But yeah. I expect soon this bubble will burst and that'll stop most of the AI push. It'll become venture funding poison instead of a plus, so it won't get funded, and if the major players exit the market, a lot of places using AI will simply stop, because they aren't doing any of this in-house, they're just using APIs from the big names. If those go, so does their AI

Some of the things currently under the umbrella of "AI" will keep being used, though. Local classification models for things like detecting NSFW images? That will continue to exist and continue to be a pain in everyone's backside.

moderation of internet content will not get any easier, so using these flawed models will continue being a tempting option for social media sites, yeah. Sorry.

so what I expect to disappear most is the conversational/agentic AIs. Your chat-gpts and such

image/video generation are also going to mostly die out, but will still remain in various communities that are using them locally.
I imagine the problem of AI art in games will remain, because it's similarly too tempting to just have a model generate a bunch of "placeholder" assets.

(and the weird pervert community: people will keep making porn with AI image generation, including the least moral ways to use it)

then give it another 20 years of computers getting faster, ram getting bigger, GPUs getting more parallel, and some grad student AI project will get good enough for this whole thing to happen again.

depending on how bad AI craters this time, they may not even call it AI, but it'll be some technology that's a great-great-great-grandchild of current LLMs

so yeah. that's what I expect will happen, at least tech wise. (what this bubble burst is going to do to the US or SF bay economies is above my pay grade).
Most AI will disappear, but some tech under the current AI umbrella will keep getting used in limited uses.

And to be clear, this isn't my "what I think SHOULD happen" manifesto, this is just how I expect this'll go having seen tech cycles before

set your calendars for the mid 2040s for the next AI wave

also I don't think the AI bubble bursting will cool the privacy invasion that is the modern web: facebook and advertising was already trying to collect all your info before AI blew up, they're gonna want to keep doing that after AI stops being a good excuse for it

and how much privacy still gets invaded post-AI is going to depend more on legislation than technology.

(commercially invaded, I mean. goverments are still going to want to spy on their citizens, that's just a whole separate issue from AI)

also at least one of the major companies is going to keep playing with it no matter what happens to the rest of the industry.

(my money is on microsoft. google jumps trends too fast and loves destroying their own projects)

much like the dot com bubble bursting I don't think this will damage the tech industry as a whole.

a bunch of smaller companies will go under, a bunch of startups will be DOA as their venture funding fails to appear, and big companies will post big losses but trudge on regardless.

and everyone else will just pull out their AI features as quickly as they put them in.

probably someone will make a killing picking up customer support contracts that were AI a year before, and they'll be replacing them with cheap human labor.

I think my underlying point is that AI is another tech bubble similar to previous ones, and this sort of bubble will keep happening and bursting at different scales so long as the tech industry works like it does.

It's a feature and/or bug of how its funded.

you wanna stop this happening? you gotta look at how tech is funded and run. If tech just weathers this impending bubble crash like its done every crash before, we'll just be on track for it happening again in 20 years, with numerous mini-bubbles in between

but again: that's above my paygrade, as is the economic impact of the bubble bursting.

but even I can tell you it's going to be baaaad

anyway this whole prediction is based on two things:
1. I don't think AI as a product can last. It's being sold as doing things it can't, and we can only lie to ourselves and each other for so long. I don't think this is fixable in short enough term to keep the hype up. It will collapse, probably sooner than later.
2. I know what tech industry bubble collapses look like: this will be one of those, just bigger than most

by "fixable in short enough term" i mean I don't think they can build an AI that can actually do what people want from an AI, not with this technology and not within a decade. Even in the worst case where the AI industry gets a bunch of Too-Big-Too-Fail bailouts, it can't last THAT long and I don't think enough progress is being made to get anywhere near the pseudo-AGI that they want

I'm not saying AGI is impossible, just that if it is, we're not close to it.
Current LLM "intelligence" is largely if not completely just anthropomorphization.

I don't know if any eventual AGI we build will have anything to do with how LLMs work today (my belief in turing completness and evolution basically says "anything could be run-doom/be AGI with enough cycles of improvement"), but if so, we're WAY FAR AWAY from this being a thing that works

so I don't think AI-the-technology and AI-the-dream-being-sold can be aligned this cycle. Maybe it'll be like "webvan" in the dot com bubble and a failed idea will come back successful the next cycle?

but I kinda doubt it. I'm no AI scientist but the history of the field says it's always 20 years away and will always be 20 years away

that's basically just on my personal beliefs, not science. I believe AGI is possible (We're here, aren't we?) so I think it'll be developed eventually.

But this ain't it. This is another Clever Hans.