This is the first part of two, in which I focus on AI. In this part, I will take a look at some of the issues surrounding artificial intelligence, that I find most problematic. Part two will focus on the positive.
It’s almost trendy to hate AI these days. I can’t jump on that bandwagon, though. After all, I enjoy tinkering with the technology myself and just a couple of days ago, I added a couple of new AI models to my Huggingface profile. I have several models and data sets there for anyone to use. Yet, I still agree with much of the criticism aimed at the industry.
Let’s make it clear that artificial intelligence and machine learning is much more than generative AI, like ChatGPT or Midjourney. AI/ML exists in many forms, has been around for decades and is already deeply integrated in various industries. But when generative AI reached a quality level, where it could produce passable content (not necessarily good), everything changed in the eyes of the public and money started pouring into developing AI further.
I have always been fascinated by simulations, automata and procedurally generated things, so artificial intelligence and machine learning are topics I started looking into several years ago. I have built countless tiny neural networks over the years, just to learn how they work and what they can do. However, I should stress that I do not have a degree in computer science.
Legitimate Complaints
The Licensing Issue
AI companies have employed some very questionable methods in order to obtain training material for their massive models. Training material they arguably should have licensed before using, in particular art in all its forms.
My personal opinion on this is that these companies should have paid industry standard licensing fees or hired people to produce training material for them under contract. Like any other business would have to do, if they needed massive amounts of art, written content, etc. The AI industry (and its main investors) certainly has enough money to pay for this.
I don’t claim to be an expert on how retroactive licensing might work at such a scale, or how they would even figure out who to pay; that’s what lawyers are for, I guess. At the very least though, they should set up some serious grants that give back to artists and creators – without a clause that grants any kind of training license.
There is no way to undo what has been done in this case, nor would it benefit anyone to do so, but I don’t think we should let the AI companies off the hook just for that reason.
Environmental Impact
The next legitimate complaint might actually be worse than the first one. I only list it second, because I hear people talk more about the licensing issue. The fact is that all data centers need a lot of electricity to run, and produce a lot of heat, which requires more power to dissipate and funnel away. It’s quite staggering, how much juice the AI compute farms require. There literally isn’t enough power in the world right now, for what the major players want to do. If you care about the environment, you are likely to be dismayed or worse by this.
The best way I can frame this, is by saying that these companies are trying to get approval to build their own nuclear power plants, just for the compute farms.
To me, this just seems crazy. There has to be a better way to scale, even if it takes longer to find? Which leads nicely into …
Greed
If the industry wasn’t in such a rush, it would be possible to take a safer, more ethical, less destructive approach. This would include research into better solutions for everything from GPU design, power consumption, heat dissipation, licensing and so on. As far as I know, all of that research is happening, but at the same time no one is pausing to wait for it.
Of course, what I am suggesting might sound like a fever dream, since it requires a long-term, holistic view of industry and environment which doesn’t exist in practice; it is in direct conflict with shipping products fast and pleasing investors or share holders in the short term.
The tech industry and its generally poor ethics is problematic and needs better regulation on a global scale, unionization and sustainability. This goes far beyond the realm of AI, into chip manufacturing, labor practices and so on. But such regulation is not likely to happen when there are billions to be made by being “first to market”. Greed nearly always wins, and I hate that.
The Bubble
My first professional job started about 10 months before the dot com bubble burst. Hundreds of internet companies went out of business, including the company I worked for. Startups were hit the hardest, but even big companies saw their value plummet in the short term.
I have been through enough “bubbles” at this point to recognize them. The AI/ML boom right now is most certainly a bubble, and when it bursts, a lot of people will lose a ton of money. But that doesn’t mean that the technology is a flash in the pan. I do believe AI is here to stay.
The world wasn’t ready for eCommerce quite yet when the dot com bubble burst, but after the crash came a new, measured phase of growth and consolidation. Today, we all buy stuff online all the time. I think the same thing will happen with AI – the bubble will burst and there will be lots of fallout, but when the dust settles a few players will remain standing, the hype will diminish and AI becomes just another industry.
Some people fear that AI is going to eliminate their jobs, and that is going to happen in some cases. Any time there is a technological advance, it ends up making some jobs obsolete. It also creates new jobs that didn’t exist before, though, but where exactly that balance ends up is a good question. I am not going to pretend to know the answer to that one.
Am I a Hater?
No. Despite the things I dislike about the industry and the way it operates, I do not hate AI as a technology. I find it endlessly fascinating and believe we can benefit greatly from it in many areas, like science and medicine, for example.
In my next post, I’ll dive deeper into what I like about artificial intelligence.
