Achievements of Artificial Intelligence (AI)

We’ve discussed the idea that AI will lead to a crisis for the discipline and careers in mathematics. That the value comes from understanding not from the solutions. What happens when these longstanding and interesting problems are all solved?

But could there be a Jevons paradox of sorts. We will need more mathematicians to work on understanding and interpreting as well as communicating what we find. And to come up with new and interesting problems. Because those are things AI may well not be able to do.

I suppose it comes back to how we choose to value mathematics and how we choose to assign resources. As with so many other disciplines.

There’s some interesting ideas in the blog below. I think it comes back not to AI or mathematics but to how we choose and assign value and what we prioritise. These seem more arguments about societal structure and economics. Just like with the industrial revolution we may be in for some upheaval here…

terrytao.wordpress.com

A value-based approach to mathematics

[This is a guest post by Ivan Corwin. This blog post was initially written in a different file format and converted using AI. — T.] Mathematics and those mathematicians who guide its developm…
terrytao.wordpress.com terrytao.wordpress.com
 
Ultimately in Asimov’s writing this wasn’t solved until sometime after resource allocation was taken over by the robots… Humans have too many problems of self interest and so on.

It may not be the solution. Maybe there isn’t one. But perhaps on the path to wherever we’re going we will need more mathematicians, more economists, more sociologists, more philosophers…
 
There is a very serious danger that the AI investment mania is going to deliver a huge crash to the world economy, and soon.

From what I can tell, those selling the dream and those investing huge amounts in it have set the whole show up so they are too big to be allowed to fail, and the taxpayers are going to be on the hook for the bill on a scale that is unprecedented.

And we all know which strata of society will get hurt the hardest to pay that bill. Hint, it won't be the rich who saddled us with it.
 
If things pop they’ll be a hit and a panic in some markets but this isn’t as significant in the real economy I don’t think.

Stocks may crash but the debt is localised. GDP has been made to look good in some places by capital expenditure. But there is no need for government or central banks to step in. They may if stupid but I can only really think of one or two that would do so. There could be a risk of it showing up who has made terrible bets and expose those with unsustainable business models or debt elsewhere but we’ll see. We do have plenty of other real economic problems and it would be great if there was more focus there but that’s distinct from and predates AI bubbles IMHO.

I don’t doubt companies like Google, Microsoft, Apple and sadly Meta will be fine, they have revenue and a business. I can’t see how OpenAI survive, previously I thought their deal with Microsoft meant they’d gain their IP now I’m not so sure. I once thought Anthropic had the enterprise market sewn up so would be okay. But despite some clever accounting to be able to report a profitable quarter they’re losing tens of billions a year. Enterprise customers are turning to running their own models while Anthropic join OpenAI in making wild valuations and claims to keep afloat.

People here must know I’m a user of the technology and think there’s great potential. I’m not anti AI. But the business side sucks and the economics don’t add up. I sometimes fear the idea of a few powerful rich people trying to hoard ‘knowledge work as a commodity’ but am not sure if they’ll be able to. They have no moats just vibes and debt.

So a positive take on this is I am looking forward to us maybe being left with the benefits of the advances which we can then go about using appropriately. Local inference may still be the present or near future as it’s not far behind the frontier. And at least the wider world has now woken up to advances coming through ML/AI so we can maybe talk about and prepare and manage things better as societies.
 
The problem, as I understand it, is the simple scale of money spent on it so far, including the exposure in investment funds, and the eye-watering amounts of return on that investment that AI is going to have deliver before long to justify it all, without showing any signs of doing so yet.

This is completely unsustainable. Which is why the tech AI CEOs are getting so cosy with the US government, to make sure they don't foot the bill for it all, the rest of us do.
 
Here's an article from Sept. 22 by two experts in AI (Timnit Gebruarchive and Emily M. Bender) that I found helpful.

And it's fairly easy to read.

"Don’t be fooled by this summer of AI hype"

www.technologyreview.com

Don’t be fooled by this summer of AI hype

Breathless claims about AGI and new capabilities fall apart pretty quickly under scrutiny.
www.technologyreview.com www.technologyreview.com
“The idea that these companies are building software that could ‘kill us all’ distracts from the real harms that are happening now,” said Emily Bender ...

People have misunderstood what happened with the Hugging Face incident, which seems to have really set off the latest round of alarmism about AI, Williams and the other critics said. The AI models didn’t go rogue or escape containment, Williams said.

Instead, the software that was trained to hack into systems and test firewalls did just that, he said — and the humans who were supposed to be keeping an eye on it didn’t watch it closely enough.

“It’s less a case of these models being so powerful that they can escape containment,” Williams said. “It is more the case that there’s a suite of tools that we’ve managed to build as humans that do what we built them to do,” he said.

Ultimately all the doomer talk distracts from addressing the real problems that AI is causing or exacerbating, he and the other critics said. The data centers used to train and run the models are increasingly contributing to climate change because they require vast amounts of energy that in many cases is coming from fossil-fuel plants, they noted.

The models were trained on creative works from writers, artists and others who frequently weren’t compensated for that. And companies have been increasingly using the threat of AI as an excuse to lay off workers en masse.
Click to expand...

PS. Folks with ME/CFS should be pretty familiar with the concept of hype not matching reality. The hype after the PACE trial was that exercise could cure ME/CFS. The reality was that the research study was deeply flawed.

Lots of people believed the PACE hype. And then the media fanned the flames rather than doing investigative journalism to see what was behind the claims.
 
It’s a good article thanks for posting @ahimsa

“It’s less a case of these models being so powerful that they can escape containment,” Williams said. “It is more the case that there’s a suite of tools that we’ve managed to build as humans that do what we built them to do,”
Click to expand...
I was thinking more about writing this because… it’s farcical the way its being presented at times. Hearing the leaders of the companies involved talk about how they’re now realising they need to bring their safety and other testing processes in earlier in their development cycle seems to be a story I heard over years of working in software development. You can’t leave this stuff until the end. If it’s an afterthought you will hit problems. It’s an age old tension between ‘new shiny feature’ and ‘making sure what we have works’. It’s risk vs perceived reward and all about pace. The incentives are broken if this sort of thing is seen as okay and trying to explain it away by ‘well this is a new thing’ is a trick some tech CEOs have been trying to pull for years to avoid scrutiny, regulation and… liability.

I wonder if liability will be what comes back to bite them in the end. It may take a while but the lawyers are involved snd there’s a lot of money at stake. This was an interesting story in the FT (free link)
https://www.ft.com/content/a5caf8d4...egmentId=7d4bcc2e-e664-92ba-62e3-5590579f1902

Maybe we need a thread on the economics of AI, one on ethics and one on achievements. We always seem to blur the lines with different people having different views whenever the topic comes up. Hopefully we can get all aspects across. Just this week I used these tools to achieve things I couldn’t have done otherwise (building some backup scripts and processes for upgrading some home automation tools). And being able to achieve things is important for those of us trapped and able to do so little, not only in the value of the outcome but the sense of achievement itself. So it’s not just big scientific achievements that matter.
 
There’s also been an interesting interview Sam Altman has just done with Politico which included discussion of risk and liability. Also some politics but I’ll sidestep that. He’s understandably trying to shape policy debate and is in favour of a more light touch approach than Anthropic saying when asked on the difference between them and OpenAI

“We believe that the world should accept some bad things happening for the benefits of this technology and people having the agency” said Altman, who characterized those harms as linked to “the lighter touch regulatory stance that we advocate for.”

“if something goes wrong with our models during training, there’s going to be some version of that we need to be responsible for.”

“This is the kind of thing that we are hoping our policymakers will all debate”

“There's gonna need to be a liability framework for companies.”

“We have more things to disclose, and because some of them involve security vulnerabilities, we give people a long time to address those, and it’s kind of often up to them to disclose it or not.”

Find the full interview on YouTube or listen to the podcast here.
 
You must log in or register to reply here.
Back
Top Bottom