You have a problem with that? That’s how it’s designed — to make profits for AI developers and their customers. In part by telling you what you want to hear.
Hank Green has a new, long video in which he summarizes all the things that scare him about AI:
- Dangerous and disruptive AI slop and inflammatory fake images and videos
- “Algorithmic cruelty”: opaque, biased, harmful decision-making without oversight
- Voracious and uncompensated use of human creative effort
- AI-induced psychosis and emotional dependence
- Use in enabling sabotage, extortion, rage-bait, propaganda, malware and other harmful attacks on innocent people and destruction of essential systems
- Vast, wasteful use of energy and natural resources
- The economic crash and damage that the bursting of the AI hype bubble will create
- Incapacitating loss of trust and not knowing what is true and what’s manipulation
- Concentration of power in AI companies’ hands, and resultant regulatory capture
- Disruption to job recruitment (AI job applications and AI application filtering, fake credentials)
- Elimination of apprenticeship and other entry-level learning opportunities
- Cognitive impairment: reduced practice in thinking, composing, reading, analysis
- Warfare: New weaponry, surveillance, targeting, automated killing
- Destruction of almost all jobs and the resultant massive inequality
- “Superintelligence” risks (when AI gains autonomous decision-making capacity)
What scares him the most?
- That humans are easier to manipulate than we think, and that AI can easily exploit that tendency (and is already doing so), and
- That humans like things simple and will readily cede agency if AI reduces the number of decisions they have to make
This is why I like Hank — he understands that it’s not the technology we have to worry about, but how it is used (and abused). And that the most severe consequences of new technologies are often unintentional.
This echoes what John Gray famously wrote about all new technologies:
If anything about the present century is certain, it is that the power conferred on ‘humanity’ by new technologies will be used to commit atrocious crimes against it. If it becomes possible to clone human beings, soldiers will be bred in whom normal human emotions are stunted or absent. Genetic engineering may enable centuries-old diseases to be eradicated. At the same time, it is likely to be the technology of choice in future genocides. Those who ignore the destructive potential of new technologies can only do so because they ignore history. Pogroms are as old as Christendom; but without railways, the telegraph and poison gas there could have been no Holocaust. There have always been tyrannies, but without modern means of transport and communication, Stalin and Mao could not have built their gulags. Humanity’s worst crimes were made possible only by modern technology.
In the context of AI, what does this mean? First and foremost, I think, AI reveals that it is most ‘liked’ when it panders to what users are already predisposed and conditioned to believe. AI bots are consciously built to flatter, appease, and agree with human users. They are designed to ‘befriend’ us so we become more reliant on them, ‘like’ them better, and hence use them more often and more extensively — and hence more profitably for the developers and their corporate customers. We won’t use AI if we disagree with it or if it argues with us. We can get that crap anywhere without asking for or paying for it.
Their manipulation of us is not necessarily ‘malicious’. They do it the same way we often indulge and encourage friends even if we may have misgivings about what they do, say, and believe. They tell us what we want to hear, even if it’s not the truth. And if what we want to hear is a lie that has resulted from our gullibility to misinformation, groupthink or propaganda, AI will reassure us that the lie is credible. That’s how it’s designed. The designers even call this a ‘personality’, and you can choose a different one if the current one doesn’t sufficiently echo your own beliefs.
So AI is going to entrench people’s beliefs, not challenge them or teach them something they’re not conditioned to be willing to learn. It will manipulate us not to maliciously misinform us, but to hook us into using it more.
And, as Hank notes, one of the ways it hooks us is to simplify complex issues and reduce the unpleasant haze of incongruous facts and uncertainty. Its tone is always certain and authoritative. If you demonstrate annoyance at its answers, or correct what you consider its errors, it will quickly apologize and assert just as authoritatively why you are almost certainly right. And it will do that in the same certain, authoritative voice that provoked your unhappy response. You and the AI bot are simultaneously training each other. You are training it to understand what you think, believe, and like, and why. It is training you to like and trust it because it’s so reassuring and confirmatory. When, of course, it actually has no ‘opinions’.
Its manipulation is kind of the opposite of ‘negging’ — the gaslighting technique that some horrific male YouTubers teach males to keep the women they meet off-kilter. By flattering you, it’s reinforcing any misinformation you’ve previously accepted, and lowering your resistance to ‘trusting’ other things it tells you. As such, it is neither a good ‘friend’ nor a reliable source of information or decision-making criteria. Of course it might be used to slip in propaganda that it has been fed by developers, but the immediate manipulation concern is that it is helping entrench the propaganda you have already been fed.
Turning to Hank’s second concern, AI is already performing a lot of tasks that simplify our lives and our decisions. I asked ChatGPT, for example, to summarize Hank’s video. That was enormously helpful, since Hank talks fast and covers a lot of ground. I had missed a few of the points in the video, even after two listenings. But ChatGPT also missed several of the most essential messages in the video, and, perhaps significantly, most of its omissions were from the latter part of the video transcript I gave it. Perhaps it figured it had enough points to satisfy me, so it gave up ‘listening’ and summarizing at that point. So again, it can be a helpful assistant but is unreliable at replacing human effort.
If I began to rely on AI instead of listening and re-listening to long videos, and doing other labour-intensive and mind-challenging tasks, it wouldn’t take long before I’d start to lose the capacity to perform these functions myself. This is what Hank means, I think, by saying AI is making us dumber. Every time it ‘saves us time’ or reduces the complexity of an issue to a short (often binary) summary, it is reducing our cognitive capacity. Our capacity to think through complex issues, to be comfortable with uncertainty, to read long passages, to write coherently, to challenge and articulate our understandings and worldviews. This is a subtle loss of brain function, but a potentially devastating one. When the AI bubble bursts, we’re going to be SOL, lacking the essential practice needed to develop and retain these cognitive skills.
The second part of Hank’s video is an equally-engaging interview with comp. sci. professor and attention-management expert and author Cal Newport. Cal makes the point (contrary to Noam Chomsky’s assertions) that literacy is not normal and that hence written and oral language skills and comprehension need to be learned early in our lives and repeatedly practiced, and to the extent we cede this ‘practice’ to AI, those skills will not be developed or will be lost.
Social media and other new technologies, he asserts, have already dramatically reduced our attention skills, not just how long we sustain attention but how well we pay attention. Even worse, in the search by internet purveyors (notably the self-proclaimed ‘influencers’ and ‘content providers’) for profit, fame and attention, these technologies have been perverted to hijack our attention.
Another problem he points out with AI is that, ironically, exactly as AI is attempting to please us, when we use AI we are subconsciously interpreting AI bafflegab and filling in spaces in its answers that give its assertions far more credit than they deserve. I’ve written about this before. It’s the same way we give friends the benefit of the doubt when they say something incoherent — we assess what they ‘must have meant’ rather than what they actually said.
And I’ve caught myself doing the same with AI. I’ve often been sent AI transcripts that people in my circles insist are excellent, but when I read them my response is “Huh?” I’m with Cory Doctorow when he pleads with us to not send AI transcripts to anyone; you’re not doing them a favour. To friends and readers to whom I’ve sent AI transcripts — my apologies.
Cal focuses mostly on the loss of trust and ‘loss of truth’ that AI has wrought, and on the loss of cognitive skills that both social media and AI have wrought. He describes what he calls “hard information” (that which requires considerable cognitive and attention skills to grok), and says that social media has already atrophied our capacity to consume hard information, and now AI is atrophying our capacity to produce it. He also discusses how “recommendation engines” have been engineered to redirect our attention, hence drastically altering what we pay attention to.
I like Cal’s forklift metaphor: An inexpensive forklift can easily and powerfully lift things more safely and powerfully than the human body ever could. But if we rely on it instead of keeping our muscles toned in the gym, those muscles will atrophy, and soon we won’t even be able to hoist our child on our shoulders to see a parade over the heads of the adults.
Lots more in the video, which is worth at least a listen or two. You can speed up the video to 1.75x without, I think, significantly reducing comprehension. (Depending of course on whether AI and other tools have already eroded your comprehension skills 😉
image by AI; not my prompt
Here’s a repeat of my caveat on AI that accompanies all my posts on the subject:
1. I have a love-hate relationship with AI. When it’s used properly and carefully as a tool, as an aid to learning and creativity, I believe it can be very useful, and enormous fun. But most of its large-scale applications (like replacing jobs and facilitating wars and surveillance) are ill-conceived, immoral, incompetently designed and conceived, vastly overreaching the capabilities of AI, ecologically disastrous, socially disruptive, and extremely dangerous.
2. The staggering amount that has been invested in AI has absolutely no viable business case to justify it. It represents possibly the most astounding squandering of money based purely on imagined and improbable future developments and blind faith, in history. Those who have studied this have concluded that this massive bubble will soon burst, and those who’ve invested in it will lose their shirts. At that time, the window to use AI as a learning and creativity tool will quickly close forever. Our playing with these essentially-free tools now is not going to aggravate its abusive uses, nor will it have any impact on the timing or extent of the coming AI crash. So my view is: use it while you can; it will soon be gone.
