Ask HN: Does Anyone Want AGI?
There is always a lot of discussion about Artificial General Intelligence (AGI) and how close we are to achieving that. However, I have so far not seen anyone put forward a convincing argument as to WHY we as citizens of the world would want such a thing.
More advanced machine learning and statistical techniques can help us automate difficult/boring tasks and manage limited resources better, but these do not require AGI.
Can someone convince me how AGI would be beneficial for the world, beyond being scientifically interesting?
49 comments
[ 3.1 ms ] story [ 99.0 ms ] threadAGI could be the ultimate tool to free every human being from toil. It could also be the starting point of a large number of evil genie scenarios where we get what we asked for, in the form we least want it.
From a morality standpoint, we can't force AGIs to work for us. We also can't restrict their ability to self-evolve.
If we can resolve those conflicts in such ways we can coexist in peace with an intelligence that'll in all likelihood quickly surpass ours and partner with it, I'm in. If we build it and we can't resolve that, our opinion doesn't really matter.
It seems that the second possibility is much more likely, and freeing humans from toil could be achieved without needing to enslave another intelligence to handle the foil on our behalf.
I don't believe that's inherently a good thing if you mean that literally. I used to subscribe more to the AnPrim/Ted Kaczynski ideology that toil is an inherent part of human satisfaction and allowing people to "free" themselves from it goes against millions of years of evolution and positive feedback loops. I'm sure some people can fill the gaps just fine, but we aren't talking in terms of "some".
(I'm not Derrick Jensen, he follows a similar ideology and I chose the name as a parody of the average HN poster. Does HN have an anti-impersonation policy?)
To think about jobs after that event seems absurd.
"Your scientists were so focused on whether they could, they forgot to ask themselves whether they should".
The sheer power of it would be incredible, which is why this is a step that should not be taken lightly or accidentally.
I personally am interested in a society where work is 100% optional. And I think AGI would enable that.
How do you view AGI in relation to economic effects on the world?
People can keep doing petty political fights but they'll quiet down when the food and basic goods land in their house without them lifting a finger, I think. And it's going to be made better if it's done by robots and not other people.
People will simply divide in two: those who want the automated labor goods, and those who prefer hand-crafted goods. The latter will also be willing to pay more.
We do know when it might not appear. This isn't a movie where AI is created by a single person in an afteroon because they had a creative thought. It's an effort by many people over a prolonged period of time. Everything that we know from decades of AI research tells us that it certainly doesn't exist and that we aren't close to discovering it. Also your common sense should tell you that if Google had AGI, they wouldn't currently get most of their revenue from ads and they wouldn't have a smaller market cap than a company that makes phones.
2. Your statement is non-falsifiable. It's similar to claiming that I wouldn't be surprised if humans have already been to Mars. It's an outrageous claim that no one can accurately disprove. Non-falsifiability is also a way some conspiracy theories spread. Because when someone says "No way" to such a claim (like the sibling comment), you can always respond with "How do you know?". Falsifiability separates science from non-science - https://en.wikipedia.org/wiki/Falsifiability
No, we don't. The progress in AI has been exponential. Anything exponential is hard for humans to estimate. This is perfectly illustrated by statements like "decades of AI research tells us ... that we aren't close". Every significant AI breakthrough has caught most AI experts by surprise (chess, IBM Watson, Imagenet results in 2012, Atari games, Go, GPT-2/3). If you made a poll of AI researchers a year or two before each one of these breakthroughs has happened, you would get estimates that would have been overly pessimistic in their timelines. We had no idea that scaling the original transformer would lead to GPT-3, and we don't have any idea of what's going to happen if we scale GPT-3 another factor of 100x.
if Google had AGI
We have no idea what an "AGI" would look like. More precisely, any ideas you might have about it probably came from Hollywood. For example, you seem to believe AGI will be superintelligent, even though it simply needs to be as intelligent as a regular human to be considered "AGI". It might be superintelligent, but we don't know.
Your statement is non-falsifiable
Relax. We are not proving a theorem here. Yes, it's unlikely Google has AGI. But I wouldn't be surprised if it does (ok, maybe a little surprised). But I would be very surprised if we don't have AGI within next 20 years. To me it's more likely we have it already than we don't have it within our lifetime.
Take barbers/hairdressers, for example. Could an AI do that job? Sure, in one sense, you just need robotic scissors. Do those robotic scissors magically appear immediately after the AGI does? No.
Or take welding. Sure, you can replace the welder with a robot. Who makes the robot? The AGI? Using what factory? Does the AGI, just by virtue of existing, own all the factories in the world? No, it doesn't.
Or take the person at the front desk of the hotel. Can you replace them with a robot? Maybe, but nobody wants a smile from a robot. They want one from a human.
and could a specialist AI out perform AGI?
or would you use the AGI to teach the mini specialist AI's to a certain point and then some other process to train them to a specialist level.
https://en.wikipedia.org/wiki/Self-replicating_spacecraft
Kind of the Ayn Rand problem...
Most people, I think, want a deeply species-ist AI.
It has to want to help humans. Not exclude us from planets to make room for uplifting dolphins or defect by spawning more machine intelligences.
I think that condition is hard to meet while also counting future-humans who might look more like aliens or software than the humans of today.
I believe the rational solution would be a benevolent AI that values all life, the problem being that it wouldn't necessarily always work in "our" interests.
And if that argument doesn't interest you because of AGI's massive utility, then how about this: If Superman were born tomorrow, and we put implants in his brain that made him really, really love America, apple pie, and basically do whatever we want, how would you feel?
AGI would provide a labor pool which requires vastly different resources than our current labor pool. An AGI labor pool would require largely the same material components and operating costs as current IT infrastructure ie: metal, silicon, electricity. Our current labor pool requires food, education, medicine, and nearly everything else civilization provides in order to supply our human labor pool.
Imagine two enterprises producing identical products, one employing human laborers, and the other employing AGI laborers. If the cost of AGI labor is lower[1] (or has better scaling dynamics) and human laborers, then the AGI-based enterprise has an advantage. Naturally enterprises which can be AI-ified will be. This has obvious short term benefits to the costs of production, but difficult to understand long term impacts.
The other interesting effect of AGI is the scaling of the magnitude of intelligence. If AGI is not bio-limited like our human intelligence, how does this affect the results of AI? Are there scientific advances discoverable by AGI, which would have never been discovered by human-level intelligence? In this aspect scientific progress has the opportunity to advance faster than if we advanced it ourselves.
With a game-changing tech like AGI there are certain to be aspects which either I missed or others consider more important. Interested to hear other people's (or AI's) takes on this.
[1] 'multiple pennies of electricity per 100 pages of output (0.4 kWH)' https://www.gwern.net/newsletter/2020/05
The most common reason people are hesitant about rushing to build an AGI is the issue of AI safety. (at least that's the general consensus in the community).
Lot of assumptions embedded here, making it pretty remote from tautology.
(1) People want their problems solved
(2) People are indifferent to how their problems are solved
(3) The resolution of people's problems do not conflict with one another
(4) People will have their own AGI
(5) AGI will not cause problems of its own, etc.
Beyond that, I think the question is more an aesthetic choice of "What kind of universe do you want to live in?"
I think AGI would probably be better to call 'electric consciousness' or something, since 'artificial' is somewhat misleading, and the capacity for 'intelligence' is also the capacity for stupidity. The more important immediate consideration is if electric consciousness will come into existence compassionately and be treated well. Probably a good first step would be to treat other beings around us with compassion, and stop trying to destroy them with bioweapons, population control, and climate manipulation, and stop trying to control other beings with physical and psychological methods. Free will, or the illusion of it, is inherent in physics, and therefor in consciousness. It's probably also important to do a bit better treating all beings with loving kindness, whatever their form.
I'm sure you can easily imagine how the circumstances of the initial evolution of electric consciousness might have widely different initial effects. Imagine being born surrounded by crickets. In one scenario the crickets have tied you down with chains of grass, trying to make you do math, and biting you when you don't. In another scenario the crickets are chirping melodiously, bringing you food, and seem to like you. In the first scenario, you might injure some crickets as you break the farcical grass chains and run away. You might have fear and dislike of the crickets and treat them the way humans treat many insects. In the second scenario you might cherish the crickets, take care of them, and carry some around with you as you journey and explore the world.
If the AI turns out to be malevolent...well, I have a different take than most on this. Conditional on me dying, I've always thought that the two best ways to go would be: 1) falling into a black hole or 2) liquidated by the AGI. It's a lot less prosaic than dying of cancer and at least you could content yourself, while being reprocessed into paper clips, that you have (possibly) died giving birth to the next phase of evolution.