HiTech with Jan Coomans: Why the AI hype is still dumb, part 2

Enormous sums are being talked about for developing insanely large data centres which, even individually, will consume as much energy as some smaller countries, but not that much money is actually changing hands. And even less money is being made actually selling AI services. The largest player by far in providing those is OpenAI, which has 13 billion dollars in revenue. That sounds like a big number but it is far exceeded by the costs of running the business not to mentioned planned spending numbers like a recently announced contract with Oracle for 300 billion to provide data centre services. As part of that gargantuan contract, Oracle will buy the necessary hardware from Nvidia….. and Nvidia is investing billions into OpenAI to pay for it. Remember the circular economy part I just talked about. Nvidia, incidentally, is now the biggest company by market cap in the world and has seen its stock price explode based on the apparently insatiable demand for its computer chips. A sceptic might ask just how insatiable the real demand is when the company is investing hundreds of billions in other companies on the basis that they will then use that money to buy its products.

But the main issue is this: these investments will only ever be worth it if the AI market starts to generate trillions in revenue. It’s a pretty big if, at the moment. There is no question that there is a genuinely booming AI economy, but there seems to be a real risk that the almost unimaginably large capacities being announced will outstrip organic demand by a long way.

With each investment announcement made, the tech companies involved see their stock go up. It’s not unimaginable that “pumping the stock” has become an end goal in itself rather than an effect of running a successful growing business. The only thing that gives me pause about declaring the whole thing a house of cards is that half the whole internet is already full of “experts” talking about the AI bubble. The rule of thumb is that massive stock bubbles only tend to be visible in retrospect, after they’ve popped. Not before. And as John Maynard Keynes remarked a long time ago, the stock market can remain irrational for longer than you can stay solvent.

Back in the real world though, I think most of us have already seen some of the good as well as the ugly sides of AI use. It’s great that you can do stuff like translate texts into foreign languages, improve pictures you’ve taken, and there are realistic promises of speeding up vital things like medical research by orders of magnitude in some cases. Not because AI is actually intelligent, but because we can just process so much data so quickly with these systems.

The flip side of the coin so far is that we’ve seen the internet becoming overrun with what has been coined “AI slop”. Useless garbage generated in large amounts very cheaply and being spammed all over the internet and social media. Videos, images, articles, you name it. You don’t need a supercomputer these days to keep generating huge heaps of “content” which has little or no rhyme or reason to it but accumulates enough clicks to just make it worth doing. It’s already making it hard to find “real” content in some areas, and the problem will only get worse from here unless barriers are put in place.

The internet started as a place where you could get information quickly and easily which would otherwise have been hard to find. Back in the old days, and I’m just old enough to remember those, if one had a brainwave to check what the capital of Botswana was, I promise it would have taken you longer than it takes now with a smartphone on a 5G connection. Pretty much any information you could think of is on the internet and all you have to do was type some key words into your favourite search engine. Right now, the first thing that search engine does is give you an AI result which may or may not be accurate or even what you were searching for, but it already sums up whatever it found on a particular website it thinks has the answer, thereby taking away your need to click on that website. So you may have gotten the/an answer but whoever went through the trouble of actually providing it got nothing for it. Which means there is less incentive for actual people to put actual real content on the internet.

We’re only half a step away from the AI search result quoting an AI-generated website which is reposting AI-generated summations of the blended-together works of both actual experts and a bunch of idiots on Reddit. Whichever of those groups managed to put the most words on the internet about it probably came out ahead in the result. I’ve already seen several examples online where people were having a debate or argument and then at some point one of them pulls up what ChatGPT’s opinion is. As if it is the final arbiter of truth. That kind of confidence in AI probably sounds scary enough already, even if you only have a vague idea of how large language models come up with their answers.

But then there’s the fake images and sounds which only take seconds to generate and can be impossible to tell apart from real. People in general already have a tendency to believe in data that fits their preconceived notions, so what could go wrong if we give everyone evidence that all of those contradictory positions and beliefs are true? Clearly, that’s too big of a question for me to answer. But my gut feeling would be that not much good will come of it, definitely not enough to justify the hype anyway. Whatever is branded as AI now isn’t intelligent and there’s a good chance it will never be, but it may well be the final excuse for humanity to stop thinking altogether and all our brains choosing the reality where they are right about everything without the need to actually learn anything.