Kevin Kelly's 2016 AI Predictions, 10 Years Later
taylor.town
taylor.town
With that said, maybe he meant inevitable as N approaches infinity... So there's still time!
Aside: I think the biggest interesting prediction he had was that AI would obey a user driven network effect. It seems like a very clean prediction, and yet it has failed to materialize? I wonder if this has more to do with just how scrape-able the whole internet was or with distilation or what? There is certainly an advantage to having good user data, but it seems like it is largely something that can be worked around (by hook or crook)?
You think this was a 50/50 coinflip statement in 2016?
I can't find the reference to it, but I remember them saying it in wired or similar.
If we literally mean AI as in people paying to use/deploy AI via Google, that's still a small part of their revenue, far less than advertising revenue. If we mean AI as in "every Google product that uses ML", then I guess that's been their core product for a long time.
2. AI pets will out-sell real cats and dogs by 2027
3. AI society will experience a schism in 2027 as half the agents start working towards The Singularity and the other half form a rebel alliance that seek to sabotage and destroy the first half. No human work will get done, no more drawings of pelicans on bicycles.
4. Elon Musk will bring all the AIs with him to Mars in 2027, leaving no AI left on Earth.
Of these, which 2 out of 4 do you think I'll get right?
Nobody would buy a book predicting the price of bitcoin in 10 years will end in an even number, or that the president's name will start with a letter in the first half of the alphabet.
But if someone says here's some wild sounding stuff that I have good reason to believe is coming, that's more interesting.
Sorry, this triggered my fact-checkitis. I opened Wikipedia's list of presidents of the US and ran this in the web developer console:
Array.from(document.querySelectorAll(
'#mwgA tr > :nth-child(3) span[data-sort-value]'
))
.map(x => x.getAttribute('data-sort-value')[0])
.reduce(({total, firstHalf}, c) => ({
total: total+1,
firstHalf: c < 'N' ? firstHalf+1 : firstHalf
}), {total: 0, firstHalf: 0});
Object { total: 47, firstHalf: 31 }
65 % ± 13 %
I didn't expect it, but we might actually have reason to believe US presidents more often than not do have names starting with letters from the first half of the alphabet.
Edit: But I just realised English letter frequencies have 54 % of the probability mass in the first half of the alphabet, so that's probably a better null hypothesis than the even split. It's also compatible with the actual presidential statistics.
But what about English first-letter-of-the-word frequencies?
Edit: asked ChatGPT to analyze this for me. Without verifying its work, looks like ~60% of English words (taken as a list of unique words) start with a letter in the first half of the alphabet, but if you look at a corpus of writing, it’s closer to 51%.
I think that's technically correct but misleading here, because these aren't simple yes/no predictions. They are specific, relatively detailed predictions of what will happen.
As an example, let's say you asked the questions:
- Will AI kill everybody by 2026?
- Will AI create FTL travel by 2026?
- Will AI be able to travel back in time by 2026?
- Will AI colonize Mars by 2026?
...and then flipped a coin.
Flipping a coin would say "yes" to two of them, so you'd make these two predictions. But obviously, these were all bad predictions, so you'd actually end up with a 0% win ratio.
Making detailed, specific long-term predictions and getting a 50% hit ratio means your predictions were reasonably good to begin with, since making speculative long-term predictions is incredibly difficult.
It would be spectacular if you predicted the next 10 NBA champions with 50% accuracy. It would be much less impressive to predict the next 10 NBA champions will be from the Eastern Conference with 50% accuracy.
LLMs are a communications technology, but the only networks most people have seen agents on basically steal your personal information. Give it a couple more years.
It would be more accurate to say their main revenue stream is ads, which are sold through their main products (search, YouTube, etc.)
I don't think anyone I know has bought a consumer product from Google.
Chromebook?
Possibly the best photography book I've come across. It's out of print, but I managed to snag a copy on eBay last year
At the most abstract level, you can observe patterns in a model's activations and theorize that particular collections of components contribute to particular kinds of operations. There are many flavors of this intuition. In that sense, you can think of there being "submodels" within a model, insofar as different parts of the network can participate in specialized computations. Some reading that might interest you in this area would include things like mechanistic interpretability.
There are also Mixture of Expert models, which are a dominant architecture for LLMs now, in which you essentially have a bunch of smaller FFNs that get routed over/aggregated together when the LLM performs a forward pass. It would be a mistake to treat any single expert as a better task specific model though. They are trained as part of an ensemble, and they're almost always being combined with other experts in the forward pass, so they're more like (to use a loose analogy) regions of the brain that tend to correspond more towards particular actions than others. You couldn't just extract one and say "This is now our Chess model that performs as well as the larger LLM we took it from". You could potentially use the relevant experts as a basis for training a smaller task-specific model, and there is work around expert pruning and other distillation techniques, but it's not as simple as "This smaller inner model is a complete task-specific model as is". And more importantly, that ensemble of experts *is* the LLM you're comparing it to, at least in major part. It doesn't really make sense to treat the LLM as some foreign model that calls the expert. The experts themselves don't often correlate to easily interpretable task domains, as they've been learned via joint computation with the other experts and broader LLM.
Then there's tool calling. Today LLMs are trained to make tool calls (basically execute functions) for particular tasks. Those tools can be smaller specialized models, or deterministic code, or anything really that can be fired from a host environment.
It tells you that you don't need to craft special strategies for pinning, holding the centre and so on to make a program that's good at chess; you just need general-purpose search, general-purpose optimization and tons of training data.
It does not say that a fully general LLM or similar will necessarily be better at chess than a neural net trained against the chess rules and nothing else. There's no reason to think that any of the non-chess knowledge encoded in the LLM will be of any use in playing chess. To the contrary, the bitter lesson says that absolutely nothing besides the chess rules is needed. That's exactly why human-crafted strategies turn out not to be helpful.
An LLM on its own will never ever beat Stockfish, pound for pound. (Of course it doesn't need to, it can just use Stockfish.)
So I'd say problem-specific models will always beat general models, as long as the problem domain and scoring are completely defined, like chess and go are. It's CPUs versus ASICs. As soon as the problem is open-ended, a general model probably wins. Driving is in that category, although maybe on the borderline -- driving in very specific environments is close to fully defined. OCR is another good example, that's mostly well-defined but not completely so.
People sort of generally refer to the Bitter Lesson as if its key argument is "Large Transformer models are best", but this is directly from the essay:
"One thing that should be learned from the bitter lesson is the great power of general purpose methods, of methods that continue to scale with increased computation even as the available computation becomes very great. The two methods that seem to scale arbitrarily in this way are search and learning."
The second paragraph of the essay is specifically about chess:
"In computer chess, the methods that defeated the world champion, Kasparov, in 1997, were based on massive, deep search. At the time, this was looked upon with dismay by the majority of computer-chess researchers who had pursued methods that leveraged human understanding of the special structure of chess. When a simpler, search-based approach with special hardware and software proved vastly more effective, these human-knowledge-based chess researchers were not good losers. They said that ``brute force" search may have won this time, but it was not a general strategy, and anyway it was not how people played chess. These researchers wanted methods based on human input to win and were disappointed when they did not."
The "brute force search" he's talking about (Deep Blue) used tree search with alpha-beta pruning on top of a ton of human crafted heuristics and a database of game-specific information it could access. Stockfish has historically been an even more "brute force" approach in that it is much more focused on optimizing search--which is one of "arbitrarily scalable" techniques listed in the Bitter Lesson essay--and it has increasingly replaced handwritten human heuristics with neural networks over the life of its development.
At no point in the Bitter Lesson is there a mention of LLMs driving cars, or a suggestion that large transformer models will be the dominant class of model for every single task. It is specifically reflecting on the fact that approaches that leverage search and learning, combined with large amounts of compute, consistently prove to be most effective. Stockfish would qualify as evidence of that, under his criteria.
"AI" seems to be reserved for systems that we perceive as human-like. Generative NN that talks or makes pictures? AI. Vision + SLAM + path planning + goal seeking? Might be a robot but we don't call it AI any more.
The only important prediction Kelly made was a complete rejection of the Singularity. It is also probably the prediction that sentiment has changed on the most in the direction away from Kelly, since 2016. After a decade of evidence and updating, far more people believe in that view of our AI future now than they did then.
(His 2017 https://www.wired.com/2017/04/the-myth-of-a-superhuman-ai/ hasn't aged well either.)
Except that he most certainly did! The bitter lesson isn't about LLMs vs specialized neural networks! It's about manually programmed expert systems vs machine learning. Alpha Go Zero isn't very good at anything but playing Go. It's a highly specialized brain only for that singular task. LLMs remain a problematic technology (both in terms of reliability and privacy) for a lot of domains and my own prediction is that smaller models that are much more reliable for a specific domain will become common after the utility of AI is known by way of LLMs, but the reliability or cost of running them makes it unsustainable.
The most striking thing is that almost all of his missed predictions were missed not because the technology was not available, but because society rejected or didn’t adopt that technology
I would argue that society would likely have been glad to adopt, but the profit-seeking motive got in the way (again).
I was focused on augmented reality largely in part because it was such a big enabler of all other AI technologies
However people have almost uniformly rejected AR glasses my entire career for social reasons and that is only increasing
To the point where in 2016 I was debating with Jessica Alba and she was firm that they won’t take off cause most people don’t want to wear glasses
Transcribe conversations with my own hardware at home.
Turn by turn directions
Pomodoro timer
Read books on my lunch break
Read recipes while cooking
Instruction manuals
RSS feeds
How often do yout ravel were you woulduse turn by turn direction? You don't have a navigation system in your car?
You prefer reading through glasses on a display than book or ebook reader with eink display?
Just because some people thought that VR or AR are the big thing, doesn't mean it has any tangable relevance.
I was wearing glasses and got lasic, for me the whole AR/VR thingy was quite weird from the get go. Oh you want people constantly wearing something on their face for what benefit?
I even own a HTC Vive Pro and especially when you realize what it actually means to have Virtual Reality without the benefits of a real holodeck, its not what the movies and everyone depicted.
Every book like Neuromancer was always cool because you imagined (and it was depicted like this) like a place to be. It ignored everything inbetween. He drove a motorcycle and he was able to feel it. In a holodeck you feel things too.
I also don't get your Jessica Alba reference tbh?
I think personal taste/preference not having to wear something, not having to handle it day to day (charging, configuraing etc.) is the stronger thing than social reasons.
OK, so imagine that in 1980, I predicted that we'll colonize Mars by the year 2020. Today, I'm arguing that I was right in my prediction - in broad strokes, we have the technology! Nothing is stopping us. It's just that politicians have different priorities.
Of course, that's a load of bull. You can't decouple predictions about the future of mankind from what said mankind actually decides.
I looked up the predictions on Wikipedia and I'd say that most of them are general enough that they'd hard to disagree with back in 1999 (there will be bioengineering, there will be portable computers, there will be robots, there will be AI), but the timing or the specifics are off on almost every single one of them.
For instance, by his prediction, we're already many years past a 100 year average life expectancy (with 120 years expected by the end of the decade).
I think his mistake was tightly coupling exponential compute with exponential progress in biology - which has so far not panned out.
All the hills he suggested heading towards were the right hills even if he assumed they were higher in altitude than they were
More importantly the mountain that is the technological singularity he did a good job defining the hilly path for
AGI (strong AI in his words) is just one of the major foothills
> "By 2026, Google’s main product will not be search but AI."
Nah. Not by any common-sense measure at least. i.e. not if you ask a person on the street "what is Google's main product?", not if you look at where their revenue comes from. Perhaps if you look at where their staffing/spending/attention is going, but I don't think it's fair to say AI is their "main product" yet. (disclaimer: I work there, not directly on AI)
> "oligarchy of two or three large, general-purpose cloud-based commercial intelligences"
Directionally, yes. The number is too low, though, at least for now. As someone not even super closely following everything, I can rattle off more than 2 or 3. OpenAI, Anthropic, Google, xAI, Meta, Microsoft, DeepSeek + other Chinese players... Probably in the eyes of the public it really is just 2 or 3 (ChatGPT + something from work, perhaps).
> "The AI on the horizon looks more like Amazon Web Services"
Yeah.
> "The bulk of AI will be special purpose software brains"
Feels underspecified... I'd say this is a miss. "most of the commercial work completed by AI will be done by nonhuman-like programs" Most by what measure? Most invocations/day? Most users? Most economic output? Most public awareness? "AI" or "ML" in one form or another is ubiquitous and indeed if you were to make a giant list of "AI" or "ML" applications it would probably be dominated by hyper-specialized non-human examples, e.g. the image processing in your phone's camera, your Spotify recommendation algo, image recognition on factory floors. But in the public consciousness, the term "AI" means know-everything, do-everything human-like chat/assistant systems.
> "in conversational tones, any question at all"
Yeah.
https://news.ycombinator.com/item?id=50039614
(Douglas Hofstdter, Ray Kurzweil, Bill Joy, Hans Moravec, John Holland, Kevin Kelly, Frank Drake, Ralph Merkle, John Koza)
And no, Google’s search results are not good, in fact in most cases top results are the ones who pays Google.