Actually, let me make another comment. The limitations of the first wave of AI that powered the first AI-fever period (1960's) was already well known by the second half of the 1960's (e.g. Dreyfus' critique, which turned to be spot-on). It took another eight years or so for the lack of the expected results to create enough doubt to overwhelm the broadly held convictions that AI would be a success. It might very well be that it is too early to expect enough doubt in society to overwhelm the convictions this time, especially because the failures are mostly hidden from view for most humans and when they are not, they are not strong enough to change convictions because the actual results hack directly into one of our standard (quick & dirty) assessments of intelligence (good language, regardless of good content).
So, I personally expect that an AI-winter this time will not be as quick, nor as devastating as earlier ones, if only because even poor Generative AI creates enough 'cheap' results that it is actually usable. But that's just my estimate, of course.
Yes, and I’d add that the “I want to believe” impulse is very strong. There is too much information and people want this to be a solution to finding the needles, so they are blinding themselves to the ways it grows the haystack.
I've been paying attention to this in my speaking/writing (e.g. yesterday in the closing keynote of EABPM 24 in London). Basically, humans are automated to the hilt. Our convictions steer our observations and reasonings far more than the reverse (we believe the reverse, generally). This holds true for all humans, from the believers in a flat earth to us. I think this is the case because of speed and efficiency demands for our organism. So, it's an evolutionary necessity. How are convictions (mental automation) come to be is an interesting subject, but they come with the same advantage/disadvantage as any automation: you win speed and power/efficiency, but you lose agility. suspect a certain level of 'total doubt' is required to flip this in society.
But the amount of capital poured into "AI" this time is way much bigger than the previous times and it would respectively require much bigger financial success than previous times, financial success that is nowhere seen on the horizon. So far nearly all success "AI" stories have been driven by investment or acquisitions, not real market success. So, I think the realization of losing a couple hundred billion dollars will have a rather chilling effect on VCs and future investments in "AI". And because the current crop of "AI" is also very capital intensive, that would effectively kill any new development in the field for some time until a new, more capital efficient paradigm comes along and spurs new hope.
Yes, the tech world is much bigger now. But do not underestimate what Japan (5th generation program), the EU (Eurotra), the US (DARPA and others) invested in AI in e.g. the 1980s-1990s leading to a lot of private capital as well. E.g. Microsoft spent huge amounts as well. But you are right, the amount of value at risk now is much higher. And, as bleak as the previous waves were in terms of realistic usability, it is now less. Not that we will get AGI (no chance). But we will probably see a lot of 'cheap' use. See https://www.bloodinthemachine.com/p/understanding-the-real-threat-generative
certain militaries are already using some form of "AI" -- of course, the militaries using this technology don't really need the thing to get the targeting right.
One has to sort out the normal ebb-and-flow of business cycles and newness-interest, from the reaction-to-limitations and failures-to-meet-expectations-relative-to-hype, to get some clarity on what’s happening. (And I’m not a business guru, just a writer-creative-philosopher experienced in the computer field); here the limitations are seen in or as the “Dawning of Reality” takes effect as some have noted, as more and more as people see what the thing actually does in specific cases, find the good and bad, and hear and see the critiques of the product, companies and players (partly thanks to writers like Gary) .
What’s unique in this scenario, besides the weird set of flawed characters and psychodrama happening vis-a-vis OpenAI etc., is that it's also a Failure of Fundament: this whole thing is exposing, in an applied way in real time (and potentially embarrassing to the companies), just how little we know about intelligence, the mind, consciousness, and how human beings and life actually operate, and what we are, or even what Reality is… Or how to be and act in a “good” way – ethical, happy, productive, of service to each other and in general … as human beings living together on a planet. In that, it is also a great opportunity… if we don’t screw it up. :)
More like a shakeup, followed by renewed vigor. This time AI has already been out in the world for a decade, with commercial success. A far cry from "expert systems".
$50 billion investment, $3 billion revenues. GenAI has not (yet) been a commercial success. Investors see this, and are starting to be more careful with their money.
It look Amazon 9 years to be profitable. Tech takes time to mature.
The current climate is more like the dot-com boom, I believe. Weaker players will be wiped. Likely not OpenAI, but probably Perlexity, etc. The firms who will find the sweet spot will survive. Surely the behemoths will, and likely will get bigger.
It is also not clear that things will stop at generative AI. I think Google and the rest are very busy seeing how to leverage the latest most powerful hardware and access to data. They likely have a lot more ideas than OpenAI.
"Fable Studio, an Emmy-winning San Francisco startup, on Thursday announced Showrunner, a platform the company says can write, voice and animate episodes of shows it carries. Under the initial release, users will be able to watch AI-generated series and create their own content — complete with the ability to control dialogue, characters and shot types, among other controls."
So many core problems with this, I don't even know where to begin. It feels for all the world like an attempt to simply cash in on the perceived GenAI gold-rush, whereupon the founders probably hope to flip it and make a fast buck.
Exciting! The winter is where a founders crank away in a garage and surprise us all with something totally new. Imagine some genius 15 year old who's read all the papers, runs models on a second-hand GPUs, and the only way they can keep the room cool enough for all the hardware is by trying out a kludge and POW... sentience. Winter in summer, too perfect!
If you know any kids who need GPUs please smash that wishlist or share it where others of us can.
The most fascinating bit of the AI story is how it can become the substrate for a new layer of informational life which is related to recent talk around effective accelerationism and the questions it raised about what activities we should prioritize on our planet. I have a chapter on platforms in my book draft and wonder if I should edit it and post it alone.
As someone who actually built an Expert System in the mid-80s, I know about the hype of AI. My system, once understood, became a normal part of the processing workflow.
But I think that Gen AI is different and has not has a chance to make its way into the fabric of our society. Let’s sere what Apple does next week to bring ChatGPT into Siri. That will most certainly have an impact on our way of interacting with AI, seamlessly! The AI companies may be transformed, and only a few may survive, but I think this is just the beginning. I am not putting on my Winter clothes just yet.
The old school "programmable" approach to AI resulted in systems that were fragile and hard to maintain. Expert systems were a good example. No one really knew all the relevant rules and the computers of that time were too slow even if they did have complete knowledge. Some systems did make it into use (like yours!) but ES's never took over as a widely used approach. This new generation of systems that use big data, deep learning and generative techniques already has impressive linguistic and image processing capabilities but when it is pushed to give the illusion of real understanding it hallucinates and goes into inappropriate conversations. But overall, like you, I see more success than failure so I'm keeping my shorts on!
Even Swisher is predicting some major company (customer side) is going to say genAI is too expensive for the minimal value it brings and announce they are reducing investment.
And the fact there is a Known/Unknown of the total amount of judgements yet to be rewarded for massive Intellectual Property infringement and theft needs to be taken into account.
It is very expected that a rebalancing will happen, sooner or later.
It is also true however that we are in the most productive era in AI's history. Methods that leverage very large amount of data and compute will crack the problem. Though likely the current methods are too basic.
Actually, let me make another comment. The limitations of the first wave of AI that powered the first AI-fever period (1960's) was already well known by the second half of the 1960's (e.g. Dreyfus' critique, which turned to be spot-on). It took another eight years or so for the lack of the expected results to create enough doubt to overwhelm the broadly held convictions that AI would be a success. It might very well be that it is too early to expect enough doubt in society to overwhelm the convictions this time, especially because the failures are mostly hidden from view for most humans and when they are not, they are not strong enough to change convictions because the actual results hack directly into one of our standard (quick & dirty) assessments of intelligence (good language, regardless of good content).
So, I personally expect that an AI-winter this time will not be as quick, nor as devastating as earlier ones, if only because even poor Generative AI creates enough 'cheap' results that it is actually usable. But that's just my estimate, of course.
Yes, and I’d add that the “I want to believe” impulse is very strong. There is too much information and people want this to be a solution to finding the needles, so they are blinding themselves to the ways it grows the haystack.
I've been paying attention to this in my speaking/writing (e.g. yesterday in the closing keynote of EABPM 24 in London). Basically, humans are automated to the hilt. Our convictions steer our observations and reasonings far more than the reverse (we believe the reverse, generally). This holds true for all humans, from the believers in a flat earth to us. I think this is the case because of speed and efficiency demands for our organism. So, it's an evolutionary necessity. How are convictions (mental automation) come to be is an interesting subject, but they come with the same advantage/disadvantage as any automation: you win speed and power/efficiency, but you lose agility. suspect a certain level of 'total doubt' is required to flip this in society.
But the amount of capital poured into "AI" this time is way much bigger than the previous times and it would respectively require much bigger financial success than previous times, financial success that is nowhere seen on the horizon. So far nearly all success "AI" stories have been driven by investment or acquisitions, not real market success. So, I think the realization of losing a couple hundred billion dollars will have a rather chilling effect on VCs and future investments in "AI". And because the current crop of "AI" is also very capital intensive, that would effectively kill any new development in the field for some time until a new, more capital efficient paradigm comes along and spurs new hope.
Yes, the tech world is much bigger now. But do not underestimate what Japan (5th generation program), the EU (Eurotra), the US (DARPA and others) invested in AI in e.g. the 1980s-1990s leading to a lot of private capital as well. E.g. Microsoft spent huge amounts as well. But you are right, the amount of value at risk now is much higher. And, as bleak as the previous waves were in terms of realistic usability, it is now less. Not that we will get AGI (no chance). But we will probably see a lot of 'cheap' use. See https://www.bloodinthemachine.com/p/understanding-the-real-threat-generative
certain militaries are already using some form of "AI" -- of course, the militaries using this technology don't really need the thing to get the targeting right.
Could be. Could be saturation (and not a drop in levels of conviction), maybe?
One has to sort out the normal ebb-and-flow of business cycles and newness-interest, from the reaction-to-limitations and failures-to-meet-expectations-relative-to-hype, to get some clarity on what’s happening. (And I’m not a business guru, just a writer-creative-philosopher experienced in the computer field); here the limitations are seen in or as the “Dawning of Reality” takes effect as some have noted, as more and more as people see what the thing actually does in specific cases, find the good and bad, and hear and see the critiques of the product, companies and players (partly thanks to writers like Gary) .
What’s unique in this scenario, besides the weird set of flawed characters and psychodrama happening vis-a-vis OpenAI etc., is that it's also a Failure of Fundament: this whole thing is exposing, in an applied way in real time (and potentially embarrassing to the companies), just how little we know about intelligence, the mind, consciousness, and how human beings and life actually operate, and what we are, or even what Reality is… Or how to be and act in a “good” way – ethical, happy, productive, of service to each other and in general … as human beings living together on a planet. In that, it is also a great opportunity… if we don’t screw it up. :)
Winter is coming.
More like a shakeup, followed by renewed vigor. This time AI has already been out in the world for a decade, with commercial success. A far cry from "expert systems".
$50 billion investment, $3 billion revenues. GenAI has not (yet) been a commercial success. Investors see this, and are starting to be more careful with their money.
It look Amazon 9 years to be profitable. Tech takes time to mature.
The current climate is more like the dot-com boom, I believe. Weaker players will be wiped. Likely not OpenAI, but probably Perlexity, etc. The firms who will find the sweet spot will survive. Surely the behemoths will, and likely will get bigger.
It is also not clear that things will stop at generative AI. I think Google and the rest are very busy seeing how to leverage the latest most powerful hardware and access to data. They likely have a lot more ideas than OpenAI.
And now everyone is losing their minds over this latest "Fable" platform...
https://www.hollywoodreporter.com/business/business-news/fables-streamer-ai-generated-content-showcases-hollywood-fear-around-tech-1235911861/
"Fable Studio, an Emmy-winning San Francisco startup, on Thursday announced Showrunner, a platform the company says can write, voice and animate episodes of shows it carries. Under the initial release, users will be able to watch AI-generated series and create their own content — complete with the ability to control dialogue, characters and shot types, among other controls."
So many core problems with this, I don't even know where to begin. It feels for all the world like an attempt to simply cash in on the perceived GenAI gold-rush, whereupon the founders probably hope to flip it and make a fast buck.
Exciting! The winter is where a founders crank away in a garage and surprise us all with something totally new. Imagine some genius 15 year old who's read all the papers, runs models on a second-hand GPUs, and the only way they can keep the room cool enough for all the hardware is by trying out a kludge and POW... sentience. Winter in summer, too perfect!
If you know any kids who need GPUs please smash that wishlist or share it where others of us can.
The most fascinating bit of the AI story is how it can become the substrate for a new layer of informational life which is related to recent talk around effective accelerationism and the questions it raised about what activities we should prioritize on our planet. I have a chapter on platforms in my book draft and wonder if I should edit it and post it alone.
Thanks for sharing my work Gary!
As someone who actually built an Expert System in the mid-80s, I know about the hype of AI. My system, once understood, became a normal part of the processing workflow.
But I think that Gen AI is different and has not has a chance to make its way into the fabric of our society. Let’s sere what Apple does next week to bring ChatGPT into Siri. That will most certainly have an impact on our way of interacting with AI, seamlessly! The AI companies may be transformed, and only a few may survive, but I think this is just the beginning. I am not putting on my Winter clothes just yet.
The old school "programmable" approach to AI resulted in systems that were fragile and hard to maintain. Expert systems were a good example. No one really knew all the relevant rules and the computers of that time were too slow even if they did have complete knowledge. Some systems did make it into use (like yours!) but ES's never took over as a widely used approach. This new generation of systems that use big data, deep learning and generative techniques already has impressive linguistic and image processing capabilities but when it is pushed to give the illusion of real understanding it hallucinates and goes into inappropriate conversations. But overall, like you, I see more success than failure so I'm keeping my shorts on!
Even Swisher is predicting some major company (customer side) is going to say genAI is too expensive for the minimal value it brings and announce they are reducing investment.
This is a normal part of the cycle of any new technology and in no way shows that interest in that technology is waning in the least.
And the fact there is a Known/Unknown of the total amount of judgements yet to be rewarded for massive Intellectual Property infringement and theft needs to be taken into account.
It is very expected that a rebalancing will happen, sooner or later.
It is also true however that we are in the most productive era in AI's history. Methods that leverage very large amount of data and compute will crack the problem. Though likely the current methods are too basic.
A little bit of reality seems to be surfacing through all the hype. What a relief.
But, when will this reality appear in the business world and a cessation of pointless PoCs?