"AI reminds me of the internet in 1999." -- Peter Thiel[1]
Smartphones followed a similar path to the internet i.e. completely new paradigm, that had immediate benefit but needed technology to improve to become widely adopted. But where the roadmaps to improve these technologies were clear and well defined.
AI is far more akin to nuclear fusion where each step along the way will require major scientific breakthroughs. And where it's not clear how to get this grand AGI end-game. Especially as OpenAI has said advancements depend on compute capacity that we simply don't have.
In 2020 I never thought an AI would code for me. And now it turns out, it's pretty good. But AI Slop is pretty much a toxic mess.
The thing is that it will take years to integrate them into existing domains and workflows. Honestly I think the most relevant comparison is the rise of desktop computers themselves. Suddenly paper-based processes had to become electronic and in some cases that took 40 years.
"Took", past tense?
You compare this to the internet or smartphones and they have been transformative across every aspect of society.
And as someone who works for a bank which is heavily exploring LLMs there is no complexity in integrating them into existing workflows. The issue is that (a) risk of privacy/security being compromised through prompt exploits and (b) risk of reputational damage if the prompt is biased or hallucinates. Issues that may well be inherent to all transformer based models.
The ability to type an email for instance, and send it instantly to someone on the other side of the planet for free was something that was in such contrast to the tools of the time it can’t really be understood by someone who didn’t live through that period.
AI to me is really hyped up by some very highly regarded CEOs with strong track records in other domains and tech enthusiasts who seem hell bent on being able to look back and say they called the next Industrial Revolution. In short everyone thinks they’re the counter to Paul Krugman saying the internet would be as useful as the fax machine. Credulity levels are off the charts. It’s gotten to the point where skeptics are automatically assumed to be wrong.
But what’s missing is the obvious amazement that should come to an ordinary person and frankly I still don’t see most people naturally gravitating to these tools.
Perhaps that could be explained by how amazingly fast technology has advanced in recent years but then that in and of itself seems to call into question whether this technology that’s being called AI is truly revolutionary when compared to what’s already available.
I barely know anyone who doesn't use ChatGPT frequently, to help wording an email or such. I agree though that in instances like this it is not transformative to society and rather one more tool that we use. We will see, IMO the impact of the current AI technology on the world is rather "medium".
ML also didnt replace manual decision making. A lot of previously automated decision making was done with what were basically encoded rules of thumb which didnt overfit much worse than "more advanced" ML models did.
ML is used everywhere.
How many ML projects for large businesses have you been on?
The world hasn’t even fully realized the productivity value of spreadsheets and email.
Also, I am old enough to remember - and have experienced in my professional career - a time when the mainstream opinion was that the internet was a tiny niche and that things like online shopping or dating would never gain widespread adoption.
It's been 24 months, where is it? And even what you can measure doesn't match the valuation of openai at all
We have AI now. We've had it for ages. We just stop using that term for things once we've gotten them working.
We also already have artificial entities that are smarter and more capable than a lone human, and have for a long time. Bureaucracy and writing are incredibly powerful technologies, especially when combined.
It just feels like we're still in the pre-Navigator days though, circa '94. All of this complex orchestration work needs to be baked into a single model. I don't want to learn how to implement a multimodal agent system. I want to tell something what to do and have it do it for me perfectly with no more interaction than a simple prompt. Fortunately this is just a software engineering problem now, not a CS one.
So are we expecting a ai bubble burst as well?