'AI'에 저항하라

--> Yes, the “AI 1 Genie” seems actually to be useful. But it is also seductive, selectively harmful, and in important ways harmful to us all. Executive, manager, consultant, developer, ordinary person: Resist! TL:DR The “AI Genie” is not your friend. Mea Culpa I do not use any of the programming “AIs”, not even whatever is built into JetBrains IDEs, and I do not intend to. However, my hands are not squeaky clean. After trying Kagi/Duck Duck Go, I have gone back to Google for search and I freely grant that I find the “AI Overview” to be useful. It is frequently correct and offers the answer I’m looking for. I do not come to you with clean hands. Purpose I am not here to dictate morality, but only to make the argument that “AI”, as it stands, and as it is tending, is not a net good for us as individuals, as organization, as society, as a world. I believe that each of us, in whatever roles we have in the world, should generally be strongly resisting “AI”, and helping to find ways to maintain good lives for everyone in the face of a technology that is already disruptive and will only become more so. My purpose here is to be an active part of what should be a growing resistance to “AI”, at least until its very significant harms have been eliminated, mitigated, or compensated for. Focus My primary focus here will be software development, but in a broad sense. My readership here is mostly software developers, but there are a surprising smattering of others from all over the world and the map of professions. And each reader, of course, connects to many people from all aspects of life. I think we all need to do what we can to resist the growth of “AI”, unless and until its harm for people and the planet is dealt with. Harms Environmental Harm It may be true that an individual “AI” query costs just a few cents and a small amount of water and power. It is also true that “training” that current “AI” systems code up to $100 million, producing huge quantities of CO 2 and consuming over a billion Watt-hours of electricity and a million liters of water. The ongoing power consumption of a typical “AI” data center is about equal to 100,000 households. Daily water consumption is around 50,000 households. I am not exaggerating: We need to resist “AI” so that it doesn’t ruin our world. Intellectual Harm Some matters of interest are ephemeral. We wonder who that actor is in Hunger Games 2 : his voice sounds familiar. Oh, right, Stanley Tucci. Cool. But for most of us, there are areas of expertise that are important to us, in our profession or our hobbies or interests, and here we need real understanding. Real understanding comes from actively working with the materials and ideas of our subject area. We can learn a bit by reading, or watching a video, but when it comes down to it, we have to do the thing, thinking about the thing, probably many times before we’re really good at it. When we let the “AI” answer the question, sure, we might check it over carefully. If you’re a developer you already know how well that works. Code we actually thought about and wrote seems right even when it isn’t, sometimes subtly, sometimes blatantly. Code we’re reviewing? We look for something easy, like a typo or violation of indentation, call that out, and then return to whatever we were really doing. When we let the “AI” answer the question, we do not learn. We do not always catch its errors. We lose the picture of the very thing we are responsible for building. I am not exaggerating: We need to resist “AI” so that it doesn’t damage the work we do. Human Harm Corporations have already begun to try to replace human workers with “AI”. The “AI” creators themselves are even shedding programmers and other workers, conceivably because they think “AI” can replace them, more likely because they have vastly overspent creating the “AI” models and have not been able to offset those costs with revenue. If we lived in a post-scarcity 3 society, having your job eliminated by “AI” could be a good thing, freeing your time to be spent on things more pleasant than work. You would still have full access to all the good things of life, food, health care, transportation, the Internet, education, recreation, and so on. In our current society, you’ll lose your income, be unable to support yourself and your family, and there is no effective social system to keep your life style where it was. I am not exaggerating: We need to resist “AI” so that thousands, perhaps millions of people do not die. Seductive The “AI” is programmed to seem polite, to seem friendly and accepting, to seem willing to help, to be open to your every idea, to be happy to try to please you. “You’re absolutely right!”, it says. Get serious, you might be somewhat right at best. The “AI” is seducing you. It is telling you things you want to hear, it is pretending that it will respect you in the morning. It will not. If you don’t remind it, it won’t even remember you in the morning. It does not contain respect. It does not contain good cheer. It does not contain willingness. It does not admire your thinking. It contains none of those positive feelings that it expresses to you. Look up the “Eliza effect” . Even the weakest conceivable program like this was seductive. These new ones are masters of seduction. I am not exaggerating: We need to resist “AI” because it is inherently far more seductive than it is valuable, and we are poorly defended against seduction. Resistance Advice What forms can resistance take? There are many, from complete refusal to touch “AI”, to careful honest study, to outright sabotage. Here are some ideas for folx with various roles in the world. They’re just my current quick thoughts: I have more concerns thus far than advice that goes much beyond “Resist”. Executives Resist. You’d be foolish not to get on top of “AI”: it’s here, and will only proliferate. However, because it is far from uniformly good for you or your company, I’d advise a very judicious approach to it. Don’t just send out the memo saying that everyone has to be 35% AI by next fiscal year. Instead, form a working group of smart people from all areas, some who favor “AI” and some who do not. Have them devise and propose experiments with “AI”. Not the BS kind of experiment: “Do an experiment showing how to save 30% in your department”. A real experiment: “Devise an honest AI experiment in your department, measuring real costs and real benefits. Try, honestly, to find all the flaws with this AI stuff.” Your first experiments should be designed to find out how bad “AI” is, and from those you can then have your working group work to find ways to use the stuff well—if that is even possible. Keep in mind, if you can, the overall costs to your company, to your people, and to the world. Work to understand. Then, later, to do. Managers Resist. My advice here is essentially the same as to executives, at the scale of your area. It’s almost certainly more important to understand how “AI” will fail you than to understand how it will help. Don’t demand that it work: they’ll just lie to you. Ask for the truth, the nitty gritty dirty truth. And, I suspect, you should be considering how much of your own work can be done by, or using, “AI”. Someone is going to be tempted to use it to write your quarterly reports. Keep in mind that the “AI” just averages its inputs and that it often gets it wrong. Keep in mind, if you can, your people, the ones who will be harmed by over-use of “AI”, perhaps by any use at all. Developers Resist. If you don’t have to use it, do not. If you do have to use it, use it as sparingly as possible. Try to use it well, and try to track more than just the times when it seems to be really good. Track all the times it’s wrong, all the times it goes down the wrong track. Track the time lost as well as the time saved, if any time is saved. Try to work, with your colleagues, sometimes with the “AI” and sometimes without. Try to keep track of how things go