I personally don’t hold much credence in predictions that AI will either destroy the world or create a utopia. I feel like this is a combination of sci fy BS and Silicon Valley marketing hype designed to encourage capital investments. But like all tech, I’m sure it will change our society in profound and unexpected ways.
So what might our future look like? Good? Bad? Incomprehensibly bizarre?
I’ll add my predictions after a few responses.
More unequal, for one thing. It is a force multiplier both for people with compute & capital, and for people who know how to use it and can train it for their own purposes.
Within a society, that can lead to a stratification of the AI haves and have nots, at a scale exceeding even the digital divide we saw in the original dot com era, I think. Would it eventually equalize, like smartphones did (almost everyone everywhere has one now and internet is commonplace, twenty years later)? I dunno, maybe. There’s a lot of effort and small successes in making AI cheaper and more accessible. Some countries are considering providing it as socialized infrastructure to its citizens.
Between societies, I think its hype factor alone is enough to drive crazy arms races between yesterday’s and tomorrow’s superpowers. Now that we cut China off from chips, they have no choice but to fast-track their own (and maybe crave Taiwan and TSMC even more), which probably means they will leapfrog us in a decade or two, like they have with all other manufacturing. And superpowers aside, where does that leave the rest of the world that doesn’t have the capital and intellectual talent pool to make their own?
Environmentally… I don’t think AI’s current footprint is really that bad, per capita, than any other major industry right now, but if it keeps scaling, it will become that bad. I think it will become financially limited before that happens, though (we’re already seeing a slowdown in investment).
That’s just LLMs. If we actually get AGI, well then, by Borg, I guess we’ll just have to wait and see what it posts here.
I think it’ll be a things of “unintended consequences” leading to a huge QOL drop for humans still in the system. Given some expectations of AGI within as little as a decade (I’m skeptical, but worst case…) we’ll have endless repetitions of AI advisors (and eventually controllers) implementing lots of decisions that improve their situation / profits of their “employers” / institutions that constantly black-blast on the population.
Why do humans need personal electronics? They’d be Healthier (MAHA) and more able to work (Ultra capitalism) if those electronic resources were concentrated on where they can do the most good/profit, ie more computational power.
Lots and lots and lots of ugly cutting the Gordian knot to comply with orders/intents, with lots of figurative and literal causalties.
It might look like some dystopian fiction, where humans live in company towns with no amenities doing work too menial and too low level to use resources for, while the AIs (and the supra-elite 0.001% of the “controllers” absent AGI) focus on making increasingly imaginary numbers “bigger”.
And that’s actually something of a least-worst scenario. Because I expect it’ll be really bad before we get there. As AI becomes more available, we may end up with a Cyberpunk-esque scenario where every major corporation builds or rents their own to engage in virtual wars over influence, resources and insider information. And given the issues we’re already seeing absent any Federal regulation (much less the rest of the world or rogue states) there’s going to be consequences to out rather delicate intertwined economies, banking, and resource distribution.
Of course, I may be overly negative, but I’d rather expect the worst and hope for better. And I’m not out there stockpiling gold, beans or bullets - just pointing out how fragile the world we live in can be with just bad faith human actors is. Adding new layers of high response low EQ AI/LLM into the decision loop is almost certainly going to make it worse for most of us.
I’m generally pessimistic as well. But as for this particular part, a part of me wonders — or maybe even hopes — that having AIs make decisions for us (as world leaders), replacing humans instead of just being servants to them, could make the world a better place.
We don’t really select (human) leaders and elites for their ethics, intelligence, fairness, or benevolence. More just ruthlessness and charisma, maybe. They’re largely sociopaths.
I’d take my chances with an AI — even just the current-gen ones — over most any human leader alive today. Today’s Claude is already more balanced and compassionate than any CEO I’ve ever met or most people you hear about on the news.
It would be cool if the plebs of the world could join forces to train an of-the-people-for-the-people AI to fight the oligarchs’ ones. Gonna happen? No, but I can dream.
The answer is we can’t know. Everything we’ve ever done or ever known, or ever imagined was done with biological cognition. We have no idea what doors machine cognition will open.
Humans are basically just chimpanzees with a few billion more cortical neurons. That’s why chimpanzees throw their feces at each other and humans have a global civilization. A chimpanzee can’t fathom what it would do with an extra few billion cortical neurons, because its never had the ability. It can’t even visualize what humans think about or accomplish.
Thats why its called a singularity. We don’t know what’s beyond it.
In the real world, intelligence is not the only bottleneck we face. So even if we have far more intelligence the other real world bottlenecks will still be there. So progress will happen faster, but there will still be limits. Maybe 50 years of technological and scientific growth will occur in 10 years after AGI.
When cognition and problem solving are not limited to 3 pound primate brains, its impossible to tell what doors it will open. Its like asking someone in the 15th century to predict what life would be like after the industrial revolution. The smartest human is still just a human, and the strongest horse is still just a horse.
If our current AI models were being built as academic projects or from disciplined sources (Such as When HARLIE was One) I’d have a bit of hope. Since current LLMs are often built from training on the internet, I think it’s more likely to reflect our worst, rather than our best. And again, from an unemotional POV, it’s easy to fix some of our problems - feed everyone balanced nutritional lentil gruels, make us walk everywhere, constant physical work (mining rare earth metals), no war because everyone works as one, fed as one, educated (to use tools) as one, etc.
That’s why I mentioned QOL decreasing for the majority if not all. We can be healthy living organisms (assuming we’re not going the road to the Matrix or Terminator franchises) while not being free, thinking, or happy organisms.
And so many problems could just be solved by reducing our numbers - we’ve had more than a few threads about improving (depending on POV, in EITHER direction) human population, but with much discussion about the effects on our civilization. An AI could make that choice unburdened by such worries, only about the easiest, cheapest way to do so. Which is a very different equation all together.
All that said, I absolutely HOPE you’re the right one in this debate! FSM make it so.
Mostly good. AI will cut through many inefficiencies, but our future society will still somewhat resemble today’s pretty closely.
Nearly all of the objection to AI is straight-up Luddite-ism, fear of the new replacing the old. Although some privacy or data concerns are valid.
I suspect we may see a deep(er) cultural divide between AI-driven high tech megapolis like New York and San Francisco and more rural working class communities. One of the main reasons IMHO is that working class jobs are more often driven by physical constraints (iRobots not withstanding) while urban jobs are often data-driven. There are limits to how much advanced analytics can help you shingle a roof or construct a highway.
My main prediction is an AI driven society that is at risk of becoming vapid, directionless, and stagnant. Fictional examples include the crew and passengers of the Axiom from Wall-E, Idiocracy, Brave New World (with AI acting as a sort of digital Soma), or the bizarre hedonistic world of Logan’s Run.
There simply won’t be enough work that actually HAS to get done (we only need so many plumbers and electricians). And even without AI, I think we have become a society that supports A LOT of highly educated people in jobs of questionable utility or value. But we also don’t really tolerate a world where 99% sit around watching Neflix and don’t at least PRETEND to look for some meaningful work to do.
Logan’s Run is an interesting example as it portrays the sort of bizarre society I might see a well meaning AI evolving over time:
Exactly. A perfect example is an AI conference I went to last year where we had to wait in a line 50 people deep for pizza. AI is not going to suddenly increase the amount of beach-front property or our petroleum reserves. Maybe it will help use resources more efficiently with better science and analysis (aside from the resources it consumes). But AI is not going to warp space and time and change the laws of physics.
Or specifically, ask that person something specific like how everyone having a phone with a 4K video-camera that fits in their pocket will transform society.
And a few centuries earlier, Gulliver’s Travels:
Gulliver is shown a large wooden contraption that allows “the most ignorant person … [to] write books in philosophy, poetry, politics, laws, mathematics, and theology, without the least assistance from genius or study.” Swift, who was deeply learned, was envisaging a nightmare world in which reading and writing — the essential foundation of wit and wisdom — are replaced by a flashy simulation of human knowledge.
But the gap between 15h century and after Industrial Revolution was massive, whereas we’ve already sort of neared the peak ceiling in terms of what technology can give us. For instance, an airplane is always going to be an airplane; we’ll only see slightly more modern versions of it. A computer is a computer, a smartphone is a smartphone, and even with AI, it’s likely we’ll only see better computers and smartphones but no true monumental difference. A ship is a ship, a submarine is a submarine, a rocket is a rocket, food is food, the human body is the human body, and the gap between year-2026 and year-2126 is probably going to be much less than the gap between the year 1926 and year 2026.
and if you take cranberries and stew them like applesauce, they taste much more like prunes than rhubarb does.
It’s hard to say given that many of those things didn’t exist in the 15th century. Many of those inventions changed so much over the last century I’m hesitant to believe that we have reached the peak of human technical progress.
But bigger ships just means we can ship more crap.
IF the political will is there, I imagine AI could be used to distribute resources and wealth more equitably.
But as you say, a car is a car, etc and these all have physical constraints that don’t just disappear because someone smarter is building them
No risk of that happening!
Competitive AIs will be worse than competitive humans in terms of a ruling class.
I see two outcomes. AIs become aggressive and eliminate humanity or AIs and humanity become symbiotic in some form.
Stuff like that will change dramatically. We’ve barely scratched the surface of fields like genetic engineering and biotechnology.
Our consciousness lives inside our brain, and our brain lives inside our bodies. Eventually we will engineer far better ‘brains’ and far better bodies for our consciousness to live in.
The brains we live in only have about 6 core emotions and all emotions are supposedly a combo of these 6 primary emotions (anger, disgust, happiness, sadness, surprise, fear). We could engineer endless emotions far beyond these 6, or things we can’t even fathom far beyond concepts like emotions. Perhaps we will find that our 4 dimensional spacetime is a small sliver of a higher omniverse reality. Who knows.
Chimps have about 9 billion cortical neurons, humans have 16 billion. Our excess cortical neurons are a big part of why we are more intelligent. We couldn’t fathom what a species with the equivalent of 100 billion+ cortical neurons could do just like chimps can’t fathom the things we do. The most advanced civilization and paradise a chimpanzee could fathom would look very rudimentary and limited to a homo sapien.
I’ve been enthusiastic about the potential for AI long before LLMs came along, but I’m also skeptical about AGI – not because I think it couldn’t be built within a decade, but because I see no persuasive or practical reason for doing so.
The model I see emerging is analogous to that of human professional specialists. A cardiac specialist knows things that gastroenterologists do not; a lawyer knows things that doctors of any kind generally don’t, and don’t need to know. What would be the purpose and function of an AGI?
LLMs are currently indeed general-purpose AIs, but they likely won’t continue to evolve that way. IBM has been following the specialist model with Watson, first training it as a demonstration that can win at Jeopardy, then training it (with little commercial success, unfortunately) to be a medical advisor, and now touting it as a business advisor.
Even if we built a humanoid type robot that is the stuff of sci-fi, for the purpose of performing menial household tasks, what would be the purpose in imbuing it with super-intelligence? It would just need to know how to keep the house in order. It would not need to know how to perform heart surgery, or argue propositions from Wittgenstein.
It depends on how their training corpus is curated and filtered. In the very early days of chatbots, Microsoft’s early effort quickly became an obnoxious jerk. Today Grok seems relentlessly racist, but we know where that comes from. But as @Reply has stated, that’s not generally true:
I agree with that, and ChatGPT is the same.
I don’t know about end of the world scenarios, but Bill Gates says it will replace most doctors and teachers in the next decade.
And taking that a step further, we can see where a whole bunch of other jobs will be at risk, too. And that’s the part that would worry me if I was starting out today. I would be looking at the trades, and even quirky stuff like morgue related manual work. (I know a young lady doing this right now)
I will just point out that Bill Gates is neither a doctor nor a teacher.
The pragmatic reason is at the state level. If AGI is possible and will iterate upon itself and make itself faster and more powerful over time then no country that has the capability to invest in the possibility of being the first to possess an AGI won’t.
I doubt AI will replace doctors in any foreseeable future, but many knowledge-worker jobs will indeed likely be at risk. I believe this is a major impetus behind the huge investments being made in AI, and LLMs specifically, by the big tech companies.
Bill Gates may well be right about teachers being at risk, too. The idea of teaching being delivered by some sort of automation is not new. Decades ago some undergraduate college lectures were played on videotape to large groups of students, with grad students wandering about just to keep order. It’s pretty obvious that a suitably certified competent AI would be a far better teacher than a videotape because the relationship would not only be interactive but also personalized (the student’s strengths, weaknesses, and skills being part of the AI’s memory).
But could one further argue that an interactive AI could – potentially, at least – be superior to a human teacher, particularly considering its access to greater depths of knowledge and detailed memory of the student’s capabilities. ISTM that a classroom of students learning from AI terminals and only lowly appointed entities like senior students present, solely to keep order, might be seen as a very cost-effective model of education.
I have the greatest respect for good teachers at any level. This is not a scenario that I hope for. But I do think that’s it’s one that is plausible. It’s also a scenario that leaves young learners subject to the social consequences of working with non-human intelligence.
One additional benefit of AI is that it is capable of teaching to the lowest-level and highest-level students at the same time. If a student is a beginner at calculus, it can teach it at the most basic layman-understandable levels, but if a student is a calculus master, it can also speak in the most arcane and advanced deep-in-the-weeds stuff too.