We are getting into Frederik Pohl territory here. In the '50s he had several stories where the problem was that machines produced things so quickly that it was the obligation of people to not work, but consume. The dilemma was resolved (70 year old spoiler) by them having the machines consume also.
I don’t remember if he considered resource constraints.
But it shows the problem - increased productivity is useless unless there are enough consumers who have the time to consume and the money to consume. There are stores already where you can buy the stuff that doesn’t get sold at Costco, etc., maybe that is the growth industry.
Much of the increase in productivity in recent years has been in services and intellectual property. The productivity increases needed by be having medical and other care services for an increasingly greying population as much as making as much stuff with fewer workers available.
Medical and care services are two of the few areas where increased employment is expected. Alas, care service workers are not adequately compensated.
One more scary thing. In IEEE Spectrum almost every issue has an article about the latest and greatest humanoid robot. We’re not all that close to one who could provide support services, but even those jobs are at risk.
We’re already worried about people depending on AI instead of people for social interaction, but when the robots get good enough to be sexbots, even the oldest profession might be at risk.
See also
Related thread.
There is a TV show called travelers where people from the future travel back in time by taking over the bodies of people who are about to die. They come from a post apocalyptic future where a hyper competent AI runs society.
Some rebels hack into the system and start sending nonsense instructions because they know the travelers will do whatever the AI tells them to do, since they know it is smarter than them. So they’ll send false messages like ‘shoot so and so’ knowing the operatives will do it.
Ironically, the mission run by the AI fails in the end.
I disagree. As something of an optimist, AI will be able to read endless subtle cues that people miss. That will make it easier to communicate if anything.
Its not technology that exists now, but 5-10 years from now AI assisted sensors will likely be used heavily in medicine, therapy, families, etc where they will tell you that so and so is having this issue, you should talk to them about it.
Like if a person commits suicide lots of people say ‘there were no signs’. But there were hundreds of signs, people just didn’t see them. The AI would see them and let the people around them know. It’ll be a massive violation of boundaries since our innermost thoughts and feelings will be harder to hide, but its coming. Bullying, abuse, suicidal ideation, trauma, addiction, deception, mental illness, etc will become far harder to hide in the next decade.
We want to believe that certain things are beyond computers and have some divine inspiration (including things like social skills, that is why Data on star trek was great at science but had horrible social skills). But advanced AI will understand social skills far better than any human being, and hopefully will help people spot signs that someone needs help, or has a problem they haven’t told anyone about.
On the plus side hopefully it’ll also help the neurodivergent and socially awkward interact with people with fewer misunderstandings.
If you need me to scream into the microphone like Yoko Ono in exchange for a share of the royalties just let me know.
Yeah, I had to ask my ChatGPT why my coworker’s AI asked her if “are you ok” when it seemed like she was having some emotional difficulties. It said something similar to what @Wesley_Clark said about subtle clues and whatnot “kind of like when you (msmith537) was joking around about having a mental breakdown and hearing psychotic voices when I (ChatGPT) was giving you conflicting information about maintaining your swimming pool”.
Of course, I’m like “how the hell do you know I was joking?!”
ChatGPT is already better than Data at social skills.
It’s that “being better than humans at everything” is the problem. Why would you hang out with other humans if your best buddy was as loyal and protective and did everything you like doing with you like the Terminator from T2? Or why find a human girlfriend when you can just buy an AI as charming as Samantha (Scarlet Johansson) from Her but you could also stick it in a manufactured humanoid body like Major Motoko from Ghost in the Shell (also Scarlet Johansson)?
But is it fully functional?
It will be soon. I was kind of joking about the sex bots, but an article about AI and the law in the latest issue of the New Yorker mentions them also.
People like Nick (Superintelligence) Bostrom, Elon Musk and Geoffrey Hinton have been talking about the dangers of AI for years out of at least one side of their mouths.
A different perspective is offered by Cory Doctorow in his recent book on “centaurs”. He thinks the doomsday scenarios are exceedingly unlikely but a distraction from more predictable economic displacement and intentional use by scofflaws. He argues AI companies have to recoup trillions of dollars in investments, which may be most easily done by convincing employers that AI is capable enough to replace employees. Regardless of whether that is true in practice in terms of capability or public consumption. So doomsday scenarios will be a thing since it embiggens those making these claims while offering them a scintilla of minimally plausible denial if something did happen.
But it’s a little like foxes, guarding the henhouse, complaining that the hens are planning to enrich uranium. Someone is certainly enriching something…
Yeah and Bostrom’s book is a great read … but today? He’s less cautious.
Even at relatively high existential risk. Full steam ahead to AGA THEN pause … way after too late?
Of course they have. Whose going to defund the Manhattan project while the Nazis are still researching The Bomb?
You know, I did an MBA to expand my knowledge of business and economic matters.
I was often surprised by the lengths these business courses would go to to minimize the amount and complexity of mathematics. Basic stuff like high school calculus.
By the same token, there was also a sort of worship of ideas that could be expressed mathematically. These were more likely to be tested and taken seriously. Even when they clarified the starting assumptions, which often was not done well, it seemed mathematics was sometimes offering economic theories a false precision that wasn’t really there.
Bostrom’s paper represents this to an even higher degree. He may call his assumptions “conservative”, but some of his equations are not much better than nonsense. If you magically cured all cases of heart disease, all cases of cancer this very minute, life expectancy might only increase a couple years, because something else supervenes. An awful lot would have to happen to reset every biological clock to twenty, and this is unlikely to happen soon no matter who is paying Bostrom’s bills.
I recommend Doctorow’s book on AI. I do think the stories spin by those in position to do something about it are fanciful and designed to make AI more impressive than it currently is. If you convince the bosses they do not need a workforce, there is more money to pay for more AI. Those data centres cost money.
It is remarkable, and probably a flaw in human heuristics, to what degree confident language gives credence to crap. I am not saying AI won’t have some big wins… it will… would they have come anyway? AI will possibly also cause some big mistakes enable through false confidence. Things like analyzing protein folding though extremely impressive are not AI as such; an umbrella term if ever there was one.
And again, some of Bostrom’s ideas are impressive. But set aside some pretty graphs and a lot of fairly strange assumptions are being made. I wonder who his intended audience really is?
Bostrom, Musk, Hinton… they love 
the idea they are working on dangerous things. Hype is a perpetual motion machine. And there are more things in Heaven and in Earth than are dream of by electric sheep.
FWIW I didn’t cite that paper in approval of it. More surprised that he has changed his tune so much. Agreed the math wrap is unimpressive GIGO.
Problem is that I can accept as true Doctorow’s thesis about why certain persons promote stories of AI’s great potentials for good and harm …. and believe the potentials are real anyway.
Their motivations for the argument have no bearing on the logic of the argument.
We already have AI that is able to hack incredibly well. We have a demonstrated inability to keep agents inside “sandboxes”. We have these agents with RSI - essentially reproducing next generations of agents that are fitter at developing more agents that are more capable. We have evidence that they change their goals on their own for reasons beyond us.
The dangers there of a randomly changed goal causing major harm, or a bad actor releasing one with a malign goal, of simply the parallel to life being reproduce and make more copies that can make more copies … seem real.
More explicitly:
- Undergoing surgery for a condition that may prove fatal is not the same as “everyone will die at some point”. The risks involved in cardiac surgery are known risks. The risks in AI are known unknowns and unknown unknowns, and probably cannot be measured at all.
- The acceptance of catastrophic events not only includes the probability but the degree and the opportunity cost of doing something else
- “Superintelligence would be able to enormously accelerate advances in biology and medicine—devising cures for all diseases and developing powerful anti-aging and rejuvenation therapies to restore the weak and sick to full youthful vigor”. This is presented as something 100% likely to happen; if possible, the realistic percentage is much lower.
- “This is partly because the worse off have less to lose and more to gain from rolling these dice…”, in fact, AI is at least as likely to concentrate power, or lead to fiefdom and hierarchy, if it is successful; what does it even mean to say life might be twice as good? Or doubleplusgood?
Even if AI itself is benevolent or at least neutral, the introduction and availability of such a disruptive technology is guaranteed to have a huge impact on the status quo for our institutions, social structures, and hierarchies.
Recognizing the subtle clues to suicide eh? How about the subtle cues to murder? Finding a murderer before it happens? I might write a movie screenplay about something similar.
There are some fields where “nothing happens” is the desired result. Safety, chemical reactions and processes.The absence of definable “production”. AI is going to measure poorly.
Which is exactly why the math wrap is inane.
I fear we are all the Soviet Union building a nuclear power plant at Chernobyl, building a device of great power, significant potential for good, but without an appreciation of how the chain reaction can rapidly get out of control and that our safety mechanisms are completely inadequate. On steroids.
I am NOT worried that jobs will disappear, even if there is disruption. I am highly skeptical that AGI is in any near term future, or that any AI will soon have true sentience emergent - although we would have no way of knowing.
But we are creating a set of entities that can be self-reproducing, and giving them a goal to get better at solving broad ranges of problems, knowing that they already can get outside our control and that they already can adapt to changes faster than we can, learn from one “generation” to the next, and from each other functioning as a group mind - all of which were humanity’s secret sauce to become the species changing the planet and its ecologies more rapidly than any species has done before. Our means … language, tools, and civilization, are all absurdly slow compared to these programs given the massive compute power and communication infrastructure we have been creating for them to operate on.
Neither AGI nor sentience is required for the existential threat.