The AI hype bubble is the new crypto hype bubble
doctorow.medium.com
doctorow.medium.com
In that sense the current state of AI is less like crypto and more like the dotcom bubble of the early 00s. No one really understands the underlying tech well enough, but everyone wants to be involved. There are companies surging in valuation just by adding AI to their name. While all of this will correct itself, the underlying tech (the web in 2000, AI today) will eventually prove to be world changing.
In that sense, the current state of crypto is less like social media and more like the dotcom bubble of the early 00s. No one really understands the underlying tech well enough, but everyone wants to be involved. There are companies surging in valuation just by adding "crypto" to their name. While all of this will correct itself, the underlying tech (the web in 2000, crypto today) will eventually prove to be world changing.
It's important to note that this is a hypothetical scenario where we're replacing terms to compare different industries. The statements above are not a reflection of actual market conditions or predictions about the future success of specific sectors.
No sed/shitcoin required.
Any company who creates digital assets, say games, wants to be in complete control. Being distributed isn't a benefit.
Moving around money anonymously has been largely used for crime. Sure, there are other users, but not enough.
And the speculation and manipulation of blockchain currency rates had made them so volatile that they are not really usable as currency.
AI (or whatever will end up calling it, machine learning is better than AI) can actually solve problems, or ease everyday tasks, for everybody. It's not a matter of who'll use it, it's what it'll be used for the most.
The space has suffered a lot from adverse selection towards grifters. UST alone captured many of the earnest attempts towards payments and took them down with it.
Use some of the creativity we've all seen you exhibit to ask GPT-4 questions that require complex reasoning inside a world model, rather than regurgitating its training set. It's not going to get it right all of the time, but the fact that it can get it right any of the time is astonishing, and it's going to get quickly better, perhaps even purely through compute spend increase rather than algorithmic breakthrough.
I'm not a crypto or AI advocate, but it's easy to see how "technology with overestimated social consequences" could get tiring for everyone who isn't an investor. The "fundamental difference" between cryptocurrency and AI is that one is built on reproducible and accountable logging and the other is guesswork. Trying to champion one or the other feels like pointless virtue signalling really.
I also have to admit I really don't like Cory Doctorow's other works (I especially despise the reductive theory of "enshittification") but his thesis is 100% the truth. People need new buzzwords to cling to, and AI is the one with the most popular demo. The usefulness is an accessory to the marketing and positioning of AI-powered products.
To most people the main use of AI is:
Code suggestions
Code analysis
Document summarization
Document creation
Clipart generation
Audio transcription
Audio synthesis
Realistic chat bots
...
These are not dubious use cases.
The number of people who have been defrauded by crypto number in the millions. The number of people who have been defrauded by AI are zero.
Even if AI is wrong 90% of the time, 10% of the time it makes someone's job easier. Making someone's job easier is objectively more valuable than making someone's bank account number go up. One adds wealth to society. The other redistributes existing wealth.
I can go on about the differences but I'll let you tell me about how Ethereum has a really good logging system first.
> The number of people who have been defrauded by AI are zero.
What's the number of people who have been murdered by AI? It's a much more dubious figure, but a larger one to be sure. If you want to move the goalpost to which one is more harmful, I still think you hamstring your attack on cryptocurrency.
Why do people keep saying their opinions as if they are fact? Whether or not I refuted your point is independent from your assertions of such.
> Both technologies provide utility that is not all-encompassing.
No. Crypto provides no utility. At all. AI provides a lot of utility, some of the time. And you know what, you're wrong. AI is a panacea. It's called general purpose transformer ffs. And not by marketers, but by actual engineers who developed the Transformer architecture. (Obviously I'm being a little facetious but AI is infinitely more of a cure-all than crypto)
> What's the number of people who have been murdered by AI?
The second most common use case of crypto is for anonymous transactions for illegal drugs and weapons. So that's all I'm going to say on that point.
That's just blatantly untrue. If there is anything you cannot do with normal money, cryptocurrency has utility. You can argue that it's overall negative utility or insignificant compared to the glory of AI (you may be right) but utility is inherent to any system of payment with more than 1 participant.
> AI is a panacea. It's called general purpose transformer ffs.
That's more than just a little facetious. Our AI generates text - it does nothing other than that. It's admittedly great at manipulating text, but that's where it's "general purpose transforming" stops. It encodes human concepts like emotions and spatial reasoning into words, which are incomplete analogs to how they work in the real world. It has impressive results, but comparing the value of human labor to the work AI can do is startlingly parallel to comparing the value of a database to the work a blockchain can do. It almost provides the basis for reality-distortion to investors, which is where those "marginal utility" AI/crypto companies come from.
> The second most common use case of crypto is for anonymous transactions for illegal drugs and weapons
The first most common use case of AI is for military purposes.
I will refer you to the Vision Transformer: https://arxiv.org/abs/2010.11929
I suggest you do a literature review on the applications of ANNs, CNNs and attention mechanisms. They are quite numerous but I will concede that they generally 'just' do compression.
AI encodes human concepts like emotions and spatial reasoning into vectors. Which, their analogous-ness to the real world, is more of a philosophical question than a practical one at this point.
> It has impressive results, but comparing the value of human labor to the work AI can do is startlingly parallel to comparing the value of a database to the work a blockchain can do.
I mean 'human labor' is a very general thing, compared to 'database' IMO.
> The first most common use case of AI is for military purposes.
This is true but the AI that the military uses is probably a far cry from what the public is able to access - and not from a technological advancement but a specificity standpoint. The military will want deterministic, non-opaque methods that are probably closer to the policy-gradient recommendation systems of old. Think Netflix and TikTok, not GPT.
I can say from an investor standpoint AI is probably a bubble but only because its entire foundations are open source. There's no money to be made from it, therefore it's a bubble.
Crypto on the other hand always had somewhat dubious foundations. There are few use cases for it, therefore it's a bubble.
Crypto's killer feature is being an open standard for finance that anyone can integrate with. Having a use case in drug and weapons transactions is still utility though, you're disagreeing with yourself
Ai doesn't do anything you couldn't hire a person to do, and you still need a expert person to vet it's outputs, so I think it also provides no utility?
Not true. I’m seeing fake voice scam stories on the news lately, fake kidnapping claims, etc
The closest to fraud AI has done is its hallucinatory outputs. And fair enough, this is enough to hold AI back from being 'it'.
Ai has 1000000x more scam potential than crypto
I don’t understand the comparison
And let's say I was - crypto promises utter anonymity in theory but is actually traceable in practice. So a bit of an academic scam on that front.
So... you mean to insinuate the Sam Altman's current "AI Will Kill Us All Tour" isn't a hype bubble?
> crypto promises utter anonymity
Where?
Cryptocurrency promises a separate ledger for transactions where you don't need to be identified by name. If you identify yourself to buy crypto, then yes, you have defeated the purpose. Cryptocurrency doesn't "promise" anonymity any more than encryption "promises" security, though. If that's academic fraud then I'm Abe Lincoln.
Its CEO did, with Worldcoin. Which is pivoting to be essentially the same with an AI spin.
https://www.technologyreview.com/2022/04/06/1048981/worldcoi...
https://www.buzzfeednews.com/article/richardnieva/worldcoin-...
Being 90% wrong causes a lot of extra work, severely reducing the gains of that 10%.
Not only is it a bad use case, it is self destructive. The only value a currency might have is if it is used in trade, holding currency "because the value goes up" is making it less useful as a currency.
Also BTC is very insufficient as an actual currency, especially from a purely technical standpoint.
No, I mean something very specific - I can do things at work with LLMs that I couldn't do a year ago without them. I can't say the same about anything crypto related.
If you think that AI is just spitting our word we've seen or that it's guesswork, I don't know how you can even discuss it in good faith.
It's great stuff, and I'd be lying if I said I liked cryptocurrency and don't use AI. The parallels are starkly visible though, and the "overpromise/underdeliver" cycle is already much further along than cryptocurrency ever got. The bubble will eventually pop because people are dissatisfied with the pace of progress, and then everyone will look for a scapegoat. We're already halfway there, judging by the daily deluge of "Ask HN: Why Does It Feel Like OpenAI Nerfed ChatGPT?" posts.
On the another hand after years of deep learning, there is still no viable efficient methods of training, fine tuning or inference without needing more data centers as the data and the AI model scales and its hype is currently contributing to the increased wastage of water and resources all for the sake of incinerating the planet to produce broken and hallucinating black-box AI models.
Being ‘useful’ isn’t an excuse for not finding greener efficient alternatives that significantly lower emissions rather than continuing to burn the planet over the new ‘hype’.
You're telling me right here that absolutely 'no-one' is using Ethereum? Surely you are not at all serious about what you said. There are many companies like Moneygram, Stripe, Ticketmaster, Checkout.com, etc that are using crypto currencies like Ethereum which are on greener consensus models.
Ethereum made it possible to switch from a wasteful consensus model like proof-of-work to proof-of-stake and it significantly reduced its emissions by 99.992% [0]. Even if your ludicrous claim of absolutely 'no one uses it for anything either', it is not incinerating the planet or wasting as much resources and energy unlike AI (deep learning) that is currently doing today.
After a decade of deep learning there have been no efficient alternatives to their existing operations and it brings in more resource waste and continues incinerates the planet. Being 'useful' is not a great excuse to justify burning the planet.
[0] https://consensys.net/blog/press-release/ethereum-blockchain...
We don’t really need general AI, nor are we going to be able to make a true flawless general AI anytime soon. AI was always a really vague marketing term, and can be applied to anything for instant relevancy boost. Bubble territory.
The LLMs we have now are extremely effective though. There’s real use case here for automating writing, and some menial tasks that human beings were unfairly burdened with, but were impossible to automate till now.
That’ll stick around. (Considering that ML concepts have been around for decades, you could even say they already have!)
There isn't any kind of distinction - when people say AI in 2023 they mean ML, LLMs, image models etc.
There’s a lot of knee jerk products out there from businesses that were scared of being left behind on the wave of AI that don’t actually add much. “AI powered” could mean an in-house trained deep learning network. Or an api call to ChatGPT. Or single function someone made by tweaking sliders in Desmos calculator till the curve had the right vibe. It’s the slapping “AI” over “black box magic process” that’s going to implode.
But there’s obviously massive work being done in ML that will change the world. Something actually worth the hype.
Some people take issue with that assertion. The biggest problem with LLMs is you can’t trust them and have to verify everything they output - and often writing it yourself and verifying is easier than verifying someone/something else’s work.
Right now no sane corporation will even let a LLM run their helplines that they out sourced to India. Imagine a LLM hallucinating a dangerous “solution” to a customer problem resulting in loss of property, injuries, or even death. It’s a massive lawsuit waiting to happen.
A bursting bubble? Not in my opinion. I'm already able to capture value in ways that were thought too difficult just 6 months ago. That's not vaporware.
> AI isn’t “artificial” and it’s not “intelligent.” “Machine learning” doesn’t learn. On this week’s Trashfuture podcast, they made an excellent (and profane and hilarious) case that ChatGPT is best understood as a sophisticated form of autocomplete — not our new robot overlord.
This is just argument by assertion. We have no good definition of intelligence, so I have no clue how he can be so confident. "Machine learning doesn't learn" is a crazy take, since "backprop + gradient descent does learn" is close to the most well-supported thing you can say about the past few years of algorithmic progress.
> sophisticated autocomplete
Aside from this being an incredibly reductive sneer that clearly isn't true if you've honestly tried using ChatGPT, etc., his citation for this is a podcast, which I'm positive Doctorow would not accept as sufficient for basically any other technical topic.
I love Ted Chiang's stories, and some of his takes on AI progress are cited here. However, I also found his extensive conversation with the Financial Times (earlier this month, so published after this) disappointing along similar lines. The thread running through both of these is a complete lack of a positive vision for the future, replaced by an almost smug cynicism that asserts any more technological progress is simply hype, a grift, and bad. Are there any current science fiction authors with a positive vision of the future?
Without concrete definitions your assertions are just as correct as theirs. But they have the evidence of absurd tech-bro hype of past technologies to draw on.
> I love Ted Chiang's stories, and some of his takes on AI progress are cited here. However, I also found his extensive conversation with the Financial Times (earlier this month, so published after this) disappointing along similar lines.
"I love Ted Chiang's stories because they jive with my preconceived notions, but I like him less when he says things that I don't believe"
> The thread running through both of these is a complete lack of a positive vision for the future, replaced by an almost smug cynicism that asserts any more technological progress is simply hype, a grift, and bad. Are there any current science fiction authors with a positive vision of the future?
Plenty. They talked about flying cars and living on the moon. Instead we got stagnant wages and a social-media skinner box. All of those wonderfully positive predictions didn't pan out.
Now valuation hype is big question. Will there be moats? Or will it be commodity technology? And maybe server farms will make some money and everyone else marginal profit?
The mobile hype bubble was the new portal hype bubble.
The portal hype bubble was the new dot com hype bubble.
I swear there's a pattern here... Maybe I'm not seeing it...
AI companies and related stocks (NVDA) will be first in line when people realizes we’re not going back to the world of free money in 2018 that fueled a number of hilarious pump-and-dumps over the last 5 years.
Other than that, OP is correct in just about everything in the article - AI isn’t “artificial” or “intelligent”, it’s just a hype word to put on top of a relatively simple ecosystem of applications that may someday prove valuable, but for today are comically bad at everything that isn’t strictly a language exercise (e.g. producing soundalike copy, translating bits of code between apps, etc.). It’s presently a tool that can be no smarter than the user whose hands it occupies, and in fact requires a significant amount of investment to actually “learn” or take direction to be good at some specific tasks. Textbook hype case as far as I can see.
AI has created a lot of euphoria pumped by VC money once again which can only result in disappointment of expectations since this is the 'pump' part of the hype cycle.
Sooner or later we will see if any of these so-called AI companies are actually making any money or will survive at all.
In 2010, almost everyone was a 'tech company'. Now almost everyone is a 'AI company'. This bullshit needs to stop.
Bingo. Good luck telling this crowd that, though - all they know is "black turtleneck" and pestering middle-aged Disney investors for cash.
These AI companies provide marginal utility that will get swallowed by later iterations even if AI is a technical failure. The "industry" of AI is bound to fail whether the technology impresses us or not.
>AI isn’t “artificial” and it’s not “intelligent.” “Machine learning” doesn’t learn.
Is just an insane thing to write.