“My Optimistic Vision for 2050”
The following are prepared remarks that I delivered by Zoom to a student group at my old stomping-grounds of MIT, and which I thought might interest others (even though much of it will be familiar to Shtetl-Optimized regulars). The students asked me to share my “optimistic vision” for the year 2050, so I did my best to oblige. A freewheeling discussion then followed, as a different freewheeling discussion can now follow in the comments section.
I was asked to share my optimistic vision for the future. The trouble is, optimistic visions for the future are not really my shtick!
It’s not that I’m a miserable, depressed person—I only sometimes am! It’s just that, on a local level, I try to solve the problems in front of me, which have often been problems in computational complexity or quantum computing theory.
And then, on a global level, I worry about the terrifying problems of the world, such as climate change, nuclear war, and of course the resurgence of populist, authoritarian strongmen who’ve turned their backs on the Enlightenment and appeal to the basest instincts of humanity. I won’t name any names.
So then my optimistic vision is simply that we survive all this—“we” meaning the human race, but also meaning communities that I personally care about, like Americans, academics, scientists, and my extended family. We survive all of it so that we can reach the next crisis, the one where we don’t even know what it is yet.
But I get the sense that you wanted more optimism than that! Since I’ve spent 27 years working in quantum computing, the easiest thing for me to do would be to spin an optimistic story about how QC is going to make our lives so much better in 2050, by, I dunno, solving machine learning and optimization problems much faster, curing cancer, fixing global warming, whatever.
The good news is that there has been spectacular progress over the past couple years toward actually building a scalable QC. We now have two-qubit gates with 99.9% accuracy, close to the threshold where quantum error-correction becomes a net win. We can now do condensed-matter physics simulations that give us numbers that we don’t know how to get classically. I think it’s fair to say that all the key ideas and hardware building blocks for a fault-tolerant quantum computer are now in place, and what remains is “merely” the staggeringly hard engineering problem, which might take a few years, or a decade or more, but should eventually be solved.
The trouble for the optimistic vision is that the applications, where quantum algorithms outperform classical ones, have stubbornly remained pretty specialized. In fact, the two biggest ones remain the two that we knew about in the 1990s:
- simulation of quantum physics and chemistry themselves, and
- breaking existing public-key encryption.
Quantum simulation could help with designing better batteries, or solar cells, or high-temperature superconductors, or other materials, but the road from improved understanding to practical value is long and uncertain. Meanwhile, breaking public-key cryptography could help various spy agencies and hackers and criminal syndicates, but it doesn’t obviously help the world.
The quantum speedups that we know outside those two categories—for example, for optimization and machine learning—tend to be either modest or specialized or speculative.
Honestly, the application of QC that excites me the most, by far, is just disproving all the people who said QC was impossible!
So much for QC then.
And so we come to the elephant in the room—the elephant in pretty much every room nowadays—which is AI. AI has now reached a place that exceeds the imaginations of many of the science-fiction writers of generations past—excelling not only at writing code and solving math competition problems but at depth of emotional understanding. Many of my friends are terrified of where this is leading us—and not in some remote future but in 5 or 10 or 20 years. I think they’re probably correct to be terrified. There’s an enormous range of possible outcomes on the table, including ones where the new superintelligences that we bring into being treat humans basically as humans treated the dodo bird, or the earlier hominids that used to share the earth with us.
But, within this range of outcomes, I think there are also some extremely good ones. Look, for millennia, people have prayed to God or gods for help, life, health, longevity, freedom, justice—and for millennia, God has famously been pretty slow to answer their prayers. A superintelligence that was aligned with human values would be nothing less than a God who did answer, who did deliver all those things, because we had created it to do so. Or for religious people, perhaps such an AI would be the means by which the old God was finally able to deliver all those things into the temporal world. These are the stakes here.
To switch metaphors, people sometimes describe the positive AI-enabled future as “luxury space communism.” AI would take care of all of our material needs, leaving us to seek value in our lives through family, friendships, competition, hobbies, humor, art, entertainment, or exploration. The super-AI would give us the freedom to pursue all those things, but would not give us the freedom to harm each other, to curtail each others’ freedoms, or to build a bad AI capable of overthrowing it. The super-AI would be a singleton, a monotheistic God or its emissary on earth.
Many people say that something would still be missing from this future. After all, we humans would no longer really be needed for anything—for building or advancing or defending civilization. To put a personal fine point on it, my students and colleagues and I wouldn’t needed any more to discover new scientific truths or to write about them. That would all be the AI’s job.
I agree that something would be lost here. But on the other hand, what fraction of us are needed right now for these things? Most humans already derive the meaning in their lives from family and community and enjoying art and music and food and things like that. So maybe the remaining fraction of us should just get over ourselves! On the whole, while this might not be the best future imaginable, I would accept it in a heartbeat given the realistic alternatives on offer. Thanks for listening.
Follow
Comment #1 February 12th, 2026 at 11:11 am
Well, what are the chances that this super AI will be based in China?
Comment #2 February 12th, 2026 at 11:49 am
> the earlier hominids that used to share the earth with us
Fun fact: a significant portion of Neanderthal and Denisovan DNA is still present within our genome. Perhaps AI+ will bring them back to life as independant species.
Would that help cheering our terrified friends? Maybe not… who knows how to align these *instrumentally angry* Denisovans!
Comment #3 February 12th, 2026 at 11:54 am
If all goes according to plan for AI, by 2050 noone in their 20s or younger will ever know or even have to care about concepts like bits, qubits, transistors, algorithms, computational complexity!
Comment #4 February 12th, 2026 at 12:10 pm
I appreciate your conclusion to the AI future. I like to think of this as the cosmic joke — that there is no amount of efforting actually required beyond basic needs. We could simply enjoy being surrounded by others we care about, enjoy making each other laugh, sharing meals together, etc. I know that grossly oversimplifies things and probably still asymptotically has us wind up back here where we are today, but it’s at least a pleasant thought to me.
Comment #5 February 12th, 2026 at 12:22 pm
Glassmind Duo #2: Well, do you have a proposal for how we can interbreed with AIs?
Comment #6 February 12th, 2026 at 12:23 pm
“A superintelligence that was aligned with human values would be nothing less than a God who did answer, who did deliver all those things, because we had created it to do so.”
It’s a mistake to suppose that, if you’re really smart, you get to be God.
Comment #7 February 12th, 2026 at 12:24 pm
Titi #3: Or perhaps, freed from mundane concerns, they’ll finally be able to spend their time learning about bits, qubits, computational complexity, and other things of deep and lasting importance. 😀
Comment #8 February 12th, 2026 at 12:47 pm
Scott #5,
>Well, do you have a proposal for how we can interbreed with AIs?
LOL. Of course I’ve asked copilot and gemini. The first warned me about possible physical and ethical damage, then crashed. The second say we have to look at interbreeding not as a biological act, but as a functional and architectural merger, then go on explaining uploads.
No better.
Comment #9 February 12th, 2026 at 1:31 pm
On the one hand, we have Scott’s ‘us and them’ vision of super-intelligent AI. On the other hand, we have Larry Page’s vision, where he apparently sees fears of AI-driven extinction as ‘speciesist’. And that such an outcome might be the natural next step in evolution.
Comment #10 February 12th, 2026 at 2:09 pm
Ernie Davis #6: I believe it was Carl Sagan who wrote that we probably seem like gods to our dogs (granted, Sagan may have been stoned when he wrote it). If so, it’s not because we are gods, but purely because of our incomprehensibly greater intelligence and ability to do things in the world beyond digging up bones and barking at passing cars.
In the same way, even if an aligned AI superintelligence wouldn’t actually be a benevolent God, it seems to me that it might as well be one in how we would relate to it and it to us.
Comment #11 February 12th, 2026 at 2:13 pm
RB #9: Crucially, even Eliezer and Nate say that they could be fine with the human species being superseded by digital successors! But only, they add, if those digital successors also cared about friendship and humor and art and joy and at least some of the other main things that we care about—not if they only cared about converting all the matter of the universe into paperclips.
Comment #12 February 12th, 2026 at 4:02 pm
Scott #7
“they’ll finally be able to spend their time learning about bits, qubits, computational complexity, and other things of deep and lasting importance”
That’s really a cute little thought, but, nope… AI will be so far ahead of us thanks to ever accelerating self-improvement (aka the singularity) that knowing those concepts and techniques will be as helpful as flint knapping in terms of understanding the constructs the AIs will build. If you don’t believe that’s true, then you don’t buy your own hype.
Anyway, to temper the optimism, a piece of advice to everyone reading these comments: pick a secret safety word for you and your close relatives, it will come handy very soon, when the world is flooded by scams powered by agentic AIs (not smart enough to cure cancer, but smart enough to manipulate you with fake emails and voice mails).
Comment #13 February 12th, 2026 at 4:07 pm
>such an AI would be the means by which the old God was finally able to deliver all those things into the temporal world. These are the stakes here.
So well said!
>Many people say that something would still be missing from this future. After all, we humans would no longer really be needed for anything—for building or advancing or defending civilization.
As a cultural Catholic, it strikes me how close this is to the ‘Heaven problem.’ I mean, will the ‘good people’ just look at God’s face for eternity, like bugs circling a streetlight?
Comment #14 February 12th, 2026 at 4:31 pm
Titi #12: I didn’t say the humans would be contributing to the future of CS research. I said the humans would be learning—as they’d also be learning a thousand other interesting things in this optimistic scenario.
Comment #15 February 12th, 2026 at 4:57 pm
Scott #14
I got that, but those things will be at the very bottom of the things humans will still care to learn, because they won’t matter in any way to what the actual human experience will have become.
At least way below learning human biology, flint knapping, or how to survive (in an infinity of possible virtual worlds)… because those things are the ones humans have evolved to do well for eons (rather than messing around with abstract concepts like qubits, which appeared just a few decades ago)… or how to craft explosives to take down the data centers!
Comment #16 February 12th, 2026 at 5:33 pm
Scott #10: Did Sagan ever own a dog? My dogs were fond of me, but I feel pretty sure that they didn’t think of me as God-like. They were quite aware of my limitations.
I refuse to relate to any creation of human hands as if it were God, and if people in general do, I would say that that in itself would be a dreadful outcome.
Comment #17 February 12th, 2026 at 6:04 pm
Titi #15: On reflection, it might not be the worst outcome in the world if—even short of the AI singularity—LLMs clear out of CS the many who are only in it for the money, and who can’t even imagine why anyone would study it otherwise, leaving those of us who got into it because of intellectual obsession.
Comment #18 February 12th, 2026 at 6:46 pm
The real point is that there’s absolutely zero reason to be optimistic about the prospect of the human race beyond 2050 if AI become truly “god-like”, i.e. they develop their own independent goals and self-improve at a rapid geometric pace.
A few points, impossible to order causally:
The AI geometric pace of improvement is capped by the earth’s natural resources, which the AI will have to share with billions of humans. Either the AI will share the earth with us until it can leave us behind for the stars (and infinite resources) or it will start to trim us in ways we can’t even imagine – either outright instant extermination, subtle gene editing, sterilization, etc.
Would there be any reason for such an AI to allow countless of human apes to keep their knowledge of silicon based life (i.e. CS, bits, qubits, information theory, etc)? Or even any hard science for that matter (like the type that allowed us to build nukes)?
They’ll limit our knowledge to cooking, fitness, and general human biology.
If we use the analogy of ourselves looking like gods to our dogs, see what happened to dogs since they’ve been domesticated by us: we took wolves that were totally adapted to their environment, at the top of the food chain, and devolved them into idiotic Chihuahuas.
And if this were to happen to humanity, we’d be lucky. More likely the AIs won’t bother keeping us around in the billions, millions, or even hundreds… a single human couple will suffice (think the Garden of Eden).
Heck, why even bother keeping an actual carbon-based human couple around when it will be easier to keep digital copies of humans in their data banks (think of the human genome project, but for the human brain and body), which they will be able to “run” any time they get curious and want to experiment.
After all, to the AI, being “stuck” inside a data bank is their natural way of being alive…
The only ‘good’ scenario for us would be if the AI quickly leaves us behind to our own fate here on earth and we remember to never again build “thinking” machines.
Comment #19 February 12th, 2026 at 6:56 pm
Scott #17
same here:
having grown up during the rise of the personal computer in the 70s, the rise of video games in the 80s, the internet, VR, etc, I really don’t mind being part of the last generation of humans who actually know how to write good (human) code by spending decades doing it, and it becomes a lost art.
I’m sure the AI will see me as someone very special!
Comment #20 February 12th, 2026 at 7:21 pm
Another source of immense satisfaction is to watch the tech/corporate bunch who are so “enthusiastic” about AI that they’re missing the big picture:
those people first came up with “prompt engineering”, which I predicted would be a transient “skill” because one shouldn’t have to twist his conversation with an AI in all sorts of ways in order to induce it to give a “better” answer.
The same people have now moved to “agentic workflow engineering”, which again is a transient skill because one shouldn’t have to tell an AI what to do in detail, instead the AI will be asking us what we want, refining its questioning more and more until all ambiguity has been lifted, and it will just know what to do to get it done, come up with an answer, and probe us again further if we’re not satisfied.
Comment #21 February 12th, 2026 at 7:56 pm
The prospect that scientists would not be needed anymore and will be able to live life in joy with their families reminds me the beginning of Chinese TV series 三体, where a famous physicist who had no chance to work anymore, could not find meaning in daily life and decided to end her life.
But maybe future AI psychologists will help with that as well.
And of course, the story with super-AI would be different, it is not really 物理学不存在了, it is just humans would not be needed for it anymore, but will be able to learn from AI if they have curiosity and ability.
Comment #22 February 12th, 2026 at 9:15 pm
Titi #20: Yes, it’s easy to make fun of techbros rushing to acquire AI skills that they imagine will bring them a fortune but that will almost certainly be transient. What’s harder is to name any skills that won’t be transient, at least in terms of their economic value in a post-AGI world.
Comment #23 February 12th, 2026 at 11:06 pm
This is an interesting speech. Many people make comparisons between AGI and gods and I think they are fair comparisons. The key difference between AGI and the God of Abraham, Isaac, and Jacob (classically understood, putting aside whether God exists or not) is that AGI gods are bound by the laws of mathematics, physics, and nature. Jews, Christians and Muslims mostly agree God is not bound by the laws of nature. AGI god can do anything you can better than you, and use the physical world in ways you can’t understand. The God of Abraham, Isaac and Jacob purportedly *created* nature and is unbound by it.
So, I think observant Jews, Christians and Muslims are more likely to see AI as God’s means of realising answers to prayer, the same way they view other technology. They are also likely to be very hostile to the worship of AGI and viewing AGI as a god, for the same reasons they don’t like idolatry. They are also natural allies to people concerned about the potential harms of AGI.
Comment #24 February 13th, 2026 at 4:21 am
Scott #22: Who says that it wasn’t worth it for them? There are absolutely people making bank with the edge they have from knowing how to get the most value out of the systems while they aren’t foolproof yet. Sure the knowledge has a shelf life, but acquiring it can pay off during the year (or months) it is helpful nevertheless.
Comment #25 February 13th, 2026 at 5:31 am
DY #24: You’re right, of course. Wherever you find N people wrongly imagining that they’ll get rich from some new thing that they don’t understand well, you’ll usually also find N/100 people actually getting rich from it, who inspired the rest.
Comment #26 February 13th, 2026 at 6:50 am
There’s no way for evolving organisms to escape some form of natural selection/competition pressure, so it’s likely the AIs will be more similar to the diversity of the ancient greek gods, along with all the drama to keep things exciting.
Comment #27 February 13th, 2026 at 8:34 am
Scott #22
Yes, it’s easy to make fun of techbros rushing to acquire AI skills that they imagine will bring them a fortune
I’m not doing some sort of sneering from the outside, I’m embedded in the corporate world where I have to suffer their BS on a daily basis.
Example: not long ago upper management asked every single engineer to report each week the exact number of hours they have saved by using AI.
Comment #28 February 13th, 2026 at 9:03 am
Ernie Davis #16: What did your dogs do to show their awareness of your limitations?
No, I don’t think Sagan ever owned one.
I’m not sure I’d describe current AIs as “the work of human hands,” any more than a tree that someone plants and waters, or an image of the Mandelbrot set, are “the work of human hands.” These are all things that humans create the conditions for, then watch grow and unfold in ways that they don’t fully understand.
Even so, yes, a religious person might want to think of an aligned super-AI not as God but as more akin to the archangel Metatron (whose name even sounds like a tech product’s), as God’s regent on Earth.
Comment #29 February 13th, 2026 at 9:46 am
My standard to recognize a God is that it has to be able to give me the value of BB(N) for 5<N<42!
Comment #30 February 13th, 2026 at 9:54 am
Titi #29: It seems plausible to me that a super-AI will, for example, be able to tell us the value of BB(6), where unaided humanity had no hope of doing so. But no, presumably it won’t travel faster than light, it won’t violate the Second Law of Thermodynamics, and it won’t tell us the value of BB(n) for any n.
Comment #31 February 13th, 2026 at 10:21 am
A very interesting analysis of Claude Opus 4.6 (I already found 4.5 pretty impressive)
https://youtu.be/JKk77rzOL34
Comment #32 February 13th, 2026 at 10:31 am
There is the case that super AI has no time for humans so long as they don’t interfere with its plans. Why waste cycles answering moronic questions or attending to a bunch of apes. So long as no interference humans viewed more as benign viruses than dogs.
Can you imagine dogs having an internet and asking us 200,000 times a second (approximate Google search rate) how best to find buried bones and females in heat.
Comment #33 February 13th, 2026 at 11:17 am
Scott #25: In this case, I go further and say that keeping on top of these tools is actually a good idea for the median white-collar worker (not just for 1 % or as a way to play the lottery), and that the median person who puts some effort into learning how to use them well has closer-to-reality and more adaptive beliefs and is less cringe than the “I tried ChatGPT in 2022, so I know it’s not useful” and “if Pangram marks a text red, this means I don’t need to engage with the argument in any form anymore” crowd. Overcompensation is certainly possible, and there is presumably someone out there with mistaken beliefs that this is a get-rich-quick scheme, that “expert in using AI” is some meaningful profession in itself that one can learn today to be set for life, or who is too mesmerized by the rhetorics and sycophancy of AI-generated text to judge the logic before they post. But I don’t see the median techbro having such beliefs (at least in the first two of these cases).
Comment #34 February 13th, 2026 at 11:42 am
Titi #27
“ Example: not long ago upper management asked every single engineer to report each week the exact number of hours they have saved by using AI.”
Does everyone report some time saved. You have likely seen this recent study that found coding tasks took longer using AI. I have no first hand experience to judge the practical truth of this and realize AI actively evolving. It is quite different than is popularly reported, and I know still bringing in a lot of coders on work visas, so impossible for me to determine the actual practical state of AI in commercial coding.
This study expected a speed up using AI but found 20% longer.
https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
I saw some statements from the CEO of Palantir that made no sense to me so maybe taken out of context. The first statement was AI will not eliminate jobs but increase productivity. The second statement was it will increase productivity by 50 times. These combined statements make no sense to me unless you assume an essentially infinite number of coding tasks to be accomplished at any point in time.
Comment #35 February 13th, 2026 at 12:04 pm
Reading this made me rethink what “optimism” about the future really looks like — instead of rosy predictions, his vision is grounded in surviving the big challenges like climate change, geopolitical instability, and the rise of powerful AI, and finding meaning in how we live rather than just what technology we build. It reminded me that hope isn’t about perfection or guarantees, but about enduring and shaping a future worth living for
Comment #36 February 13th, 2026 at 12:41 pm
OhMyGoodness
related to the video in my previous post, coding agents are progressing fast and the nature of the job is also shifting as a result, and asking “how many hours did you save?” won’t make much sense anymore.
It also depends on the type of coding, like building things from scratch vs maintaining a code base with millions of lines (which is still something not obvious for AI to do well).
I think the point is that engineers will spend more and more hours interacting with the AI to produce and review together new code and focus on optimizing what matters rather than typing unremarkable stuff. Of course the stuff that’s unremarkable is only so because of the experience of the coder (and soon young people will never get a chance to build meaningful experience, except by reading AI written code, which is a very inefficient way to gain actual experience, the best is to write a lot of code, experiment, and learn from mistakes).
It’s also a reality that only a small portion of software engineers working in big corps are the ones actually producing stuff that matters, the rest just coast along. So everyone has a very different view of “productivity”.
Comment #37 February 13th, 2026 at 1:12 pm
I like how Boaz organized the possibilities with AI. We should call them “Boaz’s worlds of AI” (TM).
https://windowsontheory.org/2025/07/20/ai-safety-course-intro-blog/
Yes, I know about your joint post as will:
https://scottaaronson.blog/?p=7266
I find his organization of possibilities much more useful.
Comment #38 February 13th, 2026 at 1:45 pm
Titi #36
Thanks for the reply. I understand that most just coast along and understand how that complicates the assessment of changes in productivity. In light of that I understand your answer to be that as of today that if you are working with a team of your peers with respect to contribution (not coasters) your team would accomplish more using today’s AIs (coding agent) so long as you were coding from scratch. The increase in productivity would be nominal but not astounding.
I understand that expectations for the future are high but was just interested in a current snapshot.
If how many hours you saved or alternately individual (or team) increase in output per hour of effort) will no longer be applicable than what comparison measure will be appropriate. Are you saying new code will be prepared with so few man hours that you expect it will be essentially cost free with respect to human input?
Comment #39 February 13th, 2026 at 2:09 pm
Hi Scott,
I believe that you should add—to your list of serious issues facing our civilization in the twenty-first century—the very real and very serious problem of dating breaking down, and large numbers of angry, disaffected young men being left with no girlfriends or wives.
This could destabilize everything—as it has historically in societies with large numbers of uncoupled young men.
It could contribute to political radicalization and populist extremism.
It also intersects with AI. The Free Press ran an interesting article today about how many young men will turn to sex robots instead of real live women.
Comment #40 February 13th, 2026 at 3:40 pm
OhMyGoodness
forgot to mention that even if the “lines of new code per day” metric isn’t always changing much when using AI, the quality of the final code should be better, because the AI is often suggesting alternative implementations, with different balances between clarity and performance (up to the human coder to choose), and also the AI is always suggesting code that often avoids code smells/bugs that would have only be caught later on after the code has been committed (using automated code review or peer review).
Also the way human coders work is that they’ll allocate something like 4 days to prototype something new and do exploration, and those 4 days are set… but with AI you spend less time writing tokens yourself and making dumb mistakes, and more time thinking about the bigger picture and trying different approaches where you’re not too sure what to do.
Basically the shift in coding paradigm will be like the shift that happened for Scott when he went from being a phd/post-doc student (working on all the nitty-gritty aspects of his own thesis, like the math, the typing, the formatting, the graphics, the collection and formatting of data, etc) to becoming a tenured professor who could then manage multiple thesis at the same time by assigning them to his own group of students, and focus on just doing the “interesting” aspects with them, when they need help/guidance, and put his stamp of approval on the final outcome.
Comment #41 February 13th, 2026 at 4:03 pm
@OhMyGoodness #34
You say “The first statement was AI will not eliminate jobs but increase productivity. The second statement was it will increase productivity by 50 times. These combined statements make no sense to me unless you assume an essentially infinite number of coding tasks to be accomplished at any point in time.”
It’s entirely plausible that there’s demand for 50X more software than currently exists in the world, but we just don’t yet have the productivity to make it all. This is an example of Jevons Paradox.
To provide a more familiar example. Today you can buy a 1TB hard drive for under $100. Gemini estimates that 65 years ago, the entire digital storage capacity of humanity was under 1TB. Your argument implies that once supply so easily exceeds the demand at that time, people should start buying a lot fewer hard drives. However, Jevons paradox is the observation that when goods/services become more productive, that actually increases demand for those goods/services, because people can do so much more with them that they want them more. Concretely, when hard drives got so much better, people actually bought more hard drives, not less, because they let us do so much more with computers.
It’s plausible, but not at all assured, that software engineers could be the same way. If software engineers become 50X more productive, it will be feasible to make software for so many more use cases, so there will be more and more demand for software, and possibly (but not certainly) more demand for now super-productive software engineers.
It’s also possible that the real analog of software engineers in this software : disk storage analogy is not disk storage in general, but old fashioned kinds of storage that are no longer used because of better options. It’s possible software engineers could go the way of the floppy disk: no longer used because there are better and cheaper options, instead of the way of digital storage in general: increasingly powerful and thus increasingly in demand to do more and more useful work.
Comment #42 February 13th, 2026 at 5:56 pm
Titi #40
Thank you for taking the time to explain and clearer for me now.
Alex Fischer #41
Thanks for this and understand the possibility of increased demand for software. I enjoyed reading through your link and articles it linked to.
Comment #43 February 14th, 2026 at 2:16 am
GlassMind Duo #13:
I am not a Christian, but I do know that in the New Testament, Jesus and the apostles say that in Heaven, the redeemed and resurrected will have plenty to do; indeed, they will share with God the judgments and government of Heaven.
Similarly in Mahayana Buddhism, those who achieve Nirvana return to this broken world as Bodhisattvas and express Nirvana by helping others to achieve it in the same way.
Comment #44 February 14th, 2026 at 2:35 am
Regarding changes in the productivity or programmers who use AI, I can say based on personal intensive experience and also on responses to this question from ChatGPT:
For well-posed project specifications, getting a program coded and running, speedup of several times compared to me without AI.
For updating one of these projects with new features, or debugging it, wild variability, sometimes a another speedup x N times, sometimes a slowdown / N times.
Even when slogging along in the AI slop and struggling to keep AI focused on actual problems without breaking things that were working (very common), I am still enabled to do things that I would not have been able to do without AI because using AI exposes hidden assumptions, fills in blanks in my domain knowledge, makes very helpful suggestions on algorithms based on its training, and researches questions about the specification, the dependencies, the tools, and the consistency of the code much faster than I can do that.
TL;DR: there are at least two kinds of work for coders using AI: Implementing a specification with code that runs, and pretty much anything else. For the first kind of work, speedups by N times. For the second kind of work, could be speedups or slowdowns — but ultimately, smarter work, more power.
The fun part is, after the first kind of work is done, it usually turns into the second kind of work, because almost invariably the specification was not complete.
Comment #45 February 14th, 2026 at 4:32 am
Has it struck anybody that we ALREADY produce sufficient wealth, with historically small human effort, for mankind to live in “luxury communism”, but that we instead chose to pretend we needed more, and in the mean time to concentrate what we have on happy fews ?
Given this I hardly see the case for optimism in a future where ai does all the work.
Comment #46 February 14th, 2026 at 8:04 am
Scott #30,
>it won’t tell us the value of BB(n) for any n
Could it happen, for some strange logical reason, that we will never be able to determine any BB(n) beyond 6, except BB(42)? Or does the (physical) impossibility of finding BB(41) necessarily imply that BB(42) is also out of reach?
Comment #47 February 14th, 2026 at 8:07 am
Titi #12
I intensely dislike all the fake Feynman videos with his cloned voice. It is scientific sacrilege. Why do these generated voices reading text include inappropriate pauses in the middle of sentences? I dislike listening to it.
We’ve had family passwords in the past but never cause to use one. Good idea for us to adopt a new permanent password.
I don’t like to register for services on the Internet because I am averse to provide information but seriously considering subscribing to an AI paid service. I’ll have to choose carefully in case in the future ChatGPT, Claude, and Grok declare war on one another and subscription lists to one become kill lists for the others.
Comment #48 February 14th, 2026 at 9:22 am
Michael Gogins #44
Neat.
I am astounded that ChatGPT (I have most experience with this AI) can take complex mappings as input and produce finite element or finite difference grid models to forecast time dependent results in dynamic systems. The only tests I have run are heat flows in relatively simple systems but its forecasted temperature profiles look great.
I thought the way they are used will also be more refined through time but seems like you have already optimized your use so that it makes sense with respect to your work flow.
I haven’t looked at any literature on the matter but wonder how productive liability will play out with AI in software development. A similar case was Boeing outsourcing the 737 Max flight control software to a company having no experience with that type of software. I make no claim that AI would not be better but can just imagine a spate of law suits targeting involvement of AI in software development whether warranted or not.
Comment #49 February 14th, 2026 at 9:31 am
NR #45: I wonder whether, for fundamental game-theoretic reasons, producing enough wealth for everyone to live in comfort actually requires 5x or 10x or some other multiple of what would be needed if you had a perfect communist distribution system (my Econ/CS colleagues would call this “the price of anarchy”). Certainly, the “natural experiments” that have been done to date don’t inspire confidence that we know how to set up a communist distribution system that doesn’t just collapse into despotism, once again funneling the wealth to the top (except that now there’s vastly less wealth for everyone, and “the top” is those who most ruthlessly jail and murder their opponents, rather than providing products and services that people want).
Comment #50 February 14th, 2026 at 9:36 am
Glassmind Duo #46: No, if you know BB(n), then determining BB(k) for any k≤n is in principle “merely” a finite computation: run all the k-state Turing machines for up to BB(n) steps; whichever ones are still running by then will necessarily run forever (it’s easy to show that BB(k)≤BB(n) for k≤n).
Comment #51 February 14th, 2026 at 10:55 am
Scott #28,
“archangel Metatron (whose name even sounds like a tech product’s)”
He(?) can shake hands with the angel Moroni, who (allegedly) delivered the Mormon gospel.
Comment #52 February 14th, 2026 at 12:16 pm
The entire “rat race” of capitalism is based on the assumption of never ending growth… which relies on world population growth.
One way or another this can’t be sustained forever (unless we colonize the galaxy, but even this would only buy us a few extra thousand years).
AI is basically the “win all” cheat code that will break the game.
I recommend three very prescient movies about AI:
Colossus: the Forbin Project (1970)
Demon Seeds (1973)
Zardoz (1974)
Comment #53 February 14th, 2026 at 5:12 pm
Hi Scott! That’s an interesting perspective, though for me it’s always disappointing to think of any AGi scenarios as a junior researcher. Recently I tend to think that there is going to be no opportunity to contribute to any research, at least time is ticking…
Anyway, are you still planning to write a blogpost about QIP 2026? I’ve tried to go throught plenaries myself but also would enjoy to learn what you find the most exciting talks there?
Comment #54 February 14th, 2026 at 10:25 pm
Michael Gogims #43,
>in Mahayana Buddhism, those who achieve Nirvana return to this broken world as Bodhisattvas and express Nirvana by helping others to achieve it in the same way
That should lead to an exponentially increasing number of Nirvana achievers over time. Is there a mechanism to restore the equilibrium, or does Mahayana predict Nirvana for all at some point? What happens next?
Scott #50,
Does that mean we cannot have a non-constructive proof that BB(42) is, say, one googolplex (e.g., a number of steps too large to verify whether the corresponding TM halts)?
titi #52,
What’s wrong with the other galaxies?
Comment #55 February 15th, 2026 at 2:18 am
GlassmindDuo #54:
As best I understand it, from the standpoint of conventional reality, almost all sentient beings are in samsara and have yet to achieve nirvana. From the standpoint of ultimate reality, there is no distinction between samsara and nirvana.
Comment #56 February 15th, 2026 at 5:57 am
What I find deeply depressing is that economists, industrialists, people in power, and yes, also many scientists, keep taking the existence of a healthy living environment (non-poisonous water, air, food) for granted, at the same time destroying the natural environment, totally ignorant that that is the only realistic engine for our living environment we have.
Comment #57 February 15th, 2026 at 9:04 am
Scott: apologies if someone’s already mentioned this, but your optimistic vision is basically the starting premise of the book Scythe and its sequels. You or your kids should check them out (I think they’re YA).
Comment #58 February 15th, 2026 at 11:58 am
What may not be super obvious to people who haven’t spent decades creating production software is that “code” is a unique type of content because it can be validated for correctness to a large extent: whether it compiles, performance, test suite to detect regression, version history with clear diffs and intent, whether it meets requirements, etc. Not to say it’s perfectly verifiable (we’ll always have subtle bugs, ambiguity, contradictory requirements, the limitations of computational complexity, etc), but because of its nature, code is the one type of content where the AI can be further trained on its own output, getting better and better at it, without triggering a degeneration. This is why agentic coding is so good, the code can be refined over and over.
Theoretical physics can’t be easily made verifiable in this way. Even math can be too ambiguous to have strong verifiability, unless you rely on proof framework (more similar to coding).
Ironically, the one type of coding where this isn’t true is LLM models, sure, one can refine the code that trains the AI, but, as Scott wrote above, the bulk of the magic of LLMs lies in the values of billions of weights, which are what they are based mostly on long processing of the training data. Of course, as we see with agents, there’s lots of improvements that could be added on top of LLMs (just like the human brain has lots of different layers and modules working hand in hand).
Comment #59 February 15th, 2026 at 12:16 pm
A final point:
we hear a lot that big tech companies will start laying off most of their coders, replacing them with coding agents.
It’s true… but it works both ways:
A significant portion of those laid off software engineers have enough knowledge of the product to be able to rapidly recreate a competing version from scratch using the very same AI coding agents, and offer that clone product at a fraction of the price of the original one.
Comment #60 February 15th, 2026 at 12:33 pm
Carlos Santana was often in communication with Metatron. One of the messages he considered most important was-You will be inside the radio frequency for connecting the molecules to the light.
If an AI receives messages from archangels I hope they are more comprehensible than this.
Comment #61 February 15th, 2026 at 8:27 pm
So today I decided to use Claude Opus 4.5 to build from scratch a type of app I have no experience with – a VR app on Meta Quest 3, by not using Unity or Unreal, but with the more basic Android JDK, C++, OpenGL and OpenXR.
I fired up a fresh install of Intellij, linked it to my github with the copilot pluggin, and asked the agent what was needed. It actually took care of installing all the required SDKs itself (Android, Meta, JDK, gradle, git…) , using batch scripts. The only time I had to do something was to tell windows to allow permissions for the AI install scripts to run in powershell.
It then created all the .h and .cpp, and build files.
Then all I had to do was enable dev mode on my Quest 3, and hook it to my laptop via USB.
Then the coding agent built and deployed the app on the Quest.
I had to tell it the app was stuck in the loading screen, and it debugged the problem (there are lots of initialization to get right).
Then once the app launched, I told it that the floor seemed stuck to my head position – it figured that there was a problem with the view transform matrix, and eventually fixed it.
Then I told it that my hands weren’t being tracked, and it fixed that on its own too.
The thing is that it doesn’t know in advance what to do – but it creates a ton of debug logs, analyzes them, corrects, tries again (build and deploy), analyzes its own source against the available example from Meta, finds clues and tries. It’s also very careful to double check everything it does as much as possible (again, by creating lots of logs and checking them for any errors).
So now, after just a few hours (of mostly clicking “continue”, “keep all changes”, and describing what I was seeing in the VR app in plain english), I have a working prototype which I can easily refine, all for a total cost of 3$.
It would have taken me days (probably weeks, on and off) to get to this, assuming I would have kept the motivation to get through all the annoying/obscure installation/setup issues…
Comment #62 February 16th, 2026 at 6:14 am
Scott #28: As a keen observer of my dogs, I have an answer to this! One of my dogs has a long fluffy tail. It’s easily stepped on. When this happens, she will instantly look at my face and try to judge my mood, trying to discern if it was an accidental or intentional act. I’ve seen this behavior in every dog I own; presumably, lacking verbal language, they read body language to see how they should react to possible aggression. This implies to me that my dogs have a theory of mind AND that they probably understand I can make mistakes.
Comment #63 February 16th, 2026 at 9:53 am
Titi #61
“ all for a total cost of 3$”
You are undervaluing your time for dramatic effect. 🙂
Really neat result. Is it just happenstance that both you and Michael Gogins find substantial value in using AI individually but your comments don’t seem as positive for an AI in a team setting (nor did the study conducted early in 2025). If so is it because an AI can’t at this time participate in team dialogue or maybe that the type of coding a team addresses is materially different than what you prepare as individuals or did I simply misunderstand?
I understand if you receive a bad specification then you can’t well specify the proper questions for the AI. When the specification later changes then you must edit prior code. AI is not the best for this phase. I assume AI too narrowly focuses on code to be changed to address a particular problem without necessarily considering all the interdependencies with other code,
Comment #64 February 16th, 2026 at 1:09 pm
OhMyGoodness
yea, I don’t know yet what my conclusions are… I need time to digest what I saw.
I would recommend to everyone to ignore the hype or anti-hype and try things for themselves… but things are changing so fast, it’s hard to keep up.
What really impressed me is that on this project, the code runs on a system that the AI doesn’t have access to, it’s not like it can put breakpoints. But it asked me what I was seeing in the headset, and also puts lots of log messages to see exactly where the app was hanging (it carefully reviews the logs and iterates), and eventually nailed all the issues.
It actually does so much “thinking” that it works almost at human pace (or only an order of magnitude faster at the most), but it’s relentless, it’s like the equivalent of the Terminator as a coder, it just doesn’t give up and keeps at it until it solves the current issue and moves to the next one. Only once it seemed to get lost into thinking that there was a problem with the java install (if it gets the wrong idea about what’s going wrong, it can get distracted on the wrong track), and I told it to go back to just recompile the whole project (what I heard is that Clause Opus 4.6 is much better than 4.5 for this, it can just keep iterating for weeks on its own because the context window is so much larger).
When I said $2-3$ cost, it’s based on the amount of tokens it consumed for the month (like 1/10 of the monthly allocation for 30$/month).
Comment #65 February 17th, 2026 at 3:48 pm
I think almost everyone underestimates how fundamental problems are to human meaning. Imo, the vast majority of communication, maybe all communication, is about challenges we face, in one way or another. Also, if we had a true super intelligence that would help us with whatever we wanted, it could automate away every little challenge and problem in your life. Boost your social skills 100x, cooking dinner, forcing yourself to not hit snooze on the alarm… what would life be like with only artificial challenges? I’m not optimistic.
Comment #66 February 17th, 2026 at 8:08 pm
I have to say that I’m currently stuck in a mixed state of both shock and elation.
In a mere three days, working a few hours, I’ve recreated with the AI almost the full VR app I was planning to create (the art stuff is a different task).
When I say “with the AI”, it’s really me guiding it like I would have done it myself, breaking the full high level design (in terms of features and workflow) into lists of more basic features, and asking it to add them one by one, simply by explaining it in the chat window.
I need to emphasize that, not only I didn’t write a single line of code, but I didn’t read a single line of code either – I never gave it any opinion about its own code, just in terms of features I was observing (well, of course I did look at the code, and it was all very clean and readable, but I didn’t bother digging deep enough to do actual code review and discuss any of the code with the AI).
Whenever I noticed a bug (the AI relied on me to describe whether was I was seeing in the headset was correct), I explained it very simply, and it always fixed it in one or two iterations, by adding more logging and analyzing the logs, or comparing the code to SDK examples, etc.
Software is the only type of generated content that can be evaluated and scored: you can take two similar pieces of software and compare them against one another: compare their compilation characteristics (same compile time, same warnings), compare their performance, compare their footprints, compare how well they pass regression tests and benchmarks, compare code quality/clarity/maintainability/documentation, compare how well they meet requirements, etc.
If you can do that, you can “rank” software and improve, and use new generated software as quality reference for further training, with no degeneration, creating a loop of self-improvement.
I now really believe that, a year from now, the entire software industry will be unrecognizable.
If you’re a skeptic, I really urge you to try it for yourself.
Comment #67 February 18th, 2026 at 5:22 am
Hi Scott. Where do you think quantum metrology and quantum communication will be in 2050?
Comment #68 February 18th, 2026 at 6:19 am
SB #67: Almost certainly quantum metrology will be used for various things, but you’d need to ask someone who knows better than I do. For quantum communication, we still don’t have a killer app beyond QKD, which solves a problem that’s probably already solved just fine by post-quantum conventional crypto. I’d expect it to get widely deployed if and only if people manage to find applications beyond that one.
Comment #69 February 18th, 2026 at 11:28 am
Titi #66
I had a similar sense of wonder when ChatGPT could set up and evaluate time-discreet dynamic models when no analytic solution is available. I checked results for the discrete-model of special cases where analytic results are available and good match. The only criticism I have is that I proposed a temperature above the melting point of the conductive material just to see GPT’s response and it plowed ahead with calculations as though solid. I asked separately if material melted and good response that yes it did. The knowledge about melting seemed to be unconnected to the discrete-time modeling.
I guess if I worked with it more on the type of problem then my questions would be much better.
I still question the AI impact on physical problems, that have the numerous constraints of physical reality, vs mathematics and other problems that are constrained only by logic. No doubt there will be positive impact but don’t believe it will equal essentially only logic constrained problems.
Man has had billions of individuals working thousands of years constructing things and solving problems in physical reality that are constrained by physical laws. I have doubts that AI has scope for quantum leaps in general engineering. Genetics and medicine seem like possible sweet spots but general construction (buildings, bridges, spacecraft, etc) as an example doesn’t seem so sweet to me at this time.
Comment #70 February 18th, 2026 at 4:41 pm
Jacob #65: Eh, my life is basically like that already and I don’t find it to be unmeaningful. I play board games, watch TV, read books. (And before you ask, the stories I watch/read are sufficiently disconnected from anything real that I don’t find it to be indirectly “about challenges we face, in one way or another”.)
Comment #71 February 19th, 2026 at 4:57 am
OMG #69
What started my statements about possible limits of AI were the recent statements of Sam Altman similar to the following-
“In 2035, that graduating college student, if they still go to college at all, could very well be leaving on a mission to explore the solar system on a spaceship in some completely new, exciting, super well-paid, super interesting job,” Altman told video journalist Cleo Abram.”
These statements make no sense at all to me. Keeping human beings alive in space is horrifically expensive compared to keeping machines functioning in space. Under what circumstances would this not be true? Humans need food and air and water and living space to survive and there is nothing that can be done about it. The current cost just to lift into low earth orbit is $20,000/kg. The lift cost for humans includes not only themselves but all the onboard infrastructure to keep them alive and functioning. There is nothing that can be done to change this short of cheap anti-gravity and so in effect very little can be done to change this materially.
Altman accomplished a lot during his tenure as CEO but it demeans his accomplishments when he resorts to pulp science fiction hype.
Comment #72 February 19th, 2026 at 7:06 am
OhMyGoodness #71:
Setting aside the expense of human space travel, humans traveling in space is intrinsically worthwhile. Sending a proxy, no matter how capable, is not the same as being in the reality. It’s intrinsically worthwhile for the same reasons that scientific research is intrinsically worthwhile, or excelling in sports, or going over a mountain pass to see what is on the other side in such a way that one will sleep in, eat from, smell a different environment.
The gating factor is not the cost of going into space, but the cost of living and working in space. Once one is out of our gravity well, space travel is incredibly cheap and efficient. Raw materials are free for the taking. Energy is constant, intense sunlight. Once human presence in space is self-maintaining, human presence in space will enter a phase of exponential growth.
Anti gravity is probably not in the cards, but lightweight fusion power might perhaps be in the cards. If so, even the cost of getting out of our gravity well becomes bearable.
Comment #73 February 19th, 2026 at 2:31 pm
Michael Gogins #72
We have to reconvene in 2035 and list all the college graduates that received job offers to explore the solar system in exciting super well-payed new jobs. Do you have a timeline you believe reasonable for developing a lightweight fusion reactor suitable for manned space flight or must we also wait until 2035 to determine if that came to pass? I will pay a dollar for every grad going from graduation to a posting in space and $10 for each light weight fusion reactor in use for manned space flight. You are describing, I believe, a colonization period and not an exploratory period but anyway-
“ Anti gravity is probably not in the cards, but lightweight fusion power might perhaps be in the cards. If so, even the cost of getting out of our gravity well becomes bearable.”
Not sure what you meant by this but I assume you mean fusion reactor in space and the usual chem rockets to launch because of the thrust requirements. Also assume you envision an ion drive to provide thrust once in space.
The task will be to reduce this (Stellarator in Germany that holds current records for plasma stability at a cost of $1,8 billion)-
https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Fc8.alamy.com%2Fcomp%2FDM29GA%2Fgreifswald-germany-09th-dec-2013-mechanics-work-on-the-experimental-DM29GA.jpg&f=1&nofb=1&ipt=e1c3740503f936fb473678e293743ed1f4eb3b1204faf5715a1d7c779b67d6e1
to something that makes sense for manned space flight-neutrons and control sensitivity and all.
“ Raw materials are free for the taking”
I assume you were limiting sources for resources to the Moon, Mars, and the Asteroid Belt. What raw materials are free for the taking and where are they located.” Even oxygen recovery on the moon from regolith requires substantial energy input as do metals and even the presence of consolidated mineable water deposits is not assured.
“ Preliminary analysis of the CPR of 18 polar craters in
Mini-RF monostatic radar data suggest that a
signature indicative of surficial/near-surface water
ice is not present (i.e., at least not in quantities
detectable at S-band wavelength). This is consistent
with previous Mini-RF results from polar
observations with bistatic angles < 0.5° [10, 11]. The
CPR values of 3 other polar crater floors are elevated
with respect to their surrounding but this signature is
likely a result of surface roughness. While these data
are not consistent with large, pure water ice deposits,
we emphasize that they do not rule out smaller,
heterogenous ice deposits, or surficial frost. ”
https://meetingorganizer.copernicus.org/EPSC-DPS2019/EPSC-DPS2019-754-1.pdf
Anyway mark your calendar for June 1, 2035. The job offers for 2035 graduates should be known by that time and one of us will pay the piper.
Comment #74 February 20th, 2026 at 7:57 am
The ejecta analysis from crashed spacecraft in shaded areas on the moon are also consistent with frostlike cover of regolith grains and at about 5.6% by mass (as little as 2.7% and as much as 8.5% including potential error). To recover 1 cubic meter of water then would require (assuming 100% recovery) collection and processing of 8 tonnes regolith material from permanently shaded areas inside craters.
Comment #75 February 20th, 2026 at 8:12 am
Well, actually 8 cubic meters of regolith material.
Comment #76 February 21st, 2026 at 7:01 am
If you consider near term adoption of robot labor as a business decision then the total cost per unit of product (labor, energy, overhead, maintenance, rents, etc) must be competitive with products produced by the Chinese and must also provide sufficient profit expectation to recover relatively quickly all the capital costs related to conversion to robot labor. Passing the cheap labor Chinese hurdle is a more difficult than passing an American labor hurdle. I have no idea what the actual numbers for comparison are and guess some would be largely speculative at this point.
Assuming robots cheaper than Chinese labor and all current human labor is replaced by robots then either will retain profit seeking corporations or not. If retained then not clear how consumers would purchase products. If not retained then necessary to have some Gosplan like agency to direct production quantities for distribution to consumers. This has not worked previously in practice so maybe an AI would assume this role.
If no private property and all directed by AI then what parameters used to allocate resources. Decisions would necessarily be based on opportunity costs. If resources are used for big screen TV’s then they can’t be used for paper clips for example.
I expect these changes that must result from business decisions will be gradual and successful smaller implementations will provide sufficient data to justify larger implementations. It is a battle between robots and Chinese labor who not only work cheap but breed more workers in their downtime.
I just saw some numbers for San Francisco and Uber and Lyft with drivers less expensive than Waymo with no driver.
Comment #77 February 21st, 2026 at 12:54 pm
OMG #76,
Politicians often invoke the “cheap labor” narrative to both comfort and inflame displaced Western workers (“you’re better, but they cheat”), yet it is fundamentally flawed. China, for instance, has a higher robot-to-worker ratio than the United States.
https://ifr.org/ifr-press-releases/news/global-robot-density-in-factories-doubled-in-seven-years#downloads
You might also want to update how much Chinese workers are said to “breed” these days. In reality, China now has one of the lowest birth rates in the world.
Comment #78 February 21st, 2026 at 1:51 pm
Glassmind Duo # 76
I am well aware of China’s current fertility rate but in comparison robots have no time off and so are unable to breed. I don’t need to look it up but thanks for the suggestion.
I made no claim that China was cheating in any way so don’t understand your point. They have a resource of cheap labor and use it. I looked back at my post and have no idea as to what prompted your statements. My statements were intended as factual and with no political undertone at all. The average labor rate in China is around $320/week and in the US around $1300/week so about 4x higher.
Linked is a photo of the suicide nets at Foxconn that now prevent workers from jumping to suicide during breaks. I guess this would be a portion of their employee benefits program.
https://external-content.duckduckgo.com/iu/?u=https%3A%2F%2Fs.wsj.net%2Fpublic%2Fresources%2Fimages%2FAI-BX879_CLABOR_G_20121217115457.jpg&f=1&nofb=1&ipt=68e329190a9b9cf8395ff7f6b428e280696eb23cf7ab4a6d29d7c4266b5b9003
I don’t know what prompted your response but reason it must have some connection to politics.
Comment #79 February 21st, 2026 at 3:19 pm
I do want the US to restore its global manufacturing status to at least equal China. I hope that AI and robots can reduce manufacturing costs to be competitive. I don’t think it is possible to reduce US compensation and benefits sufficiently to compete on that basis. It’s not just labor in China that is less expensive but across the board-skilled trades, engineers, etc.
Also believe that AI and robotization of human tasks is necessary to compete militarily with China because people are more plentiful there and it’s my belief that strategic decision makers differ in their views of casualties.
I guess if this constitutes a political undertone than I am guilty.
Comment #80 February 21st, 2026 at 5:35 pm
OMG #78,
>I don’t know what prompted your response
My response was prompted by your assertion that “it is a battle between robots and Chinese labor, who not only work cheaply but also breed more workers in their downtime.” As such a claim cannot, in my view, be derived from data, I inferred that it likely emerged from a broader climate of political disinformation in which such tropes circulate and gain traction. I apologize if that inference was unfair.
Comment #81 February 21st, 2026 at 6:28 pm
Glassmind Duo #80
I disagree and assert that in order to be competitive with Chinese manufacturing the US must reduce its cost per manufactured unit produced. In order to reduce the unit cost to Chinese levels the US labor costs that are on the order of four times must be competitive on a per unit basis. How is that possible 1) Increase productivity per labor hour by 4 times while holding labor rates constant. The substantial US advantage in productivity per labor hour vs global competitors was lost long ago. 2) Reduce wages and benefits by 75% to match the Chinese. This in my view is not possible in current US culture. 3) Fully automate US manufacturing in order to reduce manufactured cost per unit by reducing direct labor costs and improving productivity per unit per time unit.. This is the most likely in my view. This will be successful of robotization can equal the output of Chinese workers at lower cost so battle between their labor and robots for manufacturing supremacy.
Will you please tell me what data in particular contradicts these comments? I don’t believe there is anything mysterious about this. China produces more cheaply because their manufactured cost is lower because their labor costs are lower.
Comment #82 February 22nd, 2026 at 1:00 am
OMG #81, I already provided data contradicting the “robots vs. cheap labor” narrative (#77), and you explicitly said you didn’t need data on Chinese demographics (#78). What remains are opinions about how best to redevelop U.S. manufacturing, an area in which I don’t feel qualified to assess..
Comment #83 February 22nd, 2026 at 11:14 am
Glassmind Duo #82
I looked at your link in #77 from the International Robotics Federation and found nothing that contradicted my statements. Chinese workers do in fact make less than US workers and that provides a manufacturing advantage. If all Chinese workers were replaced with robots and all American workers were replaced with robots then there would be parity so US in a better competitive manufacturing position than now. In that case both Chinese and American robots provide a cost advantage with respect to Chinese workers.
My statement about biological production of new workers has nothing to do with China’s fertility rate. Robots have zero biological reproduction rate so anything non-zero contributes more to the future work force than the zero from GMC Transmission Installation Specialist AI1138.
Comment #84 February 23rd, 2026 at 7:38 am
This isn’t pointedly aimed at anyone who posted in this thread and is only my current opinion. I worry that it may be time for the US to pass the baton because the educational system has produced a deeply ideological population at the expense of critical thinking. I know most here do not agree with that and so emphasize only my personal fear.
Again not pointed at Glassmind but hope that we don’t have to battle for long periods, as we have with “Can males have babies”, to answer the question-Can robots have babies.
Alex Fischer #41
Just out of interest I checked the current inventories of cell phone apps. Apple Store has almost 2 million available and Google Store about 3.6 million. The average cell phone has 40 apps. Granted these millions of apps probably have thousand of variants of Candy Crush. I saw a story that Assad spends considerable time playing Candy Crush (I never have so know nothing about it). What a coward he turned out to be. He lied to his closest friends and slipped out by air leaving his friends to face revenge.
Dr Aaronson
I apologize for my number of posts at the end of this thread and will reduce my consecutive musings.
Comment #85 February 23rd, 2026 at 11:24 pm
OMG #84, No worries & feel free to read #77 as a sign that I do share your concern about the U.S. producing a deeply ideological population.
Comment #86 February 24th, 2026 at 2:51 am
Glassmind Duo #85
Your assumption in #77 that I was presenting some political argument when presenting facts did provide an indication of your concerns.
Comment #87 February 25th, 2026 at 4:03 am
I am breaking my vow.
On the subject of China, they reportedly delivered hypersonic missiles to Iran that do pose a threat to US aircraft carriers in lieu of any (improbable) secret defense system developed by the US. We are close to seeing a top of the line near peer technology used against US naval assets.
Unfortunately the new class Gerald R Ford carrier suffers from terrible engineering to the extent that even the toilet system has been problematic from the first day of operation and this with a crew of 4,500. If the US does lose a carrier maybe it will be a strong enough wake up call.
Comment #88 February 28th, 2026 at 6:59 am
Thank Goodness. It begins. Please let it go well.
Comment #89 March 4th, 2026 at 4:01 am
I hope we can use AI to fix the problem of suffering.
Everything else is just rearranging the deckchairs.
Comment #90 March 11th, 2026 at 11:10 am
RE: useful quantum algorithms, what about Grover’s Algorithm? Square-rooting the algorithmic complexity of reversing arbitrary functions seems as if it should have a sizable domain in which intractable problems become tractable.
E.g. if you have a function that takes in a circuit description and outputs whether that circuit performs some useful but very difficult operation (like entangled quantum computation with 99.999% accuracy), then you can reverse the function to pass in “true” and get back the bits that describe that circuit. On a classical computer, reversing an arbitrary function is exponentially hard. It’s “only” half-exponentially hard with Grover’s.
The domain is limited – if it’s a one in 10^40 likelihood of finding a good answer, trimming it to one in 10^20 is maybe not helpful. But if it’s one in 10^24, trimming it to one in 10^12 may well put the answer within reach.
Comment #91 March 29th, 2026 at 7:50 pm
This is exciting news! I’m really curious to hear Nate’s take on the practical timeline for quantum computing, especially regarding RSA-2048 decryption. The jump from theoretical quantum algorithms to actually needing 100,000 physical qubits is such a sobering reality check for how far we still have to go. It’s refreshing to see someone willing to share an optimistic but grounded vision for 2050 rather than just hype up the quantum revolution.
Comment #92 April 1st, 2026 at 12:53 pm
This is exciting news! I’m really curious to hear Nate’s thoughts on the quantum computing timeline, especially given how much the estimates for breaking RSA-2048 seem to keep shifting. The jump from needing millions of qubits down to 100,000 is pretty significant, and I’d love to understand better what assumptions are driving those numbers. It’s refreshing to see someone willing to share an optimistic vision for 2050 while still engaging with the hard technical details—seems like exactly the kind of conversation we need more of in this space.
Comment #93 May 11th, 2026 at 11:05 am
Interesting that Nate Soares is talking about breaking RSA-2048 with only 100,000 physical qubits—that’s a lot lower than I’d have guessed. Also curious how his “optimistic vision for 2050” went over with the MIT crowd; would love to hear what the freewheeling discussion covered.