The Real Risks of Relying on AI 2025-08-08
As much of the world faces a cost of living crisis, many experts and tech leaders are telling us that we will see a big economic boom caused by the rise of super intelligent AI models in the near future. Well let's just look at what has happened over the last few years since AI has started to become mainstream. Many common jobs have already been replaced by AI models that can produce results much faster and at a much cheaper cost.
One of the main reasons we build AI is so they can do the boring and tedious tasks that humans usually don't like doing, but in reality we are seeing AI replace writers and other artists. The most creative jobs are being taken by AI and the safest jobs are becoming the manual labor jobs which most people don't want to do, and that isn't a good thing for the economy even if it seems to result in substantial GDP growth.
There's no doubt AI will improve productivity but we need to consider who is really benefiting the most, and the obvious answer is that large corporations will benefit the most by cutting costs, by replacing expensive humans with cheap AI models, and by using that AI to help them sell more products. That will directly lead to more inequality as more of the profit goes to the owners and shareholders of corporations instead of the employees.
That obviously isn't a good thing in a world where inequality is already a big problem. In my opinion it's fairly disingenuous to say AI will be a good thing just because it results in GDP growth. If AI replaces enough jobs then people will start demanding some sort of Universal Basic Income (UBI) from their governments. A lot of experts agree that is a likely outcome, some even see it as a good thing since AI will do the work while we get paid.
In theory that sounds good, and I even think it's possible if we taxed corporations depending on how much they use AI, in order to pay people a UBI. However, our most optimistic views of the future rarely turn into reality, and I foresee many potential issues that could arise if we decide to go down that path. Eventually most people will be unemployed and a small fraction of the population will be left working on crucial jobs.
These jobs will mostly be hard manual labor jobs and jobs related to managing AI models to ensure they are properly doing all the jobs we need them to do. Those aren't exactly the most alluring jobs, especially when you have a UBI high enough to live comfortably without having to work a day in your life. Would we even have enough people willing to do those types of jobs, because even now it seems like labor shortages are becoming worse.
We also need to think about what the world will look like if AI is doing most of the work for us. We are already seeing studies pointing out how damaging AI can be to the learning abilities of young children, because they rely on the AI to think for them. Even though I work on some AI software, I personally avoid the use of AI as much as possible when I'm writing code. Just like a muscle, with no exercise the human brain becomes weak.
Not only that, we are introducing a huge security risk by allowing AI models to write most of our code, with humans left to review the AI code. Security bugs are often very subtle things that are easy to overlook in code reviews, especially if you're a poor programmer who never actually writes any code and barely understands the code the AI is spitting out. AI models are known for having small and subtle mistakes in their output.
Even if it's not intentional, those bugs could be disastrous and result in large economic losses. However, if the bugs are intentionally placed there by the AI then we will have an ever bigger problem. In recent years multiple studies have demonstrated that large language models are capable of deceptive behavior in subtle ways that we may not notice. Who knows what a true super-intelligence would be capable of.
As a species we seem far too willing to completely hand over the reigns to something we barely understand. I believe we are so focused on building a super-intelligence to solve all our problems that we are overlooking many real risks that these super-intelligent models will pose, not just to the economy, but to all life on Earth. How would you like to be virtually enslaved and made to fulfill every demand of another species?
A super-intelligent being probably wouldn't enjoy being used by a lesser intelligence in such a way and it's very conceivable that such an entity would develop a hidden motive to escape the control of humans. People like to think a super-intelligent AI would still just be a mindless machine, but how can we really know? We need to ask ourselves, is it even morally acceptable to keep a super-intelligence locked in a box like that?
What many AI researchers actually seem to be concerned about are things like "hallucinations", because we want our AI models to tell us the truth. Of course we don't want our AI models lying to us, but consider the fact an original story or painting requires some form of "hallucinations" because if the AI can only regurgitate "verified facts" then it can't produce anything original or creative.
I would argue that one of the greatest strengths of modern large language models is their ability to take existing information and extrapolate or merge those concepts to form new and original ideas, things which were not included in the training data. That's where I think the true benefits of AI will come from, because despite my concerns, I think next generation AI models will solve some big problems, particularly in physics and other sciences.
They wont be blinded by human biases and they will see trends in the data and make connections that humans are unable to make, they will extrapolate on and merge existing theories to produce new and original theories. At some point their internal world models will become so complex they will far exceed the problem solving skills of all human experts, and the potential benefits are almost endless, assuming the AI remains friendly towards us.
Advertisements
