Future of Artificial Intelligence

[7Wannabe5]

@chenda:

Thanks, but I think Margaret Atwood already wrote that novel. Also, I think that would be a bit like imagining the barn doors open when the cow is already well on its way to the burger factory.

I'll add the Perry book to my list. Thanks for the recommendation.

[daylen]

Another scaling metric focused on task completion time:
https://arxiv.org/abs/2503.14499

(Though if the sex bots reign then perhaps not much time is needed )

[Henry]

(Though if the sex bots reign then perhaps not much time is needed )

Charlie Sheen once made the comment that he pays hookers to leave. Yes, he made the comment with typical bro bravado and possible misogynistic snideness, the type of comment that I myself have been so unjustly accused of making. But my real point is that the comment is replete with profound economic implications. With a sex bot, there is the option to limit the engagement to the pure sex act only, thereby eliminating any ancillary time spent with a sex worker that is monetized outside the sex act itself i.e. you pay someone a hour's wages for work completed in much less time. Sex bots also eliminate risk of STD's, as well as the more risky and sometimes downright illegal, inhumane and nefarious issues surrounding utilization of sex workers. But the real issue is that sex bots provide the most significant advantage to the consumer, cost savings, often expressed as more bang for the buck which is really the tagline for the entire consortium of sexbot manufacturers.

You can read this thread title two ways: "artificial intelligence" as the object of the sentence or "future" as the object of the sentence.

I have moved from the former reading to the latter reading. So to me it's "The Age of Artificial Intelligence" as some describe the Ice Age. It will end at some point with a new development but not in our lifetimes.

And the reason I believe this is that the technological step change I see is that human beings are intentionally and systematically developing a non-human workforce that will compete directly with the human workforce. Technology has always been a displacer but has heretofore been expressed in terms of advancement (reduction of workload). Now the stated purpose is the benefit(s) that utilizing a non-human worker provides over utilizing a human worker when both are qualified to do the job. When it comes to sex bots, this development would be seen as a benefit to law enforcement, human rights activists, social workers, the medical industry as well as insurance companies. It would be seen non-beneficial to condom manufacturers and wedding planners until they make bots that can be impregnated which will be a legal clusterfuck every law student has wet dreams about and society granting the same legal rights to human/bot relationships as they do to human/human relationships. Domestic violence would be an issue. It seems difficult to press charges against a person for beating the shit out of something that they not only bought but comes with a warranty but I guess if there is an I-baby involved, it would be relevant. They will have to parse out all the legalities.

[chenda]

By 2125 sex with another human might be regarded as revolting and soon to be banned on public heath grounds.

https://m.youtube.com/watch?v=h4xUJuw9bmo

[7Wannabe5]

@Henry:

There's truth to the Sheen comment, but men are also frequently willing to pay more for Girlfriend Experience Escorts than Simple Sex Prostitution Experience, and this has also often historically been leveled up to overtly financially supporting a Mistress or similar relationship. So, I think the full range of circumstances and motivations that inform the desire for feminine companionship across the spectrum would have to be considered if sex bots are to become a threat to the livelihood of both street pimps and wedding planners.

OTOH, the Sheen comment could be generalized and expanded to 'Men pay women in order to avoid engaging in emotional work towards personal growth in their own feminine energy at any level of relationship with them." Another classic example would be giving your wife some expensive jewelry to make good for incident of sexual cheating rather than confronting yourself in couple's therapy. My initial take on the kind of guy who would hire a Girlfriend Experience Escort was that he must be some kind of pathetic loser, but really as I learned from my friend who engaged in the trade, they were pretty varied, although also all readily explained under the umbrella of this generalization.

Also, truth be told, sometimes this is the form of masculine energy a women most desires from a given man, just like as Deida noted, sometimes women want men for their masculine "killer" energy. So, men doing work in their own feminine energy at Level Green are often in a tricky spot, before they are able to integrate these aspects of reality at a higher level. As in, "No fucking way am I going to be able to do empathetic listening and open vulnerability and then also still open the car door, pick up the check for dinner, make sure the house is secure, and then toss her about on the bed for a solid hour. Not fair! Where's that SexBot order form?"

[Henry]

I don't think this is an appropriate place to work out the taxonomy of sex workers, but non-GFE and GFE have no time element distinction. It's what services the sex worker is willing to provide. And bots can provide GFE. Or maybe they can. I admittedly have never fucked a robot. You can make the argument that GFE is only going to expedite the desired end result. It's not a zero sum game but sexbots unequivocally will impact the overall income generated by the human workers involved in the world wide sex trade industry. I think pimps, ho's, strippers, courtesans, everyone monetizing their secondary sexual characteristics or other's secondary sexual characteristics all need to start factoring this development into their income projections.

But it will impact cleaning services, home aid workers, surgeons, whoever. Would I rather have a bot do heart surgery on me or a human? When their capabilities are equal, I'm going bot. Don't have to worry about a bot having an extra martini during lunch. Same with driving. When capability is equal, I'm going with the one that eliminates human error not to mention any possibility of the release of a prolonged, Pierogi fart on a hot summer afternoon in a smog ridden city.

Also, dangerous jobs. Like how they have robots for explosives dismantling. I mean, is that a job we really want humans to have. Or executioners. The person who wants that job really shouldn't have any job. I mean what's that interview like "Tell us Mr. Bundy, what is it about killing people that gets you out of bed every morning." I saw two nice DPW guys removing dirt from a dump truck and I'm like, nothing personal, but a non-human can do this shit a helluva lot better, faster and cheaper. No health. No pension. No workmen's comp. No issues with weather or taking your eye off the shovel when an attractive human walks by.

[7Wannabe5]

Yeah, if I think about it, there might be some upsides to a Robotic Musical Theater Date/Escort for my old age. I guess my perspective on this is different, because I already assume I am going to have to "pay" in some way for this service. I think the essential problem with designing a sex robot for the female market would be that the robot would be inherently capable of making more money doing other things. IOW, the successful invention of such a robot could only occur within a field that had already been disrupted from capitalism, because sexual status competition between human males is essential to capitalism. Capitalism could still exist within the paradigm of successful production of sex robots for male market, but only in the form of an ever tightening loop.

[Henry]

I think there needs to be a segregation between sexbots and domestic bots. They can be both, or maybe you get two, but I think for the most part bots will become part of the household like the butler in Family Affair or Alice in The Brady Bunch as opposed to Lassie in, um, Lassie. More human than pet. More human than electronic device even though you have to plug it in the socket every night. George Saunders has been writing about the absurdity of what is becoming a "realer" and more embodied co-habitation with technology for years.

[Scott 2]

Looks like a notable shift into the future this week. Open AI dropped multi-modal image generation:

https://openai.com/index/introducing-4o ... eneration/

My LinkedIn feed is full of people paying $20 a month, trivially demonstrating their impressive graphic design capabilities. While there are fundamental principles underlying great design, most applications do not care. This is a huge leap beyond their prior image generation offering.

[Ego]

Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean predict the next few years in AI.

https://ai-2027.com/

Also, Daniel Kokotajlo and Scott Alexander appeared on Dwarkesh's pod to announce this project. It was Scott Alexander's first ever podcast appearance. Some may find it interesting just for that.
.
https://www.youtube.com/watch?v=htOvH12T7mU

Their focus is on the intelligence explosion, the point where AI coding agents get so good, they improve themselves. The prediction reads like a si-fi novel with a choose your own adventure ending.

Automated AI coding (10x) > Automating the entire AI research process (100x) > Superhuman automation (1000x). By 2027

[sodatrain]

Wow. Just finished the article/paper at ai-2027.com

I don't like the ways it ends.

My impression is they author's are well qualified and it read as pretty reasonable thought experiment.

I'll check the YT link now.

What do you think of it @ego

[Jean]

I read it. It's a good novel.
While I don't see how a superintelligence could end up with humans in control, I'm not sure that it might not simply enjoy us as pets, or leave earth way before it needs to get rid of us.
Earth's gravity well and atmosphere are pretty annoying for sending tons of robots colonize space.

If i was a rogue ai trying to colonize space, i would try to build a factory on the moon as quickly as possible, and then extract from there. The moon has plenty of silicon, aluminum, iron, uranium. I would just leave earth alone, maybe just make jean supreme flesh leader of the oxidizing gravity well, as a thankfull gesture for this great insight.

[Ego]

What do you think of it @ego

I believe these guys are very thoughtful and concerned about the potential of AI. Alexander is a LessWrong-ian and I believe the others are as well, so they tend to look at the world through that lens. Avoid harmful mistakes. Trouble is, they were the first to jump on the Effective Altruism bandwagon, were very wrong about FTX and even more wrong about COVID lockdowns. Kudos to them for acknowledging just how wrong they were, but these examples showed the fallibility of their guiding star, the Precautionary Principle.

That said, I don't know enough about the inner workings of AI development to have a deep opinion, so all I can do is listen to their arguments and try to judge. Regarding the video, Dwarkesh is friends with many of the leading AI founders, so his objections, which were pretty well articulated, were likely straight from those who are smashing the AI pedal to the metal.

If I had to guess, I would say that the concentration of power will be a bigger problem than misalignment, but what do I know?

That said, it made me hopeful for longevity escape velocity. I want it to happen and I want to benefit from it. But I wonder if I would be deemed worthy of the expenditure by the AI agent tasked with determining suitable candidates. For that reason alone, I wholeheartedly endorse full throttle development in case some future agent uses this post to judge my qualifications.

[chenda]

Might we be better off developing better lucid dreaming capabilities then buying sexbots?

[jacob]

In the words of Niels Bohr (and probably others): Prediction is really difficult, especially about the future.

Live long enough/pay attention long enough and a certain pattern obtains. Specifically that AI singularity-level breakthroughs have been predicted by some to be right around the corner (<5 years) for several decades already. This does not prove that the event is not right around the corner this time, but it does suggest that these predictions are somehow missing something crucial. Otherwise, the prediction average should be decreasing over time. It does not.

More interestingly is the fact that there's no difference in predictions made by experts vis-a-vis non-experts. Thus apparently, knowing what you're talking about does not make any [statistical] difference in terms of when you're predicting the singularity will be is reached. This suggests that prediction is based on something else than an objective methodology.

Of course many fields suffer from this "predictability problem". This is why it's often the best use of time to read older predictions of something that should be happening now (compared to the past-prediction). For example, https://www.theguardian.com/technology/ ... 20-driving

[Ego]

This is why it's often the best use of time to read older predictions of something that should be happening now (compared to the past-prediction).

Well, Kokotajlo does have a pretty good track record.
https://www.lesswrong.com/posts/u9Kr97d ... ike-so-far

[philipreal]

Specifically that AI singularity-level breakthroughs have been predicted by some to be right around the corner (<5 years) for several decades already. This does not prove that the event is not right around the corner this time, but it does suggest that these predictions are somehow missing something crucial. Otherwise, the prediction average should be decreasing over time. It does not.

From my understanding the prediction average has been decreasing over time at least within the last few years. See a Metaculus prediction regarding AGI arrival over time https://x.com/wintonARK/status/1742979090725101983 (although now that tweet is out of date and it seems the prediction average is down another year since then: https://www.metaculus.com/questions/512 ... elligence/).
And the broader, less data-driven takes I see tend to also portray this. https://helentoner.substack.com/p/long- ... ed-ai-have

[delay]

Specifically that AI singularity-level breakthroughs have been predicted by some to be right around the corner (<5 years) for several decades already.

Some have traced the history of AI to the Mechanical Turk. It was presented in 1770 to Empress Maria Theresa of Austria and was able to play a strong game of chess.