Comment by jerf | Hacker News Reader

Comment by jerf | Hacker News Reader

So the piece exhorts us a couple of times to try to just think neutrally about this stuff, and I get the point, but at the same time, can anyone in 2025 think that it's just going to be best most altruistic people armed with highly persuasive AIs who just want to use those AI to persuade us to act in our own best interests, rather than the people who own and are running the AIs?

Like, I have a hard time even strawmanning such a position. Of course people armed with highly persuasive AIs will task those AIs into doing what is best for the people who own the AIs and the odds of that happening to line up with your own interests are fairly low. What the hell else are they going to do with them?

But then, keep gaming it out. This particular thing isn't exactly completely new. We've seen bumps in persuasiveness before. Go back and watch a commercial from the 1950s. It's hard to believe it would have done a darned thing, but at the time people had not yet had to develop defenses against it.

We had to develop defenses. We're going to have to develop more.

What I forsee as the endgame is not that we become mindless robots completely run by AIs... or, at least, not all of us... but that we enter into a world where we simply can't trust anything. Anywhere. At all. How does any being, human or otherwise, function in an infosphere where an exponentially large proportion of it is nothing but persuasion attempts, hand-crafted by the moral equivalent of a team of PhDs personally dedicated to controlling me? Obviously humanity has never had a golden age where you could truly just trust something you heard, and we can argue about the many and sundry collective failures to judge the veracity of various claims, but it's still going to be a qualitative change when our TV programs are designed by AIs to persuade us, and the web is basically custom assembling itself in front of us to persuade us, and all of our books are AI-generated to persuade us, and literally no stone is left unturned in the ever-escalating goal to use every available channel to change our behavior to someone else's benefit.

When does trying to become informed become an inevitable net negative because you're literally better off knowing nothing?

What happens when the bulk of society finally realizes we've hit that point?

Here is my Bayesian version of this: If you have lies coming at you in high enough volume, you cannot update your priors at all, or else you will eventually come to believe the lie.

But then you have the problem: If you won't update your priors, and neither will someone else, but they have different priors than you, how can you talk to them?

But I'm maybe a bit less cynical than you. I think (maybe I'm kidding myself) that I can to some degree detect... something.

If someone built a parallel universe of false journal articles written by false experts, and then false news articles that referred to the false journal articles, and then false or sockpuppet users to point people to the news articles, that would be very hard to detect if it was done well. (A very persistent investigator might realize that the false experts either came from non-existent universities, or from universities that denied that the experts existed.) But often it isn't done well at all. Often it's "hey, here's this inadequately supported thing that disagrees with your priors that everybody is hyping". For me, "solidly disagrees with my solidly-held priors" is enough to give it a very skeptical look, and that's often enough to turn up the "inadequately supported" part. So I can at least sometimes detect something that looks like this, and avoid listening and believing it.

I'm hoping that that's enough to avoid "When does trying to become informed become an inevitable net negative because you're literally better off knowing nothing?" But we shall see.

It's the team of PhDs dedicated to me personally part that gets me.

In the current world, and the world for the next few years, the amount of human and CPU time that can be aimed at me personally is still low enough that what "real reality" generates outweighs the targeted content, and even the targeted content is clearly more accurately modeled by a lot of semi-smart people just throwing stuff at the wall and hoping to hook "someone" who may not be me personally. We talk about PhDs getting kids to click ads, and there's some truth to that, but at least there isn't anything like a human-brain-equivalent dedicated to getting my kids, personally, to click on ads. I have a lot of distrust of a lot of things but at least I can attack the content with the fact that needing to appeal broadly still keeps the content somewhat grounded in some sort of reality.

But over time, my personal brainpower isn't going to go up but the amount of firepower aimed directly at me is.

The good news is that it probably won't be unitary, just as the targeting today isn't unitary. But I'd like something better than that. And playing them against each other gets harder when the targeting becomes aware of that impact and they start compensating for that, because now they have the firepower to aim at me personally and do that sort of compensation.

If I may, I’d suggest reading the latest Harari book on this topic, Nexus. Great read with interesting ideas.

Zvi's an AI-doomer, so he sees the endgame as AI becoming smart enough that they don't need us anymore and then killing us to take our stuff, but your scenario is also pretty bad.

Speaking in generalities, some source of "doom" are really covert marketing, where Thing X is so amazing and magical that nobody can risk not investing in it, paying attention to it, or constantly talking with their friends about it.

I don’t think you can really accuse the lesswrong crowd of being in it to hype up OpenAI.

They don’t profit from it (well at least the current doom crowd), they have been saying that this is super-risky since “forever” (15 years), they strongly argue against e.g. OpenAI going private … I really don’t understand where this whole strand of thinking of hidden ulteriour motives for AI doomers comes from, AFAIK there was never a single clear case where this happened (and the arc of people invested in AI is they they become less of a doomer the more financially entangled they become with AI, see Elon or Altman).

It's so common for people to throw out those accusations, despite zero evidence from the lesswrong crowd, because they need explanations that are way more comforting than facing down the scary implication that lesswrong "doomers" are right.

One thought path leads to business-as-usual comfort, happily motivated by trillions of $ in market cap and VC.

The other is dread.

It makes sense why people take the easy road. We're still battling climate change denialism after all.

They could be be playing into such a narrative, even though they have different motivations. Certainly Altman has had a few moments where he is saying "this stuff is getting so good it is society-scale dangerous". His motivations are to create FOMO with investors and users, and to try to form the regulatory landscape in their favor.

I just don't understand the line of reasoning at all. It sounds to me like postulating that oil executives have started admitting climate change is real in order to create FOMO. Investors and users live in society, why would they fear missing out on destroying it?

> like postulating that oil executives have started admitting climate change is real

The subtext isn't "our product has bad side-effects and can be avoided with known alternatives", but more like: "Our product's core benefit is too awesome at what it does; If we don't pursue it then someone less-enlightened will do it anyway; Whether you love it or hate it the only way to influence the outcome is to support us."

So the oil-executive version would be something like "worrying" that petrochemical success will quickly bankrupt every other power-generation or energy-storage system overnight, causing economic disruption, and that eventually it will make mankind too satisfied and too comfortable so that it falls into existential torpor... But at least DinoCo™ has a What If It's Too Awesome working group to study the issue while there's still time before our inevitable corporate win of All The Things.

Why would the oil executive benefit from saying that if it's not true? Wouldn't it be better for his stock price to say that disruption will be limited, and the DinoCo takeover will usher in a golden age like Amazon has for shopping? If anything, that's my criticism of the guy; as OpenAI revenue has risen, he's become much more vocal about the benefits of AI and much more confident that the risks and costs are solvable.

If enough people believe it is true and act on this belief, it will become true (to a larger degree, at least). There is a massive shift of the direction of money happening right now, away from workers into "AI" (LLM based systems). The more of that shift can be produced now, the stronger the AI companies will be in the future. Because they need massive capital investments in compute, engineering investments in efficiency, and to reach economies of scale, to be able to really compete with human-based workforce.

Yeah it's partly a "this is the inevitable winning team" - join before it is too late. The earlier the better! There are some parallels to how MLM and ponzi schemes are sold.

One of the ways it can work is that people may reject the most extreme case (say end of humans beings in control), but accept some milder version (most jobs will be done by AI, or most jobs will be AI assisted). The fear of an extreme can cause people to rationalize the "milder" outcome - independently of whether there are good arguments for that outcome, or even whether the outcome is desirable, or better than status quo.

The investor class are not dependent on wages, so their livelihoods are not at stake. Same with big corporate partners, they are hoping to improve competitiveness by having fewer employees and CEO take a bonus for that. Regular users in fear of their jobs may act on that, in hope that they can reskill and transition by being AI savvy.

I do not agree with the argument they make, but I understand what they are are playing at, and that unfortunately it can be effective.

Another aspect is for distraction. OpenAI for example have had periods where they talked a lot on existential risk (humanity is doomed due to AI overloads, yadda yadda). This can serve as a very useful distraction to avoid talk about more plausible risks in the foreseeable future - such as massive concentrations of wealth, increased inequality, large amounts of unemployment, large scale personalized psychological manipulation, etc.

I'm hoping that's intended as: "That group exists, but for clarity I want to say the lesswrong crowd aren't part of it", as opposed to my initial reading of: "Your 'generality' is a false covert accusation, you cad."

I think you get to the point of authoritarianism. A trustless society is not a functioning one, unless people are forced to function.

Reply on news.ycombinator.com