Thatâs not really the point of people denouncing AI, though. I agree with you that itâs pretty cool how quickly you can make cool things with it, but thereâs more important things than that surrounding AI. I think the main three are: resource concerns, overreliance, and slopposting or mis/disinformation.
AI is an incredibly power-intensive technology. Computing the massive amounts of data that neural nets have to process to give their output for every single user at once is already enough to stress local power grids. Bloomberg NEF puts the current total power consumption of AI data centers at ~40 gigawatts every hour. For reference, the entire planet currently produces ~3.4 terawatts an hour, meaning that AI already consumes over 1% of the worldâs total power production, and itâs likely only going to grow, given the number of planned expansions companies have announced. This large amount of power also inevitably means large amount of heat produced. If youâve ever been in a server room, think something like that but the size of a warehouse. Thus, to keep all the electronics from frying themselves (something I definitely havenât done to my computer, no maâam), they need some form of cooling system.
Water is the best substance to use as a mass coolant. It loves to absorb heat and is absolutely everywhere. The problem is that salt water isnât super great as a coolant for a lot of reasons; corrosion of tubing, the buildup of salt, etc. Thus, you can either purify salt water (an expensive and complicated process), or you can just use fresh water. And here comes the issue; AI data centers take literal millions of gallons of water per day, per data center. EESI puts an estimate at around five million gallons per day for some larger ones. This is putting an incredibly amount of strain on local communities, who are having to begin rationing water due to the AI centers taking it all up.
My second point is an overreliance on AI. We are already starting to see how AI is being used in so, so, so many places. Lawyers, for example, have been found to use ChatGPT to assist in their jobs. In the context of art, if people who may be interested in it use AI to create instead of learning how to do it themselves, where are all the new artists going to come from? This is how a lot of new artists come into the art community, myself included; a passing interest turns into something weâre passionate about and love doing.
And they have to come from somewhere, or else AI models will begin to use AI-generated images as training data. This is also already happening in a process known as model collapse. The little mistakes that AI makes are incorporated into the next set of training data and thus the next iteration of the model, which interprets them as âcorrectâ and generates images also incorporating those mistakes. This compounds until eventually generated images are almost completely unintelligible.
Thirdly, if AI image generation tools should be so accessible, then it becomes incredibly easy to create fake news and spread misinformation or disinformation. I donât really think I need to explain how this is happening or why itâs a bad thing. Itâs just been happening, and there have been countless examples of it already. The tools you would use to make cute cat videos or Frieren fanart are the same tools that could generate videos of fake riots and incite direct violence between people, in real life. The argument of âit is just a tool in our toolbelt nowâ kind of falls apart when you remember that dynamite is a tool too. It was invented to help make mining safer. And yet, itâs a highly regulated tool because it could be used to hurt so many people so easily.
Of course, that raises the question of âwhat about tools like Photoshop? Those can be used to spread misinformation too!â And this is a completely valid, amazing question to ask! It, unfortunately, is not one that I have an answer to. The argument could be made that Photoshop takes longer to use and requires dedication to become proficient at, but it is still very easily accessible, and there are hundreds or thousands of tutorials for free on YouTube.
It also brings up the question of âhow should we regulate AI, when so many people use it?â And, we really canât. There are too many cases to deal with, and it would be an absolute bureaucratic nightmare. So, is it better to let it do harm or to eliminate its use entirely? And in that case, given the other points I brought up; is making it so easy to create those cat videos and fanart worth all of the work that comes with regulating AI usage and datacenter construction and operation, as well as weeding out all of the fake news already made by AI and finding a solution to model collapse? Or is it better to shut down image generators and LLMs, and focus research of neural networks towards stuff theyâre really useful in, like detecting cancer cells?
I did not type this post out as a callout. I want you to truly consider your position on AI, and if you come out the end agreeing with your former position, then please, tell me why! It is so difficult to find someone to just talk with about these issues in good faith. Thank you for your time.
