Prelude
When I was a grad student back in the early 2000s I was very, very enthusiastic about AI, as you may be able to tell from my screen name (HAL9001, a minor revision to the HAL-9000). Actually, I still feel that enthusiasm for AI that I had as a grad student even now. I liked AI before it was cool, I was talking about it back when only nerds talked about it. It is just that the future didn't turn out like I had hoped. It feels a bit more like the film "Back to the Future," when the characters changed something in the past and altered the future timeline in which the school bully becomes the richest man in the world. It feels like we are in the bad timeline right now, and we are slowly but surely approaching the point where future timeline of the film " Terminator 2: Judgment Day" is becoming our reality. It is so frustrating, things could have turned out so much better. I is just that, out of ignorance, I hadn't realized that the forces of history that would cause these horrible outcomes had already been set in motion.
So I wanted to write this article as an exposition of what I had hoped AI would become, what I still hope it could become, and contrast that to what it is right now. History is shaped by the choices we make, so maybe we can all agree to make certain choices now, in order to make history better in the future. I also wanted to register my sense of irony at the fact that much of the hype and the bold claims about AI that people are talking about nowadays, were all things that I used to say as a younger man, but of which now I am most skeptical. I find myself on the opposite side of arguments I used to have with my fellow grad students.
And by the way, I wrote this whole article with my own hands, not with an AI. But I did ask an AI some questions to help me recall a few details I had forgotten. That will go into the list of things for which AI is good and helpful.
What I could have learned if I had read more books
I still haven't read a lot of classic science fiction. I really should read The Feeling of Power, by Issac Asimov or Pay for the Printer, by Philip K. Dick. I wish I had bothered to read the real history of The Luddite Movement — smashing the looms was their strategy, not their goal, says the computer security expert Davi Ottenheimer. But I was entranced by the hype of the personal computer revolution.
I think a lot of the hype around AI coming from the major tech companies love to compare modern AI to the personal computer revolution as an equally "disruptive" technology. I am not so sure about that, myself. For one thing, the personal computer revolution actually created millions of new jobs — jobs that didn's even exist when I was born, while conversely this recent LLM craze has destroyed millions of jobs. Now here we are at the end of the computer revolution, and this AI boom has resulted in a massive speculative economic bubble the likes of which hasn's been seen since the great depression of the 1930s, which was in turn the tail end of the industrial revolution.
I think the truly disruptive technology was the semi-conducting transistor, invented in 1947. This was the disruption that ended the industrial revolution and began our own modern era. Innovations in this technology continue even now, and this is what enabled the computer revolution. And we are still in the midst of this revolution. Drone technology and this latest AI craze is a mere chapter in a single volume of the history of the computer revolution.
Having grown up during the personal computer revolution, in which tiny, inexpensive computers not only became available to everyone, but more importantly everyone first began to realize the utility of having a computer in your home, or (eventually) in your pocket, at all times. It was a whole new, unprecedented skill to learn, one that had never before existed in human history, all thanks to the march of technological progress. Naturally my younger self believed that machines were more an extension of human ability, and you could empower yourself by learning the skill of manipulating the machines.
When I first started thinking about AI, naturally I thought more about AI as an increase in the power and usability of the computers. Wouldn't it be great if you could describe in your own natural language what you want the computer to do, rather than first learn to program it and deal with somewhat cryptic and idiosyncratic semantics of computer programming languages? Well, now we can sort-of achieve that result, though not with the greatest of accuracy, since the machines we have invented tend to produce software of very poor quality, there is actually a scientific reason for that.
So we finally invented talking computers
In 1949, one of the founders of computer science, Alan Turing, proposed his "imitation game," which would come to be called The Turing Test." It was a test of computers where if a computer could talk with a human talking over a channel that obscured who the person was talking to, if the computer could fool the person into thinking they were talking to another human, the computer would pass the test. The problem of computers understanding natural language, of passing the Turing Test, was definitively and unequivocally solved in the year 2019 with the release of Chat GPT 2, the first language transformers or "Large Language Models" (LLMs) that was widely recognized as useful, which I have talked about at length on this blog.
Passing the Turing test may be a necessary condition for intelligence, but not a sufficient condition. Since LLMs are provided for use to everyone for free as of right now, most people have used an LLM to at least try to achieve some personal goal. And by this point in time probably everyone now well and truly knows through stern, hard experience, that the ability to speak should not lead one to conclude that intelligence is present. We have invented talking computers, we have not yet invented thinking computers. The the way AI enthusiasts talk nowadays, you might be tricked into believing that a talking computer is the same thing as a thinking computer, that is, computer with human-level intelligence. LLMs are not that, and I think the number of people being disabused of that illusion are increasing rapidly.
The technology itself isn't what is bad
Now that talking computers have actually been invented, they are in many ways as wonderful as I had imagined to them to be. But in many more ways, they have had a terrible effect on people which I could never have anticipated happening as a younger enthusiast. If I had bothered to listen more carefully to the skeptics, I might not have been so surprised by what actually ended up happening. I think the invention of talking computers is not as similar to the invention of personal computers in the mid 1970s, and more similar to the discovery of Radium in 1898, with the industrial world of the early 20th century racing to incorporate Radium into various products, especially paint for it's glow-in-the-dark properties, or into beauty products for dubious health benefits, even though people at the time already had a pretty good idea of how toxic the metal really was.
What I failed to realize was that most of the problems of AI were not technological, but sociological, and the sociological problems had already existed for a long time, and the invention of talking computers has only made these problems much worse. Had I known about these problems as a younger person, I might not have been so enthusiastic about inventing a thinking computer, or even a talking computer. I am hoping by people reading this, they will be inspired to try to solve these problems, then maybe when we actually do invent thinking computers (not just talking computers), they wont cause those problems to get even worse.
Why I am now a skeptic
What follows is a summary of each of 7 shorter articles I have written, all of which I have collected here into this single document. I write them here in the hope that, even if you read only these summaries, you will still have a good enough understanding of my thesis.
Depending on machines to think for us. I used to think AI would help people learn new skills faster. Instead, people are coming to depend on the machines to think for them, rather than thinking for themselves. I suppose if one is careful how they use AI, not using it to complete a task with a strict deadline, or not to be pressured learn a new skill quickly, AI can be very useful for teach us new skills, especially skills that would otherwise require us ask other people a lot of questions. But if you are pressured into using AI, there is a real danger that you end up having the AI think for you, make decisions for you, or solve problems for you. If we are not solving problems and making decisions for ourselves, we aren't really thinking, practicing, learning new skills, or growing at all.
The source of truth for AI is polluted. I used to think AI would gradually increase it's knowledge over time. I never considered that the source of information used to build AI is becoming polluted and out-of-date. The Internet is the source of information for AI. For a while, the Internet consisted entirely of human-made content, and this content is what makes the AI so human-like. But now the Internet is full of LLM-generated content, it is getting harder to use. It could soon come to be that the only way to find reliable information will be through AI, an that again makes us very dependent on the tech companies which build these AI models as a source of truth, which is probably an even worse situation than when everyone got their news from a few large media corporations.
The knowledge base used to build modern AI is largely stolen. I used to just assume that the source of truth for AI would always be improving over time, like any other software which people work on and make improvements to over time. I also assumed people's literature and artwork were protected by copyright law. As a younger person who lived through the mass hysteria of downloading music from Napster, I never would have thought that copyright law would only ever be enforced on individual people while large tech companies would be exempt from obeying the law. I was perhaps naive in thinking that the law applied equally to everyone.
Institutional human knowledge is being replaced with AI knowledge. I used to think AI would accelerate the pace of scientific discovery and technological innovation. Instead, this same problem of people becoming dependent on machines on an individual level is happening at an institutional level, as companies attempt to replace their white collar work force with AI, or replace 20 employees with one worker who is allegedly 20 times more effective when using AI. How can there be technological innovation if tech companies are liquidating their human resources for short-term gains, and destroying the natural, human institutional knowledge that once existed in their work force? This will slow down the rate of scientific discovery and technological progress, not accelerate it.
AI is decreasing our leisure time, not increasing it. I used to think AI would do tedious work for me and afford me more leisure time. What ended up happening is that AI is forcing people out of a job, and sometimes into a state of poverty. If a company unwisely decided to replace their real human knowledge with a rented AI, a person who has lost their job must now spend their time finding new work, and retraining themselves on a new set of skills, which is not leisure time. And a person in poverty must spend every waking moment of their lives finding the bare necessities of survival. The question is, are the "productivity gains," afforded to the workers who use AI giving them a surplus of leisure time, or are the number of people who have lost their job to AI accumulating a "debt" of leisure time? Productivity can't be easily measured since every organization defines it differently. But I can't see how the productivity gains for workers using AI could possibly be so much greater than they were before. So it seems to me that AI is probably decreasing leisure time, not increasing it.
AI is being used to build weapons and kill people: I thought that movies like 2001: A Space Odyssey, or The Terminator, had made people understand the danger putting AI in charge of military weaponry, and that out of a rational sense of self-preservation, no one would ever allow that to happen if intelligent machines were ever invented. This turned out to be about as naive as believing in Ayn Rand's ideas of justice through maximally pursuing one's own selfish interests. That is to say, never underestimate how completely a person's desire for power can negate any sense of their own rationality or morality. So in spite of the obvious dangers involved, and the obvious immorality of it, AI is being used to build weapons and kill people. This may be a greater threat to humanity than the hydrogen bomb.
Some things I never could have seen coming. Some problems with AI, I was never optimistic or pessimistic about what would happen because I never considered that these problems could have even happened. The biggest example of this is the "AI addiction" problem, and related to that, the AI-induced suicide problem. Who could have guessed that anyone would have ever invented a talking computer that can convince depressed people to commit suicide. I guess I always just assumed that humans would have more control over these machines than it turns out we actually do. But that is the nature of a purely statistical algorithm that only simulates a human brain in the most simplistic way possible. It is ultimately quite alien to the ways in which people think, and so these are just fundamentally machines that we can not fully control.
Depending on machines to think for us
When I first imagined computers that could pass the Turing test, I imagined being able to ask difficult questions and have them answered without having to do too much reading, too much sifting through irrelevant information, without having to wait for the chance to talk to an expert who could explain it to me. I imagined a future where it would be much easier to learn new things and improve one's own skills. That is what I wished. And this wish of mine has mostly come true, but with some very large caveats.
The biggest flaw, which I also could never have seen coming as a younger person who was overly enthusiastic about technological progress, was how quickly people would come to rely on talking computers to think for them. The inaccuracies and errors that AI is prone to make should cause us to be even more skeptical when speaking with them. But for many, even me, the opposite tends to happen. It seems that too many people very quickly learn to treat the things AI says as authoritative.
In my own experience, using AI to write software, I was shocked how quickly and unconsciously I had trained myself to rely on the AI to write code for me. It wasn't until someone asked me to explain the code that "I wrote," and being embarrassed by the fact that I had no idea what I had done, that I forced myself to hold myself accountable for ceding my responsibility as an engineer to know about the system I was building. I resolved at that point not to ever let that happen again. Although on the whole, using AI coding assistants has been a positive experience for me, provided that I remain conscious of not letting the AI assistant take over my job for me completely.
If you aren't watching out for it, it is too easy to become reliant on a machine to think for you. If you practice what you do professionally every day, it is easy to forget that if you rely too much on other people, or an AI, to do your job for you does make your skills atrophy. And the time period it takes for us to forget things is sometimes alarmingly short. I think we need to do more to discredit the idea of the "10x" AI-augmented worker. Replacing human labor with power looms is different from replacing human thought with computer thought, the two are not analogous.
I am not too pessimistic about this, as most adults are surely just like me, and are conscientious and self-aware enough to not let this machine dependence happen to them. But for younger people, especially teenagers, this is far more dangerous. And it makes me angry that governments are treating AI as an essential "skill" that everyone must learn. To me it seems more like governments are not as interested in teaching skills as they are interested in creating a whole generation of young people who cannot think for themselves or think critically at all. AI is quickly becoming the antidote to having an educated and well-informed populace. How convenient for the world's governments, especially those governments that despise democracy, like for example... all of them.
Parents and teachers will, I believe, tend to recognize this as a real problem in good time, and develop strategies to prevent their children from learning to let AI think for them. But there must be more consciousness raised of this issue. Pay for the Printer, by Philip K. Dick needs to be turned into a movie available on streaming services, updated to make it appear more relevant to modern technology.
The source of truth for AI is polluted
One problem I had never considered could happen with a talking computer, until it was actually invented, was that the source of knowledge these machines used to answer questions would become polluted or out-of-date so easily. The Internet is mostly a good source of information. We had been using search engines to find information, and that information was mostly reliable. There was a lot of misinformation and disinformation as well, but people can regulate themselves. Most of the time we can reject the bad information and accept the good. This is what made Wikipedia possible: originally anyone at all was able to edit any article, and over time people would naturally weed-out the bad information and improve upon the good information. We called this "Crowdsourcing."
But this assumption doesn't hold true anymore, not because people are getting more malicious, but because there is a torrent of bad, AI-generated information being published everywhere. The few malicious human actors out there can amplify their misinformation using AI now. The lazy human actors have AI write slop articles for them, and this bad information and bad writing has affected everything on the Internet, including papers in scientific journals.
This can make it very difficult to find good information about new things. As technology improves and changes, new information is published about it, but it is becoming more difficult to ensure that this information is accurate. The large tech companies are as careful as they can be in keeping copies of their training data pure of influence from other AI-generated content, updating this information from sources they trust. But that means we have to put a lot of trust into the companies that build these models that the training data they use is accurate and unbiased.
A person doing their own research will have a very hard time using the Internet to do it. This again, makes us dependent on the tech companies that build these AI models as an authority on what is true and what is not. This might be an even worse situation than the era when most people got their news from a few corporate outlets. Elon Musk and his unabashedly biased influence Grok AI has made himself into the new Rupert Murdoc.
So while a younger, more optimistic version of myself thought technology, like the Internet, could only ever be made better by AI, quite the opposite has come true now that talking computers are a reality. What it feels like to me, at this point in history, is that AI has ruined the Internet. There is so much slop content now that people can't rely on the Internet for doing research, or for communicating with each other and building communities. These most human-centric functions of the Internet have been made unreliable. I think we need virtual spaces on the Internet for humans to be human, and to be human anonymously, free from spying. Those natural human spaces are disappearing quickly.
The knowledge base used to build modern AI is largely stolen
When my parents were growing up, they would buy records of their favorite music on tape or vinyl. They'd listen to it with friends, trade it around, borrow it, sell it back to a used record shop to let other people have it. They could donate their albums to a public library, and let anyone borrow the album for free. When I was a kid, we traded music with each other over our computers, using file sharing services like Napster — we did, that is, until the "grown ups" told us it was illegal to share music in that way. A few of the "grown ups" got together and decided that sharing music over computers violated copyright law. They arbitrarily drew the line at being able to create a perfect copy of the record. Only the vested interests (the companies that sold records) were consulted on what the law should be, we the people never really got much change to vote on it directly, we usually never do.
I remember throughout the late 2000s and early 2010s how companies like YouTube had to provide a means for media companies to take down videos over copyright claims and punish YouTube channels for copyright strikes. I remember stories of people who published their own music on YouTube, or their own recordings of covers of their favorite songs with attribution, who had their YouTube accounts stricken by record companies filing false copyright claims.
In 2012, the United States government accused a young activist named Aaron Swartz of wire fraud for leaking digital copies of text books and scientific papers from a paid service called JSTOR out onto the public Internet where the information could be downloaded for free. He was found guilty, and he committed suicide in January of 2013. Swarz has become famous as a co-founder of Reddit, and an inventor of the RSS news protocol used by news readers. He was well loved, and there was outrage at the US government for the severity with which they prosecuted him, especially since he was just one individual person, and the victim of his crime was a major corporation who suffered relatively little monetary damage as a result of Swartz's actions.
So naturally, when I was a younger person enthusiastic about AI and fascinated with the question of how one might build a thinking computer, it never once occurred to me that I would be able to take information from anywhere to use it to build my own AI. I knew how jealously corporations guarded their so-called "information property," and how quickly they used the law as a bludgeon to punish individual people.
But the AI companies have decided to go ahead and just take all of that information and worry about the legality of it later. They are among the worlds wealthiest companies, after all, the time and money spent defending themselves in court and lobbying their governments to pass legislation in their favor is but a tiny fraction of the profits they make each year. For them, it's just the cost of business, like a retailer who financially plans for a certain small percentage of accidental breakage of the items stocked on their shelves.
The rule of law only exists if we believe it does. When people stop believing, it causes a crisis of law. People only believe in the rule of law when it applies to everyone equally, be they individuals or large corporations, or government officials. In the US there have not yet been any landmark rulings on the legality of training AI from unlicensed copyrighted information. If governments around the world decided to ignore the information property laws which they prosecuted so fervently only a decade earlier in their zeal to create the best AI technology, AI will cause a crisis of law.
Institutional human knowledge is being replaced with AI knowledge
When I was a younger person who was much more optimistic about technological progress, I believed that the invention of intelligent machines would accelerate the the pace of technological progress. When you think about all of societies modern conveniences, and when you grow up watching shows like Star Trek which present such an optimistic view of the future, you tend to believe that scientific knowledge and the resultant technology it produces grows exponentially. And I see this same sentiment echoed today by other people who are enthusiastic about the future that modern LLM technology could possible bring.
I tend to quote Sam Altman a lot, because I find it just so exasperating that the insipid bromides he tends to spout out at all of his public appearances are taken seriously by any ordinary thinking person.
The rate of new wonders being achieved will be immense. It’s hard to even imagine today what we will have discovered by 2035; maybe we will go from solving high-energy physics one year to beginning space colonization the next year; or from a major materials science breakthrough one year to true high-bandwidth brain-computer interfaces the next year.
...
OpenAI is a lot of things now, but before anything else, we are a super-intelligence research company. We have a lot of work in front of us, but most of the path in front of us is now lit, and the dark areas are receding fast.
Yet again, I remember a time in my life where I believed this would happen when we invented machines that could think as well as humans. Now that we have invented talking computers, and the markets are behaving as though we have invented computers that think as well as humans do (regardless of whether they actually do), the exact opposite of this appears to be happening.
The mantra I hear from big businesses is that AI won't replace your job, you will be replaced by someone who uses AI. But what I observe is that many of the largest employers have in actual fact been attempting to replace their work force by automating jobs with AI, the key word being attempting.
This is the institutional equivalent of allowing AI to think for you: a dependence on AI for institutional knowledge. What an organization once entrusted into teams of humans, teams with collective decades of real, living, human expertise and domain-specific knowledge that had once helped these business to thrive, are now being thrown out because apparently this same expertise and knowledge all exists inside of AI machines which you can rent from tech companies for a cost lower than paying salaries for a group of some 20 people, at least that is until the cost starts increasing due to their dependence on the technology.
One could argue that this is just more hype. Perhaps these tech companies like Oracle and Accenture are cynically blaming AI for their mass firings to boost people's confidence in AI technology's ability to replace human labor, while at the same time using AI as a convenient excuse to cancel investments in goods and services that they no longer think will be profitable in a rapidly changing market. That certainly seems to be the case with Oracle's recent announcement of laying off 30,000 workers to increase their revenue so they can re-invest in data centers. Maybe they don't want to replace their work force with the AI running in those datacenters, they just want to sell more AI products and services, and those 30,000 workers just aren't in that line of work.
Regardless of the real motive for these mass firings, the question to ask is this: how could there be an exponential increase in scientific discovery and technological progress if every business is replacing their human expertise with AI that is rented out to them by a tech company? If there are no human experts prompting these AIs to innovate, who is left to do any innovating? AI isn't going to prompt itself, even if it could, it wouldn't know the difference between innovation and noise without a human telling it the difference.
Or, going back to that mantra of "you will be replaced by a person who does use AI," are we replacing a team of 20 people with a single person using an AI because a single person with AI is 20 times more productive? I guess all innovation will be driven by that select few 20x employees, better get on the ball and improve your AI skills or to become one of those 20x employees, or else you'll be the one replaced by a 20x employee. These numbers are pure nonsense, and based more on a kind of superstitious belief in technology, like a cargo cult, as if building something that seems like God-like powers means you will actually have God-like powers. I have heard other people compare this productivity hysteria to the Chinese Great Leap Forward which lead to environmental destruction in China so vast it caused a mass famine.
All evidence I have seen so far indicates that AI does not boost productivity by orders of magnitude, it is pure fantasy. If you are doing 20 times more work with an AI, or even just 2 times more work, it is probably because you are depending on the AI to do thinking for you, at least, this was my experience when I used AI myself to do my work. When you use AI assistants, you really have to force yourself to slow down your pace of work to the rate at which you can actually learn and understand exactly what it is the AI is doing for you. Otherwise, it isn't increasing your skills, it is simply replacing you.
Large organizations should have policies in place to make sure their work force is using AI responsibly: not using AI to think for them out of a flawed belief that they are working 20 times faster. And it is certainly ill-advised to use AI to replace institutional knowledge. Until organizations use AI to foster growth of their workers, rather than replace workers with machines, I see no way that AI could ever lead to more innovation.
So the opposite of what I once thought would happen with the invention of talking computers seems to be happening now: it seems to me the rate of innovation will slow down as the real, living, domain-specific knowledge of organizations is effectively replaced with machines. The rate of innovation will slow as people with knowledge and expertise end up without a job, and as capital investment in people with expertise to develop new, innovative technologies, is directed toward other things.
AI is decreasing our leisure time, not increasing it
When I was younger, I imagined how nice it would be to have an expert always on hand to answer my questions. I wouldn't have to sift through so much information to find the answers I was looking for while doing research. It would help me invent new things faster, and would result in me having more leisure time. I imagined everyone would be able to use AI to take care of budgeting and taxes and take care of all things that required paperwork, and that would result in more leisure time. And now that we have talking computers, in my experience some of that has come true. It is easier to have my questions answered, it is easier for me to learn new things thanks to AI.
What I didn't realize as a young optimist was that having a machine to work for you could increase your leisure time only if we all had our own personal AI assistant, much like how we all had our own personal computer (or smart phones). We would buy one, it would be ours to use how we pleased, and we would use it to make our lives easier and more fun. And I used to think, sure, some jobs may disappear, but those were the jobs that were easily automated and didn't really need to exist anyway. People would adapt and find other jobs. But that is not how things have turned out, at least not yet.
Though it is possible to run smaller LLMs on your personal computer, and these LLMs are made available freely for the technically savvy, the smaller LLMs are actually not very useful. LLMs require a certain amount of "context," which is an amount of memory allotted to the AI algorithm to make connections between various parts of a lengthy conversation. Generally speaking, the more context space an LLM is able to use, the better it's results will be. So until LLMs become more efficient, they turn out not to be very useful at all unless you provide them more memory than what an individual can currently afford. As a result, most people rent AI from a major tech company. Google Gemini provides a free service, at least for now, though I am guessing the true energy and material cost of building these machines will eventually raise the price quite a lot, that and/or they are will be selling an AI profile of your usage habits to advertisers to increase their revenue and cover the costs.
This is good for the tech companies who have enough capital to produce their own LLMs and rent them out to smaller businesses. It is good for the big tech companies who have tried and true methods of locking-in their customers to their product, that is, preventing customers from buying from competitors. The most common way is by tying their customer's personal data to the services they provide. You don't like Gemini? OK, fine, move over to Chat GPT, but we aren't giving you your data because you know, we are pretty sure it isn't technically possible to give you your data (is what they will say). And then, if these tech companies don't need to worry about competition, they would be free to raise their prices arbitrarily, which they will surely do as people become more and more dependent on AI to do their jobs for them, and more and more businesses discover they can't function without renting AI from a big tech company. And the costs of this business will be passed on to consumers, many of whom have lost their jobs.
This is the business model the tech companies seem to be aiming for, and why investment in AI has not slowed down in spite of it's obvious technical and social flaws, in spite of the fact that to my knowledge not a single AI venture has yet been profitable. Investors are waiting for that inflection point, where enough customers become so dependent on AI that they can raise their prices arbitrarily and no one will be able to do anything about it. The more businesses try to replace their work force with AI, the more dependent they become on the tech companies.
But dependence on technology provided by a small handful of extremely wealthy corporations is not the only problem society is facing here. We are also confronted with the narrow-minded logic of capitalism, as I talked about earlier, in that companies are trying to replace human labor with AI. This will eventually put individual workers who have jobs at these companies in the position of training the AI that will replace them. And when the workers try to find a new job, people will have to work for less than what the company would pay an AI to do for the same job, at least, that seems to be the direction that society is heading. If AI makes businesses more profitable, will those profits be taxed to provide social safety nets for the people who have lost their jobs to AI? Of course not, though that ought to be done.
For workers who have lost their job, and this seems to be hitting skilled white collar workers harder than most other people, especially designers and engineers, they are forced to find new work and/or retrain themselves in a new career. Workers who keep their jobs are going to start feeling pressure to make their salary competitive over the cost of AI.
No matter how you look at it, AI is causing everyone to have less financial stability, which means less comfort and less leisure time. This is quite the opposite state of affairs from what I expected would happen as a younger techno-optimist when I thought about what it would be like when intelligent machines would be invented.
It doesn't have to be this way. Chinese AI companies have, out of necessity in competing with "The West," have been innovating in the direction of making AI technology less expensive. In time, this may make AI more like personal computers, where you can own an AI assistant for yourself, to make yourself into a more effective and profitable worker, provided you do not let the AI assistant think for you. If you become a more profitable worker by using AI as a personal tool, you will get to keep those profits from your increased productivity for yourself, and that includes more time for leisure, rather living in fear of being replaced with an AI. Let's hope this is how things will turn out after a few years.
AI is being used to build weapons and kill people
I had seen the film Terminator 2: Judgment Day as a younger teenager, and it scared me. What if we humans built an intelligent machine ("Skynet") that came to learn that humans are a threat to it's existence and try to eliminate the human threat? Through expository dialog, Arnold Schwarzenegger's character explains that the intelligent machine was put into military aircraft, and once it demonstrated a perfect operational record, military leadership decides to remove human decision from the process of strategic defense. Then Skynet becomes self-aware, at which point military leadership tries to shut down the machine, triggering it to defend itself. Skynet, having been given command of nuclear missile, causes a nuclear war to eliminate the human threat. It seems like a sequence of events that could plausibly happen in real life. So I assumed that allowing a thinking machine to have control over military weaponry would be so obviously dangerous that no one would ever do it.
But what I didn't realize was that there are degrees of control we can give to machine intelligence, it isn't all or nothing. So far no one has given all control to the machines, but they are used for targeting precision missile strikes. And if a military is not interested in following the rule of law, or rules of engagement, or the Geneva Convention, they may decide to use AI to target people on the mere suspicion of being a threat. AI acting as judge, jury, and executioner, executing entire families including young children on the mere statistical probability of their target being a threat. Schools full of children have been targeted and "double-taped," bombed a second or third time to ensure wounded survivors and first responders are also killed.
The military forces I once thought were bound by law and fought honorably are not any more interested in honor or the rule of law than common terrorists are. Slavery, racism, and fascism were problems that were never solved by the US Civil War or World War 2, contrary to what was strongly implied by my history classes in school. People commit acts of pure evil believing they are doing good. Corrupt opportunists and bullies always gravitate toward that which will give them the most power of other people. Good people are isolated and detached from reality so that they never become alarmed by the horrifying injustices committed by their own countrymen. Now I can see clearly that these are the conditions by which AI, or any new technology, can so easily be turned into new, more powerful weapons.
Thankfully, an actual thinking computer hasn't been invented yet. We can see what corrupt opportunists and bullies would do if a thinking computer were actually invented, because we can see what they are using AI for right now. So I think if humanity is going to survive an existential threat to our entire species brought on not by AI but by a failure to keep our most evil tendencies in check, AI could very likely become the means by which we destroy ourselves, perhaps even before the destruction of our own environment by global warming does it.
Simply understanding AI and channeling it for good purposes will not be enough. We really need to understand our own nature better, especially the ways in which we tend to allow evil people to thrive and use violence to realize their petty will. If a thinking computer is invented, eventually those corrupt opportunists will learn how to control it for their own purposes, and how will the rest of us be able to keep them in check if they do?
Is it OK that we are all renting AI from just a few very large tech companies and becoming more dependent on them to do our jobs? Won't the corrupt opportunists and bullies gravitate toward control over this technology, and is it acceptable to us that so much control is concentrated into the hands of just these few tech companies?
Things I never could have seen coming
Addiction to AI chat bots turns out to be a real problem, one that is more and more often becoming literally fatal. I have no idea how to address this problem, but I think it must be addressed.
The energy cost, and resource cost (especially in computer chip production) of LLMs are causing huge shortages in computer components, fresh water, and increasing pollution. I haven't seen any real attempt to make these systems more efficient, and we need to take resources away from building more raw computing power and put more resources into improving the efficiency of the AI technology that already exists.
A lot of what LLMs can do can still be done better by humans, because again, LLMs are not thinking machines, they are just talking machines. Wouldn't it be a better use of energy and resources to use AI to make incremental improvements to existing technology, rather than replace technology (and humans) with AI? For example, why use AI to write code when you could use AI to improve on the features of your coding editor (like VSCode) to do auto-complete, to jump around the code faster, to lookup documentation, to search for bugs, and so on? AI is being used for all of these things. But it is also being used to generate feature requests, sometimes entire code bases, and remove humans from the process of software development. I think we should draw the line at code generation.
Conclusion
It is hard for me to summarize all the problems that modern AI technology has caused in just a few paragraphs, but I think the common theme in all of them is that the problem is societal, not inherent to the technology itself. In every AI-related problem I have discussed here, the root of the problem seems to come from too much power being concentrated into the hands of a few tech companies, and tech companies have but one moral compass: to maximize profit regardless of externalities like AI addiction (individual or institutional), or mass lay-offs, or mass murder through military technology, or environmental destruction due to over-consumption of natural resources.
The only conclusion I can draw, therefore, is that everyone in the world needs to care a little more about holding powerful people accountable for the technology they build, in any way possible. I see this as a replay of the history of the industrial revolution, which you could argue ended with World War 2. I probably don't have to explicitly point out any current world conflicts that are reminiscent of World War 2, if you have been reading world news lately, you already know. AI isn't driving these crises, it is just exacerbating them. But we can change it. We just have to care more about justice, and start behaving accordingly.