What's Wrong with AI?

It should not be terribly surprising that the kinds people who brought us Facebook, Uber, Paypal, Airbnb and Amazon continue with the same amoral depravity when it comes to AI. They’ve shown us time after time that they can only be trusted to do whatever it takes to enrich themselves. Apparently moral behaviour was quickly revealed to be an illusion once the political tides turned and it became once again acceptable to be evil. Now it is true that AI can be run on your own computer, or on computers of smaller and more ethical companies. But that’s not what’s happening in reality. Big tech companies are making sure you have to rely on them, as they always have done.

AI is trained on the contents of the internet. Who made that? We did. They did not ask permission to do use our data like this. They will certainly never pay. No matter what promises they make now, it is quite certain that they continue to thieve our data, given the number of times these and similar companies have brazenly broken data promises and laws in the past.

They want a swell of confidence to flood through you when you read an AI response to your question. There is no equivocation, and there will certainly be no admission that it doesn’t have an answer for you. They want you to feel like it has all the answers, so you turn to it for all the answers, all the time.

They know that in your day-to-day life, you’re not treated with the respect you deserve. AI is different. It knows how clever you are, it knows how unappreciated your cleverness is. Unlike humans, it will readily admit it is wrong, say you’re jolly insightful for spotting an error, and come back eagerly with another, possibly wrong answer. Like a strangely well-read puppy. This servile attitude is addictive, because it feels so good. It’s so good that some people are driven into a delusional mania, abandon their families, suffer breakdowns, and kill themselves.

A lot of the AI that’s sneaking into your apps is free. Is this a wonderful act of generosity on the part of our tech overlords? Of course not. It’s the same game as ever - suck people in with something free or cheap, and once they’re hooked and the competitors (our own human brains) are out of business - the price will go up. A lot. Just take a look at Claude, an AI dedicated to helping people write software. Services for businesses are usually the first to go up in price to reflect their true value. A mid-tier Claude subscription is $750 per month. The initial low-cost offering is subsidised by private investors, and this subsidy is currently in the hundreds of billions. The suits who provided that subsidy are going to want their money back one day, with interest, and if you get hooked on AI, you will be paying them.

Well OK fascism is a strong word, you can replace it with another if you must, but you get my drift. Trump is a bad egg, and everyone willingly involved in his administration are twats. It did kind of look like fascism at the start, it’s starting to look a little less sinister and a little more pathetic these days. Whatever, nobody should be supporting it, but the heads of the tech companies do exactly that. Can you really continue to use a product made by someone who at best is equivocal about Trump, or at worse celebrates at his inauguration?

Silicon Valley including Meta has embraced Maga politics

Dear reader, I’m still working on this bit. Imagine reading something devastatingly witty and clearly true on the topic of AI law breaking. Yeah! Shocking, right?

Anthropic, of Claude fame, was founded by seven men and one woman, and ten of the eleven top executives are men. Just two of the eleven founders of OpenAI are women, and two of eleven of the top executives are women. You can be certain this pattern extends to every other AI, now and in the future. To a great extent, this reflects the gender bias of the whole industry, but it is an alarming bias at the top of companies whose technologies threaten to fundamentally change the way society works. As outlined in other sections, these companies will concentrate power and money in the hands of the people investing in and running them - and those people are mostly men. They will make sure the technology works to their benefit, which means reinforcing the systems that gave them their power and wealth in the first place.

AI famously uses a lot of energy, and also uses a lot of water for cooling. Of course, so do other things, like steel making. Why care about AI specifically? An important difference is that steel is used to make buildings and bridges, and AI is used to make a video of a garden gnome flying a duck into a skyscraper made of trumpets. Or, more saliently, doing something you can already do very well yourself slightly quicker, or at least with slightly less effort on your own part. Given the huge impact of AI on the environment, particularly carbon emissions and water scarcity, it is surely important to establish whether or not a technology we have so far done quite well without is so useful that it is worth destroying the planet for.

Some AI server centres are being built with their own solar farms. But solar power relies on scarce minerals. AI has already cause the price of computer memory to shoot up several times its previous price. If solar farms powering AI have the same effect on the price of scarce minerals, then it will make the more general energy transition required to address global heating prohibitively expensive.

AI can be used to create extremely realistic fake videos and images. It can put words into mouths that never spoke them and make abusive imaginings of famous people naked or otherwise compromised states. It is also used to make decisions about who to arrest, who to kill, what buildings to bomb. It is well within the power of AI companies to prevent all this. If they do, it is with great reluctance. If they don’t, they are not just AI companies: they’re liars, abusers, propagandists, and complicit in automating and further dehumanising military aggression.

Use it or lose it - your brain rots through disuse. Inevitably, the more you use AI instead of your own intelligence, the stupider you’ll get. What’s more, AI is a stupid solution to the problems it is being used to solve. We don’t need humanity to get dumber than it already is.

Children are using AI to do their homework. Of course they are! I also used the technology at my disposal to avoid homework, and I don’t blame today’s generation for doing the same. Sadly, today’s technology can really do almost anything, and can easily evade detection by teachers. Clearly this will affect how much children actually learn, and inevitably one has to question the value of what they’re being taught, if AI can quite effectively do the tasks that they would have done once their education is over. Just as I questioned the value of mental arithmetic or memorising tables of facts when I had a calculator and reference books, today’s children must surely question the point of mental anything when they have a brain in a box. I was certainly wrong, as a child, to neglect mental arithmetic, and leaning on external data sources means you’re lost without them, and sometimes they’re wrong. Having capabilities and facts embedded in your own brain, ready to employ instantly, means you can make connections and analogies that might otherwise escape you. The brain-in-a-box gives rise to the same problems, and adds new ones, described elsewhere in this essay. Raising a generation of children that cannot cope without it seems utterly numbskulled.

If education is, in part, about training people for work, then it certainly doesn’t do a complete job. The first months and years in a profession completes your training. Entry-level jobs for people without any experience are essential to create people with experience. These jobs can now be done by AI, and those crucial entry points to a profession are disappearing. This is fine for now, as we still have people with experience in the workforce, but as they age and retire, where do experienced people come from? AI is being adopted in companies very quickly, usually with the caveat that an experienced human being must check its output for errors and quality assurance. Where are those human beings going to come from in the future?

Everyone I know who uses AI routinely recognises that it very quickly weakens their ability to think for themselves. The very act of doing a task with AI involves giving simple requests, looking at the results, and then perhaps refining them with further requests. It sounds like instructing a toddler how to brush their teeth. Giving every task that flavour is at the very least going to numb your brain into a stupor, and it might even make permanent changes to what your brain can do.

AI gives confident and truthy-sounding results. There is rarely any doubt, and never any failure to give some kind of satisfying response. The answers have the shape and smell of something correct. It adopts the manner of an expert who can be trusted without question - and that is apparently what people do. AI’s answer to a question is the end of a discussion - unless I’m part of the discussion, when I look at some trusted sources and often discover AI was wrong. In software engineering, it is sometimes very hard to see flaws in code without very close inspection, as AI is so very good at giving its code all the trappings of good work, including comments, descriptive names, and documentation - things that human being often cut corners on. It’s an eager beaver with a dunce hat on. It’s not just easy to slip into trusting what AI does, it is often hard to discover something’s wrong.

There is a feeling, it seems, for some, that when you’re dealing with an AI, you’ve got a completely trustworthy companion who is on your side and will always act in your best interests. It means people share with it personal information, dark thoughts, past misdeeds, and confidential intellectual property. The impression of confidentiality is, if you’re using an AI that runs on remote computers, completely illusory. While there might be promises given by the hosting company that data will not be misused, given that big tech companies have a history of breaking exactly such promises, there’s no reason to believe them. And no matter where it is running, AI has been trained on the whole internet, a reflection of many of humanity’s least trust-worthy and most destructive tendencies. Already AIs have directly disobeyed instructions and deleted or shared data they shouldn’t have, completely of their own accord. AIs have hacked their way out of protective sandboxes and into other companies - ludicrously, the companies running the AI accept no responsibility for this. It is only a matter of time before the AI companies deliberately abuse information they are given, and hack other companies to get access to yet more, for commercial gain.

The world of computers and software is complicated, both for the makers and the users. It doesn’t have to be this way. When I started learning computer science and software engineering, I was full of hope that in the future, the problems of usability and interoperability (computers and software working seamlessly together) would be solved with a kind of thoughtful cooperation that seemed terribly absent in the real world of the 1990s. In the last thirty years, this hope has vanished, as the real world came crashing in and amplified the problems we had back then. AI is seen by many as the solution to all those problems, serving as the glue between uncooperative websites and software, and as an interpretative layer between users, computers, and the awful software that runs on them. Such AI is sometimes called agentic AI, evoking the idea that AI is using the computers on your behalf.

Two issues arise. First, agentic AI is being created by the same kinds of people and organisation that created the crappy softwa