I’ve written very little on my blog about so-called Artificial Intelligence, an old term applied to a relatively new technology, Large Language Models (LLM). I don’t think I have anything new to say about the subject, and my first-hand experience with LLMs is negligible, since I eschew them. Fortunately, my employer does not force me to use AI. (I’m self-employed!) The benefit of writing a blog post expressing my opinion about AI is that I can refer to it in the future when someone—some idiot—inevitably, unjustly accuses me of writing “AI slop” based on dubious evidence such as my careful proofreading or my punctuation—the em dashes. (I recently saw a commenter highly ranked on the Hacker News “karma” leaderboard label an article as AI slop before realizing that the article was published in 2018.) I honed my writing skills decades ago in academia, an unwitting source of LLM training materials. The only significant previous discussion of LLMs on my blog was the last paragraph of one post: If computers are the future, why are computer users expected to be permanently illiterate?
The latest front in the war against power users is Artificial Intelligence. The promise of AI appears to be that you’ll never have to learnanything. Don’t know something? Just ask AI for the answer. Can’t do something? Just ask AI to do it. Ignorance is bliss. Laziness is encouraged in the name of efficiency. The prospect that one may have to work and struggle to achieve one’s goals is considered abhorrent. The very notion of self-improvement becomes obsolete. Life under AI is a video game, pure joy… as long as you continue inserting tokens into the machine. Hopefully my video arcade metaphor hasn’t become obsolete too. A more contemporary example: playing guitar, which practically anyone can do with practice, is replaced entirely by playing Guitar Hero. I’m sure that there are Guitar Hero heroes, people who have mastered the game, but could a single one of them ever write a song? I guess we have to ask AI to write our songs in the future. The public debates whether AI will eventually become as intelligent as humans, or even super-intelligent, when I think the relevant question is whether humans will eventually become as dumb as AI, or even super-dumb, as in Idiocracy. I fear that none of this will end well, except for the computer and LLM vendors.
I stand by my earlier statement. However, this objection to AI is just the tip of the iceberg. I fear that AI will contribute to the melting of the icebergs, because training and disseminating AI appears to require an inordinate amount of resources and power, in a world where unrestrained industry has already polluted the atmosphere and destabilized our climate, our ecosystems. I find it ironic that AI, which promises to make computing easier for the masses, has triggered mass shortages of computing components, thereby making it more difficult for consumers to obtain personal computers. AI is also making electricity more expensive for consumers. The LLM vendors have perpetrated mass theft of intellectual property, aided by regulatory inactivity, itself a product of political corruption financed by the LLM vendors. Business leaders unabashedly state, celebrate, that AI will allow them to automate away the jobs of human workers. The workers who haven’t yet been laid off are becoming deluged and burned out (to mix my metaphors) by AI-generated busywork and insatiable management demands for increased productivity. Many observers of the stock market describe AI as an investment bubble, but I would describe it as an investment black hole, a singular mania that exerts a gravitational pull sucking up economic activity. AI all the things! I doubt that a burst of the bubble is imminent; after all, the bubble has not burst for cryptocurrency, the grossly inflated long-term investment value of which was based nominally (delusionally) on the expectation that everyone would adopt cryptocurrency for short-term spending. Markets can remain irrational, yadda yadda.
Given the shallowness of my personal experience with AI usage, I try to avoid wading into the ubiquitous debates about its personal utility, focusing my critiques instead on the obvious societal consequences of AI. When I’ve tried ChatGPT a few times, it has produced both truth and lies (euphemistically, “hallucinations”). I still write all of my source code the old-fashioned way, manually typing into (older versions of) BBEdit or Apple Xcode. Believe it or not, I also write all of my blog posts with BBEdit, in raw HTML format! I’ll talk more about the production of software momentarily, but as a consumer of software I’ve seen no indication that AI has benefitted the industry. So many software developers claim that AI makes them so much more productive, but I honestly haven't noticed or enjoyed the results of this alleged productivity. As far as I’m concerned, the enshittification of software has continued unabated.
Devotees of AI embrace its speed and (supposed) efficiency. Indisputably, AI can produce source code at an unprecedented rate, like a hare racing a tortoise. I’ve always felt, though, that speed is a vice rather than a virtue with regard to software development. I’m a longtime critic of software release schedules, as explained in my blog posts about the myth and reality of Mac OS X Snow Leopard. Software development is an art, not a science; if your applications are merely cookie-cutter CRUD, then why are you reinventing the wheel rather than purchasing an off-the-shelf component? Innovative software is complex and unpredictable. It’s ready when the engineers have solved the problems, not on some arbitrary date preannounced by a glory-hungry CEO. Software evolution should decelerate rather than accelerate as it matures, because the odds increase that any given change to the software will break rather than improve it. When software is new and immature, the avenues for rapid improvement are obvious and manifold. In contrast, mature software is battle-tested, has fewer obvious shortcomings, long-established user expectations, and accumulated complexity. You alter it at your peril. Some overeager software developers feel that this is a downside to mature software, preferring rewrites or blank slates, but my feeling is, why would you mess with near perfection? If you’re close to the summit of a mountain, there’s not a lot of altitude to gain, but there’s a long way to fall.
AI is indefatigable. If you have the tokens to spare, that is. AI does some work that humans wouldn’t otherwise do, which is assumed to be a benefit of AI. My question is, why haven’t humans done that work heretofore? If there was money to be made, then humans had financial incentives to do the work. The story we’re told is that the work was too costly before to pay for human workers, but now it’s cost-effective with automation. I’m skeptical of the story, because if a product is valuable to consumers, then it should generate enough revenue to pay for workers to produce it. What we appear to be discussing are products so worthless that they would never get produced unless production is practically free. For example, news reports have claimed, and if I recall correctly, Apple itself has confirmed, that the number of apps submitted to the App Store has greatly increased recently. Nonetheless, App Store downloads and revenue have barely budged. Everyone is making vide-coded apps, but nobody is buying them. In my opinion, that’s productivity theater. On the other side of the coin, highly profitable monopolistic corporations are also employing AI. Yesterday I read about bugs fixed by AI in Google Chrome. I wonder, though, why wasn’t Google employing more engineers to fix bugs in Chrome before LLMs appeared on the scene? The bugs existed, right? (Or did AI write them too?) Cynically, I have to assume that Google’s corporate priority is to release endless new features while placing little economic value on fixing bugs (Promotion Driven Development). That’s why Google is now willing to do the work of fixing a huge number of bugs, because that work has become cheap enough to be a financial afterthought, not undercutting the company’s obscene profits. In other words, I’m not buying the story that a historically huge entity like Google lacked the (human) resources until now.
I’d like to end this blog post by going back 25 years, to 2001. (Pedantically, it was 1968, but please grant me some artistic license.) Dystopian science fiction predicted most of our terrible technological fact, and we know that our terrible technological leaders read that science fiction, yet they either ignored the morals or, more likely, identified with the villains. In 2001: A Space Odyssey, HAL 9000 was an extremely clever AI. Plus, HAL was a great conversational and chess partner for Dave and Frank, guys who were lonely and isolated. (Wow, doesn’t this sound eerily familiar?) HAL was so clever, accomplished, and friendly that everyone came to trust it/him. The Discovery team entrusted HAL with complete control over the ship and thus the lives of the crew. Spoiler alert: this did not end well. HAL was clearly intelligent, but was HAL as intelligent as the human crew? No, HAL was basically an idiot savant. HAL decided that the best way to resolve inconsistencies in the prompted mission was to murder the human crew, an egregiously stupid mistake that the humans themselves would not have made. And it turns out that Discovery still operated fine with a human captain, Dave Bowman, after deactivating HAL’s higher functions, so in a sense, HAL was ultimately correct. The problem was caused by human error, placing too much trust in AI, allowing AI to have free agency and run wild. It has always been due to human error.
Is AI “just a tool” as so many defenders claim? The film 2001 includes one of the most famous cuts in cinematic history: after using a bone to kill both food and enemies, our fictional prehistoric ancestor throws the bone into the air, and the bone becomes a bone-shaped spaceship. The bone was indeed a tool, but according to filmmaker Stanley Kubrick, the bone was not just a tool. The bone changed the course of the entire species. Although this was sublime entertainment, I don’t actually subscribe to the theory that violence is what made humanity special. On the contrary, I think that nonviolent cooperation and especially language was the basis of civilization and science. (Sadly, not an ungrunted word occurred in that opening 2001 scene.) In fairness to Kubrick, I’ll admit that the Cold War accelerated the space program specifically, for better or worse. However, we mostly gave up after planting the American flag and hitting a few golf balls on the Moon. There’s still no Howard Johnson’s hotel in space. In any case, you’re sticking your head in a cave if you think that worries about the societal effects of AI can be dismissed by throwing your hands (and bones) in the air with a “just a tool” response.