The AI Contradiction
I’ve begun to notice that a lot of endlessly repeated statements about AI by the vendors of hosted AI inference and their sympathizers are strongly contradicting reality, and yet, they keep making these statements. A short list of the ones I’ve seen recently (there may be others as well):
AI will make us all very wealthy. This may well be true. In the extremely long term. But the trend is pointing exactly in the opposite direction. AI is making us all much poorer. We’re losing jobs, of jobs in the USA alone, thousands world wide, every day now to AI and it seems as if that trend is accelerating rather than decelerating. It is true that it is making the AI vendors a lot wealthier, and maybe that’s the ‘us’ they were talking about. But until I see people who lost their jobs to AI finding better ones because of AI, I’m not buying this argument at all. Don’t worry, they will say, ‘There will be UBI’. UBI stands for ‘Universal Basic Income’. So even if you don’t work, you will still get money. That, too, may well happen. Where I live, we pretty much have this already. But that does not qualify as ‘rich’ even if you have lots of time. It’s not even enough to live a normal life. You won’t be buying a house with that - ever - and it may just be enough to rent yourself the smallest dwelling you can stay alive in (you may have to pool your income with others). If you’re lucky there will be a tiny little bit left over for your food. It doesn’t begin to qualify as rich.
AI will lead to the ‘post-scarcity’ economy. In the meantime, AI has made certain resources scarcer than they ever were before. DRAM has jumped up so far in price (500%!) because of the AI service vendors buying up all of the production of RAM for the next couple of years that it is becoming quite hard to provision computers for other uses than AI (or for AI applications that are not served up by the usual suspects). GPUs, especially the more powerful ones, have increased in price. First twofold, then they kept rising. Definitely more slowly but still quite steadily. And now they are at roughly 250% of what they were before. Video cards that were made for gamers are bought by the pallet by businesses jumping on the AI bandwagon. Big AI companies have cornered the GPU market just as they did with the RAM. It may be smart. But it puts the lie to the ‘post-scarcity’ economy. To them RAM in your computer is something you shouldn’t have because you are not using it efficiently enough (Sam Altman, during an interview in October 2026). You can’t make it up, really. Yes, it is true. But you know what, I have more cutlery than I usually need too. There is probably a CPU idle in my work room right now and I may even have a spare room in this house. But it is up to me how I allocate my resources. I wouldn’t tell Sam Altman that he can’t have four seats in his car just because he’s driving it alone.
AI will cure cancer. This is such a bold statement. How could you be against a technology that will cure cancer? Here’s the problem with this: it’s possible that AI will cure cancer. We do not know if it will or not. But let’s assume just for the sake of the argument that it will. What makes you so sure that if AI actually will cure cancer that that cure will be gifted to the world? I think what is far more likely is that if AI does in fact cure cancer that it will be priced out of reach for the bulk of the population, the same way healthcare is expensive right now. For the wealthy, cancer will be cured. But what about the poor? Today, your chances of getting cancer correlate with your exposure to carcinogens and it just so happens that poverty is a very strong carcinogen. And the poor, typically, can’t afford the cures for the diseases they are dealing with, cancer or otherwise. And AI will bring poverty to large numbers of people, so unless they manage to cure cancer and to do so in a way that makes it free or so affordable that even the poor can partake of it, more people will die of cancer due to AI.
AI will make software more secure. I find this one the hardest to believe. AI is super good at finding ways to crack existing systems. But just like with people, the ability to find flaws in something that exists does not immediately translate into a more secure environment in general. There are several reasons for that. We have built our collective IT infrastructure on top of protocols and software (including programming languages) developed in the 1970s and the 1980s. Rewriting all of that stuff with or without AI is incredibly costly. So is analyzing all code (both old code and new code produced by AI) for security flaws. As well as figuring out which flagged issues are the true positives and then to find fixes for those issues. And finally, ensuring you are not accidentally introducing new flaws (or functional differences). So the presence of AI gives an incredible advantage to attackers, especially those attackers who have large amounts of financial backing. You see, security has always been a matter of budget. For an attacker, going after a particular target has to be profitable. So if they spend less than what it will take to fund the (successful!) attack, then they’re making money and will continue to do so. But if they spend more, then they are losing money and sooner or later will run out. AI has substantially reduced the budget an attacker requires to breach a system; at the same, time AI has increased the budget a defender will need to ensure attackers won’t get through. The practical effect of this is that, for the foreseeable future, you will see many more successful hacks of existing systems. This could easily take a decade or more to play out.
AI will stop climate change. But first, of course, we will have to build out our data centers, which will consume resources and energy, which is bad for the climate. And then we need to operate those data centers, which in turn adds an insane (in an absolute sense) amount of power consumption to the total, which is bad for the climate. Proponents of AI will tell you that data centers are only 1.5% of the global electricity consumption, and that not all of that consumption is for AI. But we’re conveniently ignoring a number of things here: (1) that data centers consume resources and power during construction as well and that this can be a considerable amount of the total over the life span of a data center. (2) that that 1.5% is not spread evenly. A very large fraction of the total data center construction and associated power consumption is in the ‘developed world’. In the United States, 4.9% of electricity consumption goes to data centers. In some states that share is more than 10% and in Virginia more than 25%. And demand is rising fairly rapidly, fast enough that it offsets a large fraction of the gains in renewables. So at least for the foreseeable future AI will make the climate change situation far worse. It directly leads to more consumption of hydrocarbons (some of these are fed by local generators, comparable to running a jet engine 24x7), and thus to a further increase in greenhouse gases, which could not have come at a worse time.
Overall, from a technical perspective, I’m impressed by AI, by what it can do and how far and how fast it has come. But I’m sick and tired of the whitewashing and the degree to which the short-term effects are being ignored. In the short term, AI is not going to get us to a better place, and very likely it will get us to a much worse place. What the longer term holds, nobody actually knows. It may be better, it could also be worse. The trends are - technical achievements aside - not super positive. It reminds me more of an arms race than of a technological development that will benefit us all, a land grab combined with a massive power play. It would be nice if the proponents of AI would at least recognize the contradictions between the promises and results so far.