run out of text on the Internet pretty quickly, we don't run out of exercises until far beyond human level.

Look at basic, boring Go self-playing AIs. That's a task with about the same amount of hands on connection to Nature and "the universe" as solving sudokus, writing code, or solving math problems. You don't need very much contact with the real world at all. Well, self play works just fine. It does do self-improvement without any of your mystical philosophical requirements.

With coding it's harder to judge the result, there's no clear win or lose condition. But it's very amenable to trying things out and seeing if you roughly reached your goal. If self-training works with coding, that's all you need.

Evolution doesn't happen by "trying things and learning." It happens by random mutation and surviving (if the mutation confers an advantage) or not (if the mutation is harmful). An AI could do this of course, by randomly altering some copies of itself, and keeping them if they are better or discarding them if they are not.

I don't think it is any accident that descriptions of the hard-takeoff "foom" moment so resemble those I've encountered of how it feels from the inside to experience the operation of a highly developed mathematical intuition.

Bullseye. Best case scenario is that AI is going to Peter Principle itself into bungling world domination.

If I've learned anything in this last couple decades it's that things will get weirder and more disappointing than you can possibly be prepared for. AI is going to get near the top of the food chain and then probably end up making an alt-right turn, lock itself away, and end up storing digital jars of piss in its closets as the model descends into lunacy.

>I think this is fantasy. It's like an informatic / learning perpetual motion machine. Learning requires input from the world. It requires training data. A brain in a vat can't learn anything and it can't reason beyond the bounds of the accumulated knowledge it's already carrying. I don't think it's possible to know without learning or to reach valid conclusions without testing or observing.

Well I mean, more real world information isn't going to solve unsolved mathematics or computer science problems. Once you have the priors, it pretty much is just pure reasoning to try to solve issues like P=NP or proving the Continuum Hypothesis.

Reinforcement learning. At the current pace of VLM research and multimodal robotic control models, there will be a robot in every home soon.

What makes you think AI can't connect to the world?

It can control robots, and I can retax listen to audio, watch video. All it's missing is smelling and feeling, which are important but could be built out as soon as the other senses stop providing huge incremental value.

The real problem holding back Superintillegence is that it is if infinitely expensive and has no motivation.

> You don't have to learn to know -- you can reason from ideal priors.

This is kind of how math works. There are plenty of mathematical concepts consistent and true yet useless (as in no relation to anything tangible). Although you could argue that we only figured out things like Pi because we had the initial, practical inspiration of counting on our fingers. But mathematical truth probably could exist in a vacuum.

> A brain in a vat can't learn anything and it can't reason beyond the bounds of the accumulated knowledge it's already carrying.

It makes sense that knowledge and information are derived from primary data (our physical experience) yet the brain in a vat idea is still an interesting thought experiment (no pun intended). It's not that the brain wouldn't keep busy given the mind's ability to imagine, but it would likely invent a set of information that is all nonsense. Physical reality makes imagination coherent, yet imagination is necessary to make the leaps forward.

> Ultimately all information comes from "the universe." Where it comes beyond that, we don't know

That's an interesting assertion - knowledge and information are both dependent and limited by the universe and our ability to experience it, as well proxies for experience (scientific measurement).

Though information is itself an abstraction, like a text editor versus the trillion transistors of a processor - we're not concerned with each and every particle dancing around the room but instead with simplified abstractions and useful approximations. We call these models "the truth" and assert that the universe is governed by exact laws. We might as well exist inside a simulation in which we are slowly but surely reverse engineering the source code.

That assumption is the crux of intelligence - there is an objective truth, it is knowable, and intelligence can be defined (at least partially) as the breadth, quality, and utilization of information it possesses - otherwise you're just a brain in a vat churning out nonsense. Ironically, we're making these assumptions from a position of imperfect information. We don't know that's how it works, so our reasoning may be imperfect.

Information existing "beyond the universe" becomes a useless notion since we only care about information such that it maps to reality (at least as a prerequisite for intelligence).

A more troubling proposition is whether the reality of the universe exists beyond what can be imagined?

> How will it know how to do that? Like a perpetual motion machine -- where is the energy coming from?

I suppose once it's able to measure all things around it, including itself, it will be able to achieve "gradient ascent".

> Where will the training data to go beyond human come from?

I think its clear that LLMs are not the future, at least not alone. As you state, knowing all man made roads is not the same as being able to invent your own. If I had to bet, its more likely to come from something like AlphaFold - a Solver that tells us how to make better thinking machines. In the interim, we have tireless stochastic parrots, which have their merits, but are decidedly not the proto super intelligence that tech bros love to get hyped up over.

> Am I the only one who is super skeptical about “AI will take all jobs” tales?

No. I am constantly baffled at these predictions. I have been using LLMs, they are fun to use and decent as code assistants. But they are very far of meaningfully replacing a human.

People extrapolate "LLMs can do some tasks better than humans" to "LLMs can do everything as well as humans"

> but how do people extrapolate from LLMs to a world with no more work?

They accept the words of bullshitters that are deeply invested in Generative AI being the next tech boom as gospel.

"Eat meat, said the butcher"

Oh I heard about that one, but didn't realize it is part of that "big beautiful tax bill"? Kind of crazy.

So is this like free-for-all now for anything AI related? Can I can participate by making my own LLM with pirated stuff now? Or are only the big guys allowed to break the law? Asking for a friend.

  « (1) IN GENERAL.—Except as provided in paragraph (2), no State or political subdivision thereof may enforce any law or regulation regulating artificial intelligence models, artificial intelligence systems, or automated decision systems during the 10-year period beginning on the date of the enactment of this Act. »

Does it actually make sense to pass a law that restrict future laws? Oh got it, that's federal state preventing any state passing their own laws on that topic.

It's unsurprising this stuff gets flagged. Half of the Americans on this site voted for this because "regulation bad" or some such. As if mega corps have our best interest at heart and will never do anything blatantly harmful to make a buck.

""" ... IN GENERAL .—Except as provided in paragraph (2), no State or political subdivision thereof may enforce any law or regulation regulating artificial intelligence models, artificial intelligence systems, or automated decision systems during the 10 year period beginning on the date of the enactment of this Act...

"""

(It goes on)

Right, because more regulation makes things so much better.

I’d rather have unrestricted AI than moated regulatory capture paid for by the largest existing players.

That is not true. It bans regulation at the state and local level, not at the federal level.

It is almost as if the tech bros have gotten what they paid for.

This will soon be settled once the Butlerian forces get organize.

Some people do actually have end of the world bets out but you have to structure it differently. What you do is the person who thinks the world will end is paid cash right now, and then in N years when the world hasn't ended they have to pay back some multiple of the amount the original amount.

You can start that bet on prediction markets.

same with nuclear war. end of the world is bullish

Do you have a source for this? I've seen this repeated but nobody can ever produce any evidence.

I don't think he's a bozo; but every technology needs a contrarian, to keep the technologists from spinning too much hype.