How did AI advance in 2024?

It’s been another bumper year for artificial intelligence, with AI pioneers scooping Nobel Prize awards across different disciplines. In fact, it’s only been two years since the first public iteration of ChatGPT was launched, prompting excitement in some quarters and existential dread in others. So, how have these predictions turned out, and who are now the major players when it comes to AI? I’ve been speaking with the University of Oxford’s Mike Wooldridge.

Mike - So I think the first thing we've seen is that the debate, I think, has calmed down a little bit. People have used these tools, they've experienced them, they've got to know a little bit about what they're good at and what they're not good at. And I think, the initial, 'oh my goodness, we're on the edge of the end of AI'. That kind of feeling has sort of died down a little bit. So that actually I think is a huge relief. We've still had AI o the front pages very, very regularly, but not quite in the same slightly hysterical tones. If I look back, what else do I see? Well, one remarkable thing throughout the year has been the rise of NVIDIA. So NVIDIA's core technology is graphics processing units, GPUs. Now GPUs were invented so that teenagers could have realistic renditions of explosions in whatever computer game they were playing. That's what they're for. They're for processing graphics. But around about 2012, it was realized that exactly that technology was ideal for training neural networks, neural networks being the core technology underneath all of modern AI. And by using GPUs you got in 10 times more bang for your buck than you did using conventional CPUs. The CPUs, the central processing units that you have on a regular computer. Now, when that happened in 2012, we saw the progress in AI start to accelerate very quickly because we were getting 10 times more computer power available for the same amount of money.

Chris - I hadn't realised it's one of the world's most valuable companies, valued in the trillions. And, as it's been said in a gold rush like this, you don't sell gold, you sell shovels. And they're selling shovels.

Mike - They are selling the shovels that absolutely everybody wants. They specialise in this technology. And the reason that they are, I think at the moment, the world's highest value company has got nothing to do with teenagers playing computer games and everything to do with the fact that they sell GPU AI super computers. And if you want to build your GPT class model, you are going to need tens of thousands of those AI supercomputers. So the current valuation for NVIDIA is larger than the GDP of the United Kingdom. That's an extraordinary thought. They're larger than Amazon, larger than Google, larger than Apple. That's absolutely mind blowing.

Chris - NVIDIA are definitely the winners here because they are selling those shovels in this gold rush. But everyone else appears to be just peppering the world with AI, in what appears to be in a desperate hope that they'll hit the golden nugget of this is the killer function or the killer app or whatever. Is the bubble going to burst on this at some point? Because this is expensive, isn't it? The amount they must be investing in doing this, it must be bank balance breaking.

Mike - To be blunt, I don't think anybody really knows what's going to take off. But the hope is if you put this everywhere that you'll manage to land on the one that ends up winning. Another recent example is Apple intelligence, and this is Apple's foray into generative AI. And it's exactly this model that they're embedding large language model technology in particular across all of their products, into their email and sending text messages and their notes apps and all of that. In the hope that it takes off. There is no guarantee that it will take off. If we go back to the 1990s, I don't know if you remember Clippy, and I don't know if you remember how you felt about Clippy?

Chris - That annoying paperclip that kept popping up. 'It looks like you're trying to compose a letter. Let me help.' And it would screw up all the settings and you, it took half an hour to unpick all the damage it did. And then you wrote your letter and wished you'd never touched it. Yeah, yeah. I remember Clippy <laugh>.

Mike - So Clippy was Microsoft's 1990s attempt to embed AI in their product suite. And it didn't just fail, it attracted ridicule. It rapidly became an embarrassment and was very, very quickly dropped. But actually what we're seeing at the moment is not very different from Clippy. It's much more sophisticated technology, unimaginably more sophisticated technology. But the mode of using it as a kind of an assistant that helps you do things is very, very similar. So it may well be that people just don't pick up this technology. But yeah, what we're seeing is this just rush to put AI, generative AI everywhere in the hope that companies land on the killer app. And then how do they monetise it? How do they recoup these incredible returns on investment?