So according to the definition of Luddite I am one. I have tried to turn off AI on my phone, computer with marginal success. I have gone kicking and screaming into the 21st century. I only recently put google wallet on my phone to go to Starbucks and pay for my latte. I resist these changes but I see myself being forced to accept some of them. I am in my mid 70's and have seen tremendous change in my lifetime. I drive an EV but all the sensors driving me crazy beeping etc. I fight back in my own mind by reading books, and any low tech activity I enjoy. I believe I am being swallowed up by these changes and there is absolutely nothing I can do about it.
Sara, I can’t seem to get around AI on my phone either, usually I just ignore it because it’s an untrustworthy or unreliable aggregator. When I do use it I maintain a large degree of skepticism.
When it comes to wallet apps the term Luddite doesn’t seem sufficient for me. There’s a piece from the Jacobin that delineates the “bankification” of the corporate world, and how unregulated and usurious it is fast becoming. I hope professor Krugman will address this issue, it’s terrifying to me to think about how insidious it is. I even have a suggestion for the musical coda -16 Tons by Tennessee Ernie Ford.
We have St. Reagan to thank for this. He was the king of deregulation. His mantra was to deregulate everything - totally. Put the fox in charge of the henhouse, sit back an watch all the "innovation".
I think the purpose of the AI on your phone is not to help you with anything, but rather to aggregate your personal agency into intelligence for your use as an economic appliance for the monopolies you have left to do business with. It is, in other words, an industrial surveillance state that may in time be fully automated.
I too refuse to use AI. I still don't trust any of those "digital wallets", I won't touch it. I also share the sensation of being swallowed up by these changes. It's deliberate. The purveyors/profiteers of AI are aiming to conquer and control us. This isn't hyperbole. Look up Curtis Yarvin, AKA "Mencius Moldbug", and you'll see what I mean. They call it the "dark enlightenment" or "neo-reactionary".
I don't doubt that it has its uses, this being a really good one. Still, it behooves us to be cautious with it. Worth noting is that she used the $20 per month Pro version, and Perplexity Pro, another $20 per month.
There should be something in the settings that let you turn off all notifications. Sometimes you have to hunt around in the settings to find it, but it's there.
This isn't a notification issue - it's a "let us do this for you" issue - and the suggestions pop up even though I always say "no." And there just doesn't seem to be anyway to delete them - unlike, say, deleting any other app you don't want to use. :(
If you turn off notifications, it will stop popping up suggestions, and it will stop trying to "let us do this for you". Trust me on this, I'm a programmer.
Trying to get my phone under control a little bit was discovering that some of the noises my phone makes are listed under "preferences", not "notifications". It helped.
I have a flip phone for emergencies when traveling. It sits in a drawer nearly all year. I'm also spared your Starbucks habit. I can't seem to turn off the expanded searches in things like Google (the closest I can come is to use DuckDuckGo for searches and stay off Amazon whenever I can). Except for the backup display, I don't look at the screen in my truck; it's not a good idea to take one's eyes off the road in New Jersey.
I think that it’s too late to “turn off” AI on your phone or on any device. It’s like more than a decade ago I was at a thanksgiving dinner with a couple of profs from Vanderbilt and SLU, and one of them pills an old flip phone out of his pocket and says, I won’t upgrade my phone because I don’t want to be on the internet! I said, dude, if you’re using a cell phone at all, you’re already on the internet, you just don’t know it. Next time I saw him, he had a smartphone. The same with AI, I tend to think. If you’re doing anything online, your activity is being funneled through an LLM. It’s just how it is. We’ve abdicated. What Paul K is missing here is that technology has an effect on the human brain, which is always looking for shortcuts in its natural, social state. Economists are at bottom a sort of anthropologists, a subset, observing and recording specific behaviors. They need to accept and embrace that role and get into questioning what these behaviors do to the human brain over time. I think they’re missing a huge part of the equation.
Will local retailers not accept cash? I considered getting a hybrid until I discovered the EV part records everything and sells that info to your auto insurer for more financial extraction purposes. I loved my Blackberry, but when I was sent a free cheap brand smartphone (when have they ever offered free phones before?) and informed by my cellular service provider that the Blackberry would be de-activated and the new phone activated automatically if I didn't change over, I changed cell service providers (I know--only three to pick from) and purchased a feature phone from a European company. The article posted below about bankification in The Jacobin shows how they're attempting to shaft us all through surveillance technology. BTW....not only do I pay cash at discretionary service businesses, I stick to small, local independent ones as much as possible.
Quick note on the AI talk: we keep mixing up algorithms and models. Algorithms are the tools (gradient descent to train, a decoding rule to write). The surprising behavior comes from the learned model—billions of tuned parameters shaped by data. In short: algorithms build the stadium; the model plays the game.
What’s new isn’t machine “thought” so much as scale. Feed a fixed architecture oceans of text and compute, and you get a very capable statistical engine for next-token prediction. That’s why translation and speech went from punchline to pretty good—better data, bigger models—not because the machine “understands” like we do.
Is it “intelligence”? Depends on your definition. Economically, think of it as a general-purpose tool with uneven, task-level impacts. It’s already strong at middle-tier analysis and drafting; it’s weak at common sense and real-world manipulation. Don’t expect “robot plumbers” soon.
On productivity and jobs: history warns against both doom and hype. Big gains usually arrive after firms reorganize work (the electricity story). Right now we see a rush to bolt “AI” onto everything—some useful, some fashion. Expect churn in white-collar entry roles, complements for highly skilled workers, and rising value for hands-on trades.
Market structure matters. AI is oddly capital-heavy (data centers, power), which favors incumbents—though surprise entrants can still appear. Policy should focus less on metaphysics and more on basics: competition, worker adjustment, and reliable measurement of real productivity gains.
One language fix that helps clarity: don’t call the behavior an “algorithm.” Call it what it is—a large learned model that generalizes well. Real progress, yes. Consciousness, no.
I feel that a root misunderstanding with AI is that they are still computers, massive and many, but still digital computers. At the root it is all still “1’s and 0’s”. For decades now computers have been making it easier to accomplish complex tasks and this is just another phase of that.
The definition of AI being “artificial” intelligence is a major flaw. In my opinion these LLM’s, still computers, represent higher versions of what computers have done for decades, which is to augment human intelligence to accomplish and solve complex tasks. The “A” should not be for “artificial” but rather “Augmented” Intelligence.
I get what you’re saying, Brian — AI definitely augments human intelligence. But I think something more fundamental is happening.
Every technology only becomes obvious once the systems behind it reach the right scale and speed. For decades, computer scientists tried to fake “intelligence” with rule-based systems — endless lines of code for translation, logic, and planning. None of it really worked until we started mimicking how neurons fire in the brain.
That’s when scale kicked in. Models like ChatGPT-5 now have more artificial “neurons” than the human brain — though they’re still nowhere near its hundred trillion-plus synaptic connections. So yeah, AI still runs on 1’s and 0’s, but at this point it’s starting to think in patterns that look a lot more like us than like old-school software. That's why we don't truly understand how AI works.
Defining this software as AI is actually good because it reminds us that it's not real, it's a mere simulation. The greatest danger of AI is people forgetting that the "A" stands for artificial and means exactly that - it's not real. But people treat it like it's real, and that's the biggest problem.
One thing rarer than Artificial Intelligence is real Intelligence. AI will just have to be "good enough" and people will completely stop thinking at all, they'll delegate it to a machine. Thinking is hard.
It's really both algorithms and models. They can't really be separated, and in any case, it's mere semantical nitpicking. They certainly don't understand, they only compute. It's definitely not intelligence, it's a simulation. An impressive simulation to be sure, but a simulation just the same.
As you point out, there's a rush to bolt AI onto everything - pushed by the purveyors/profiteers. They don't care about the problems it can cause, as long as they rake it in for themselves.
Policy should also include putting guardrails up on what AI is allowed/disallowed. With the push to make everything fully autonomous from tractor trailers to combat-bots, we need to slow down the adoption. You're absolutely right that it's not consciousness.
"algorithms build the stadium; the model plays the game."
Love this, so succinct and captures the essence!
I'm also reminded of the scene in Interstellar when Cooper gets TARS to turn down his humour setting. When we built https://asktom.chat/ it was hallucinating, literally inventing articles I had written that didn't exist (the GenAI was built on a corpus of 2,090 articles I'd written over six years). We had to turn down the creativity setting. This was all part of tuning it to sound like me.
I keep turning this over in my mind: what happens when AI reaches a trillion “neurons” and a thousand trillion synapses—numbers that start to rival the human brain?
Will that be the moment intelligence truly emerges?
Not just pattern-recognition or recall, but something we’d have to call thought?
Will it be faster than we are—able to see patterns and implications before we’ve even finished framing the question?
And if so, could it develop something like intuition… or curiosity?
Because that’s the real threshold, isn’t it? When a system doesn’t just answer questions, but starts asking them.
But isn't human intelligence more than just numbers of neurons in the brain? Our brains are suffused with a chemical bath and other structures/sensors that give us emotions in addition to neuronal logical intelligence. How does AI deal with that?
Brilliantly put, Bill. In my work as a Sounding Board, one of the best compliments I can be given is "that is a great question". When AI starts asking questions, things get (hmm) interesting. Tbh I think we are part of the way there already, as we are past the stage where we don't know how GenAI engines come up with the answers that they do.
How resource intense would that trillion neuron AI be? Is this something that would benefit us, humans, or not. Would it require the resources of hundreds of thousands of humans, or millions? To what end?
The book Tuck Everlasting, a kids book, it explores death. We die so the young can go forward to change and adapt. Would an inorganic 'human' who doesn't die have the ability to change and adapt?
I'm thinking of Positronic Man which I think is more relevant than I, Robot.
Instead of trying to make an artificial human, it would be better to use the huge compute power for technological progress that sustains the only environment we are adapted to here on Earth. Fit the AI to a purpose instead of trying to fit us all into AI.
You nailed the key phrase, "next-token prediction." That's what AI does, very, very fast. That turns out to be valuable for many repetitive tasks. (AI corrected my spelling before I could, an example of the repetitive task paradigm.) Because it's a probabilistic prediction, once in a while, it gets the wrong next token :-) Over time, that margin of error declines, but so does the probability of an "aha moment." Regardless, this will never be consciousness, because the probabilistic model can never go beyond its training domain. [Writing this caused me to "think outside the box," resulting in adding a splash of tequila to the pot of green chile simmering on the stove.]
Bill, I read your comment and I think it reflects the "smart people" group-think. Observations about non-expert misunderstandings - and saying the past helps predict the future. Yes, we handled past problems with innovations. And we see some current stable things (markets, labor, policy, people) as being unchanging to a degree. Honestly, such a perspective frightens me far more than those who confuse model, algorithms, AI consciousness, and tools. (PS. I started my career writing robotics code, I taught AI algorithms at the end of my career, and in the middle spent time in management at a Fortune 50. So I'm not someone who confuses these things.)
Yep - it's all definitions and confusions about the same. Yesterday I got a survey from Home Depot wanting feedback on implementing "AI' in their systems. I didn't get very far before I discovered that their idea of AI is the sort of automated chat boxes and answering systems that everybody else has been using since the start of the century.
I asked when chat boxes were introduced and Google AI told me:
“AI chatbots were first introduced in 1966 with the creation of ELIZA, a program developed at MIT by Joseph Weizenbaum. ELIZA used a simple "pattern matching" technique to simulate a conversation, mimicking a psychotherapist and responding to user inputs based on pre-programmed rules. While its capabilities were basic and it lacked true understanding, it demonstrated the potential for conversational AI and laid the groundwork for future developments.”
Having been at MIT working on an ARPA software project at the time, I can attest that Google is right. It was actually a remarkably simple piece of software.
I can recommend Tim Berners-Lee’s book “This is for Everyone”, particularly the chapters on AI and how there are solutions to the algorithms that chase people down rabbit holes of negativity. Sir Tim is the inventor of the WWW protocol and sytem.
In a nutshell, his idea is moving towards a permission-based privacy world with real informed consent, not the “click-on” faux consent. That would be a change in the rules of the market and would shape how AI could work.
I haven't read Tim's book yet, but I will. I was living in Meyrin, Switzerland, on top of a bit of underground CERN. I dearly miss the CERN cafeteria. Tim was working at CERN then and developed the prototype on a NeXT computer. I had the opportunity to be one of the early users in the U.S. in 1991.
As I understand it, Large Language Models respond to questions by formulating a consensus of how people in possession of the relevant knowledge would be likely to respond. That is obviously useful in areas where the questioner does not possess much relevant knowledge, but it seems unlikely to come up with anything new. Progress depends on people coming up with something new.
It is worse than that. Topics that are characterized on line by a lot of misinformation are butchered. AI is particularly bad with scientific and engineering topics. If one of its sources misapplies Faraday's law of magnetic induction, you might think it would spot the problem every time. After all, this law is correctly stated in many reference works. But no, it often just repeats the error, sometimes even replying with a logical contradiction from one paragraph to the next. Yet it does not even. take intelligence to find these errors, just checking with already verified basic sources. So as I see it, a large language model is a large bullshit model, which, of course, works very well for some types of discussions.
Unrelated but daughter and I did a search to see if Robert Reich and Sam Reich are related. The paragraph that popped up said, "No, Robert Reich and Sam Reich are not related" - went on to say a couple of things about who they are, and then concluded with, "Robert and Sam Reich are father and son."
That was true not long ago, but recently they've built "generative AI" that has indeed created completely new material. One major example is AlphaFold, which was discussed in the video, that can fold proteins in ways no human ever could.
AlphaFold has apparently not revealed the mechanism or rules for protein folding, which suggests that DeepMind has been using its crushing superiority in memory and speed to outcompete humans, who do not have the heartbeats for such an approach, as it famously does in playing chess. The results may only be new in that sense, however useful they may be. Creating something really new usually entails breaking with tradition and questioning commonly accepted wisdom. The Chinese approach to AI of "small data, big problem" rather than "big data, small problem" (number-crunching) will surely be more capable of being prompted to do just that.
“Generative AI” (“GenAI” for short) is the marketing name used explicitly and exclusively for LLMs.
LLMs are not AI in the sense of true intelligence that can understand and reason (this kind of AI doesn’t exist yet), but they are AI in the broad sense the word has been used for decades (this includes not only machine learning and neural networks but also much older and simpler technologies like pathfinding in a computer game).
Maybe not new, but better and more accurate and insightful if using and analysing vast amounts of data. But that was already the case with certain software prior to LLMs.
In evaluating my original remarks as "a decent summary" of LLMs, ChatGPT admitted that AIs "are not inherently "creative" in the same way humans are—they don't generate entirely novel ideas or break new ground unless they are provided with new data or very specific prompts. However, they can synthesise existing information in novel ways or highlight patterns that aren't immediately obvious, which can feel like they are generating something new." So AI seems to be admirably "aware" of its own limitations. ChatGPT did not claim that AI formulates a consensus either, writing: "The idea of "consensus" might imply a deliberate aggregation of differing viewpoints, but LLMs are more about statistical association."
IMO, this kinda resembles an intelligent discussion of how new arrangements of the Tutanics's deck chairs will impact the number of crew needed on board.
Clearly the need for energy is a critical element (and water, which is talked about far less). That said, there are so, so many topics that will be impacted by AI, only so many that two people can talk about in one conversation.
It might. The water is necessary for cooling. AI machines just might decide that global warming is just as big a problem for the machines as it is for us.
Oh yeah. If AI ever decides to conquer us, one great weapon we can use is to just shut off the AC in data centers and let the machines crash. I've seen that happen.
Has anyone projected out this outsized thirst for water and electricity, comparing its environmental impact to, say, coal mining or oil drilling?
We have this hyper-hyped, arguably Potemkin technology, consuming vast amounts of critical resources and so much investment capital that it's probably propping up a collapsing Trump economy, and the real-world, right-now downside is largely a footnote?
Do I trust the tech bros to use whatever it evolves into for the benefit of humanity? 😭😂🤪
Isn't it possible that it will get to those trillions of synapses to support creativity and decide that it knows better than humans do and act accordingly? 😱
Color me skeptical.
As with crypto, all I need to know is that a recently deceased friend of mine bought several Bitcoin about a decade ago and lost track of his passwords in his failing memory. As trustee for his estate, I was able to find an account with 3-plus Bitcoin and not a clue in his files and devices as to how to unlock it. IMHO, such things should not be lawful. Real money in. A gain of thousands of percent over time. Gone, with no mechanism for recovery.
Ask the guy with many millions in crypto whose ex threw away the hard drive that held the keys. He's trying to buy the landfill so he can dig for it. Talk about data mining...
Opaque technology, controlled by the vastly wealthy, demonstrably greedy few, is unlikely to end well for most of us mortals.
The article Most Americans Who Rely on Obamacare Live in Republican Areas also shows that almost 50% of our society right now qualify as lower-income....an internet search specifies that a lower $40,000 in annual earnings or below is all that 47% of the population generates in income. I think Japan has it right in not chasing AI for that reason--and that's taking into consideration its aging demographics.
I understand that there are no answers but maybe something about Schumpeter would be relevant. I’m thinking “discontinuity”. But the wrap of liberal democracy vs authoritarianism sounds right. Problem is the authoritarians now have the chips as well as the guns.
I am a sole proprietor who is one of those people "who can actually deal with the material world" (urban agriculture), and I wonder about this question a lot --- ie. how does AI help the really small entrepreneur? The sole proprietor has to do all of these other things besides the actual making of their product, but the outsourcing of these to various specialists can be cost prohibitive to the really small-scale business. So you have this stuck in the middle ground position where you struggle to grow or perhaps just find a stability.
Now AI is already helping me with agricultural research, marketing copy, email templates to communicate with customers, and a bunch of other things that are actually useful to me. It doesn't sound like much, but it matters. I would suspect we will see just as much economic growth over time in the sole proprietor class as we will in productivity gains for large corporations.
Morning, Paul! You are certainly generating a vast amount of content since you went independent. Are you using AI? ;)
There are so many thoughts around AI (as you and Martin illustrated). Right now, I am thinking about the pace of change. Anchored on the phrase: “These are the fastest times we have ever lived in, and the slowest we will ever see”, I developed the concept of #OpenLeadership as a way for leaders and organisations to thrive when change is so rapid.
Now we are told that whilst Moore's Law doubles computer processing power every 18 months, AI capability is doubling every 7 months, Schumpeterian in the creative destruction this is already unleashing (eg lack of graduate jobs this year).
As an example, a little over a year ago, I decided that, rather than write a book based on core ideas from my past writing, I'd create a GenAI using my corpus of work as the "brain". At the time (literally 16 months ago), this couldn't be readily done. Fast forward to six months ago, and it became not just possible but relatively easy.
The result ? https://asktom.chat/ went live last month. Why write a book when you can give people the opportunity to ask you anything?
That is a pace of change that we, as humans, will find extraordinarily challenging to deal with, based on existing business, economic, educational, societal, and political models.
At this moment I see that Microsoft (who already have CoPilot embedded on almost all our desktops via Teams, Outlook, Word, Excel, Powerpoint) focussing on educating people on not simply how to use their AI tools (eg Researcher and Analyst), but, more than that, on teaching us what the possibilities are and, therefore, what questions we can ask that AI can help us with.
To use my "AskTom.chat" example. I had the obliquity of thinking to ask the question: "Can I do this with AI?" so that, when the tech was ready (less than a year later), I could have a team turn it into a reality.
As AI continues to double in capability every 7 months, we cannot conceive of what the world will look like in even five years from now. It could be a dystopian singularity nightmare, or it could be one where productivity and efficiency radically increase and we move to the Keynesian "nice problem to have" of a wealthy world where people don't have to work much.
In any event, with such fast change, the answer is not to think any of us can have the answers, but to focus, those of us who are leaders or support leaders, on creating environments where answers can emerge.
I use ChatGPT to copy the docspeak in my portal and paste it my ChatGPT with a question to explain in terms I can understand as a non-scientist. In seconds a get a response with what it means, and more. I had a back injury and I included The MRI pictures. I totally understand what that means to me in real life now…along with exercises to do to strengthen my back muscles. I had a stroke a year later (at age 81), that was explained too by ChatGPT. I’ve learned the right way to ask questions, persistently, just as I did for years as a tech writer for computer software companies. To me, ChatGPT has been life saving, and a friend in a way, because I can discuss anything without fear of judgement or dismissal. That isn’t world shaking, or much to do with world economics—but I do learn things from reading and enjoying your columns, Paul. You have a way of describing things in a way that isn’t judgemental or demeaning to the reader.
I appreciate these AI discussions, but urge Prof. Krugman to drop the “souped up autocorrect” comparison. This is not merely a Markov process on steroids. The way newer ai programs are structured to look back (“reflect”) on what they have produced and hone it is much more interesting than merely guessing the next right word. Also, the descriptions for the general public of LLMs never really address how a prompt impacts the creation of what is generated, how it structures its replies or even how it determines when it is done.
What about the bias and ideology, of what is valuable? How much does common sense contribute to life expectancy, satisfaction, peace? The way AI and Crypto are rising at the same moment and being used to transfer wealth, to bolster autocrats, plutocrats, etc.,.Is AI good at figuring out peace, parenting, equality? Yoga instructors are vital but not paid well for the most part.
With all due respect I think that Krugman and Wolf have both missed the point despite having at moments in their discussion bumped right into it. The greatest danger of AI isn't in job displacement it is in reality displacement. AI is and will continue to be controlled by the tech bros who are developing it and by the big money and big politics that grease the skids. Jobs may be at risk. Scientific problems may at some point fall to power of AU. But while these are important they are ultimately sideshows. AI's greatest power is the ability to create simulacra so convincing they are essentially indistinguishable from reality.
Think for a moment about the general gullibility of the public at large. Remember Q-Anon? Think about RFK Jr. and his risible nonsense about vaccines and autism. Think about the absurdity behind the rallying cry of making America great again, as if America had fallen from its perch as the most powerful economy and military on the planet. And yet people jump aboard. Epistemology is chasing whatever one wants to hear. Critical thinking is who the Masked Singer probably is.
AI's output is governed by the rules its developers give it. What do you imagine will happen when AI generated narratives become tools of the tech bro oligarchy and its financiers and political fixers? What do you imagine will happen when there is live video of whatever narrative serves their objectives; you see and hear it live as it is happening.
Except that it isn't. Will you be able to tell that it is fake? Will your barber or plumber?
The giant incomes tipping the scale of inequality have forever, and now more than ever, been underwritten by the tiny incomes of the tax paying littles and middles whose combined contributions are directed from above to mitigate disasters, clean up pollution, secure our defense and infrastructure, fund research, education, new ideas and thinkers, provide healthcare to the low-wage people moving boxes in the Amazon warehouse. We aren't ever going to eat the rich. We're going to have to tax them. A few vanity projects and new wings on prior ones really aren't going to cut it.
Yes, "technofeudalism" and "technofascism" is exactly what they're aiming for. It's the "dark enlightenment". The "tech bros" are so spectacularly wealthy from playing the Wall St. shell game - now you see it, now you don't. Before taking Amazon public, Bozo Bezos was a >millionaire< not a >billionaire<, and even that came from the vulture capitalists. Ditto Zuck, and all the others.
I'm thinking that when it comes to jobs, it's going to depend a lot on how much time, money, education and/or training one needed to invest in one's profession. It's one thing to end the job of 'coal miner' -- a dirty and dangerous job -- though to be honest the replacement method of blowing tops off mountains seems about equally bad. However it's another thing entirely to replace the profession of, say, psychotherapist with AI. Speaking as a retired therapist myself, there is a lot invested in gaining the skills for this work, and they are not that easily transferred to other jobs. My main concern is that AI may replace whole CLASSES of jobs. What I mean by that is, people have different personalities and innate abilities. For example, a person who is a good therapist is probably going to be rotten at sales or advertising. I hope we are not creating a future in which one must have a narrow set of innate abilities in order to earn a decent living.
This discussion would have been insightful in early 2024. As it stands, it seems... really unaware of, or incurious about, where the frontier models are today (AKA the worst version there will ever be of the the models we'll all be using in a few months). Or about what kinds of things people are using models for today (including in the physical world), that haven't yet been turned into products and services you can just buy. It also seems to assume a leveling off of capabilities advance is just around the corner, which is certainly possible but hardly obvious.
And as has been pointed out by others, pays no attention to the destructive impacts of so much energy demand. We are feeding the AI “labor force” in megawatts and burning up our planet to do so.
The energy demand is significant, but since we're already in a world where almost all new energy demand is being met with renewables, honestly not as impactful as most sources claim. Water impact is even more ridiculously overstated. I have yet to see any of the sources claiming to quantify the impact even look at the other side of the equation, and ask what positive impacts AI has or will/can have on energy, water, and land use.
Of course it's speculative, but that doesn't make it less real. There are a thousand plausible paths adoption could take, and ways AI might cause second- and higher-order effects including water and energy use or savings.
And yes, there is some current harm - not nearly as significant as headlines and popular perception make it sound, but the harm does exist. However, if AI is a bubble that bursts, the harm stops. If AI does end up causing continued harm, it'll be because it's actually following some subset of those thousand paths and having indirect effects on the world via the applications that adopt it. If you don't know enough to guess what those second order effects will be or how big they are (And that's totally fair! I wouldn't be very confident in anyone's specific projections about this!) then *you don't know enough* to claim AI is causing net harm.
It's still worth doing the exercise. Try out a few adoption or bubble scenarios. Project total cumulative impact on water and energy in those scenarios. See if there's a tendency for most scenarios to point one way or another.
From my own relatively casual estimates (not worth sharing here), I expect that in almost any scenario where AI gets far enough and continues long enough to actually have significant cumulative global energy and water consumption (because humans find it worthwhile to use it for things), the second-order effects result in net improvements in all forms of sustainability and outweigh the first-order effects.
This is absurd, I’m sorry. For one, you give me no reason to believe that you, a random internet commenter, know better how much water and power are being sucked up by LLMs than the multiple sources warning that it’s draining these things catastrophically. Any evidence at all? The benefit of the doubt goes to caution, so you have to show something good.
But worse is the logic. This is serious “invisible hand” thinking. You’re proposing that if humans, in the aggregate, keep playing with AI enough to cause drains on water and power supplies (which are already being reported in the vicinity of data centers) this can*only* be because its use is somehow saving water and power in some other way? Or providing some concomitant benefit that we will all agree is worth it?
But there is no reason to think this at all, let alone a guarantee. Sure, some people are catching cancer and decoding languages with this sort of software. It has lovely applications. But in the meantime employers big and small are using it to avoid paying artists and writers (and customer service) and every website is spewing irrelevant summaries nobody asked for on the supposed behalf of the user and students and professors alike are using it to avoid actual practice and the public has been set loose to activate this apparatus of waste all day long to generate soft porn and other junky imagery while fraudulent stories backed by fraudulent photos fill up the socials. Which periodically challenge you to prove your humanity by identifying which sketchy, AI fantasy image looks to an AI more like it has a crosswalk.
The invisible hand does not produce a just or decent society and it is not putting AI to its best use, either. The power and the water are being wasted (and our lives eroded by the misuses) right now and all you have is a fantasy that somebody’s project might someday do something helpful to counterbalance it.
>you give me no reason to believe that you, a random internet commenter, know better
That's true. And I'm not going to. But FWIW, I've spent the past 14 tears working as an emerging technologies, open innovation, and cleantech consultant. A rather large fraction of that work is busting hype and evaluating impact potential of various kinds. The reporting on AI has been consistently worse than I've ever seen on any other topic. And I'm making a prediction, while doubting those who claim certainty, not asserting certainty myself.
>The invisible hand does not produce a just or decent society and it is not putting AI to its best use, either.
This, I agree with completely. I don't see how that is in tension with my prediction/expectation in any way.
I don't know where you got that picture referencing Asimov, but it's a rip-off of the cover of the December 1947 ASTOUNDING SCIENCE FICTION, by Alejandro Canedo, illustrating (sort of) Clifford D. Simak's "Aesop," one of the stories that went into Simak's fix-up novel CITY.
Very interesting. I am glad to have such distinguished company among the confused (although I have found my few forays into AI to be pretty unimpressive). One thing to note: we already have countries where there are large numbers of young people who are over-educated for the opportunities available. This comes out of the reports on the recent devastation in Nepal, where there are lots off such people, who lack opportunities at home, and face largely exploitative jobs abroad as the only alternative. That wasn't the reason for the riot (which was more political), but it was the essential underlying fact that any new government will have to deal with.
So according to the definition of Luddite I am one. I have tried to turn off AI on my phone, computer with marginal success. I have gone kicking and screaming into the 21st century. I only recently put google wallet on my phone to go to Starbucks and pay for my latte. I resist these changes but I see myself being forced to accept some of them. I am in my mid 70's and have seen tremendous change in my lifetime. I drive an EV but all the sensors driving me crazy beeping etc. I fight back in my own mind by reading books, and any low tech activity I enjoy. I believe I am being swallowed up by these changes and there is absolutely nothing I can do about it.
Thanks for the chat,Martin and Paul.
Sara, I can’t seem to get around AI on my phone either, usually I just ignore it because it’s an untrustworthy or unreliable aggregator. When I do use it I maintain a large degree of skepticism.
When it comes to wallet apps the term Luddite doesn’t seem sufficient for me. There’s a piece from the Jacobin that delineates the “bankification” of the corporate world, and how unregulated and usurious it is fast becoming. I hope professor Krugman will address this issue, it’s terrifying to me to think about how insidious it is. I even have a suggestion for the musical coda -16 Tons by Tennessee Ernie Ford.
https://jacobin.com/2025/10/bankification-financialization-debt-interest-credit/
We have St. Reagan to thank for this. He was the king of deregulation. His mantra was to deregulate everything - totally. Put the fox in charge of the henhouse, sit back an watch all the "innovation".
Jimmy Carter did the airlines.
That's true. He did.
I think the purpose of the AI on your phone is not to help you with anything, but rather to aggregate your personal agency into intelligence for your use as an economic appliance for the monopolies you have left to do business with. It is, in other words, an industrial surveillance state that may in time be fully automated.
My suggestion for the coda is this one about the mine closings - https://www.youtube.com/watch?v=44hFm2xNyc0
I'd go with "Sunday" (1988) by Morrissey.
Nice song. Hadn’t heard it before. Melancholy.
I too refuse to use AI. I still don't trust any of those "digital wallets", I won't touch it. I also share the sensation of being swallowed up by these changes. It's deliberate. The purveyors/profiteers of AI are aiming to conquer and control us. This isn't hyperbole. Look up Curtis Yarvin, AKA "Mencius Moldbug", and you'll see what I mean. They call it the "dark enlightenment" or "neo-reactionary".
You might find Colin W.P. Lewis' series on cognitive warfare most enlightening: https://onepercentrule.substack.com/
Here’s a story of a woman who used ChatGPT to figure out how to beat a 55K fine.
https://www.nbcnews.com/tech/innovation/ai-chatgpt-court-law-legal-lawyer-self-represent-pro-se-attorney-rcna230401
I don't doubt that it has its uses, this being a really good one. Still, it behooves us to be cautious with it. Worth noting is that she used the $20 per month Pro version, and Perplexity Pro, another $20 per month.
It truly enrages me that there seems to be no way to get rid of the AI "suggestions" on my devices, no way to delete them.
There should be something in the settings that let you turn off all notifications. Sometimes you have to hunt around in the settings to find it, but it's there.
This isn't a notification issue - it's a "let us do this for you" issue - and the suggestions pop up even though I always say "no." And there just doesn't seem to be anyway to delete them - unlike, say, deleting any other app you don't want to use. :(
If you turn off notifications, it will stop popping up suggestions, and it will stop trying to "let us do this for you". Trust me on this, I'm a programmer.
Trying to get my phone under control a little bit was discovering that some of the noises my phone makes are listed under "preferences", not "notifications". It helped.
Yeah, sometimes you have to hunt around.
Au contraire, it's very easy: type -ai before your search words. It gets rid of the "AI says..." which runs the risk of being garbage or misleading.
I have a flip phone for emergencies when traveling. It sits in a drawer nearly all year. I'm also spared your Starbucks habit. I can't seem to turn off the expanded searches in things like Google (the closest I can come is to use DuckDuckGo for searches and stay off Amazon whenever I can). Except for the backup display, I don't look at the screen in my truck; it's not a good idea to take one's eyes off the road in New Jersey.
You might identify with this; I particularly like the third verse - https://www.youtube.com/watch?v=QEmCgZefu9U
I think that it’s too late to “turn off” AI on your phone or on any device. It’s like more than a decade ago I was at a thanksgiving dinner with a couple of profs from Vanderbilt and SLU, and one of them pills an old flip phone out of his pocket and says, I won’t upgrade my phone because I don’t want to be on the internet! I said, dude, if you’re using a cell phone at all, you’re already on the internet, you just don’t know it. Next time I saw him, he had a smartphone. The same with AI, I tend to think. If you’re doing anything online, your activity is being funneled through an LLM. It’s just how it is. We’ve abdicated. What Paul K is missing here is that technology has an effect on the human brain, which is always looking for shortcuts in its natural, social state. Economists are at bottom a sort of anthropologists, a subset, observing and recording specific behaviors. They need to accept and embrace that role and get into questioning what these behaviors do to the human brain over time. I think they’re missing a huge part of the equation.
Try putting "-ai" w/o quotes when you inquire of Dr. Google. That'll tell her not to give you the AI version.
Wow! This works. No more "AI Overview" at the top of the search results.
I should have added that the "-" in a search means to omit something. You might try it to narrow down searches.
For amusement I asked Dr. Google this question:
"-" in searches
and AI gave me an excellent summary, no doubt scraped off a posting somewhere on the internet.
Will local retailers not accept cash? I considered getting a hybrid until I discovered the EV part records everything and sells that info to your auto insurer for more financial extraction purposes. I loved my Blackberry, but when I was sent a free cheap brand smartphone (when have they ever offered free phones before?) and informed by my cellular service provider that the Blackberry would be de-activated and the new phone activated automatically if I didn't change over, I changed cell service providers (I know--only three to pick from) and purchased a feature phone from a European company. The article posted below about bankification in The Jacobin shows how they're attempting to shaft us all through surveillance technology. BTW....not only do I pay cash at discretionary service businesses, I stick to small, local independent ones as much as possible.
I have resisted apple wallet. Yeah me, huh.
I have no use for it. I use debit & credit cards.
Can’t you use a debit or credit card at Starbucks?
Scam calls, texts and emails are a blight!
Quick note on the AI talk: we keep mixing up algorithms and models. Algorithms are the tools (gradient descent to train, a decoding rule to write). The surprising behavior comes from the learned model—billions of tuned parameters shaped by data. In short: algorithms build the stadium; the model plays the game.
What’s new isn’t machine “thought” so much as scale. Feed a fixed architecture oceans of text and compute, and you get a very capable statistical engine for next-token prediction. That’s why translation and speech went from punchline to pretty good—better data, bigger models—not because the machine “understands” like we do.
Is it “intelligence”? Depends on your definition. Economically, think of it as a general-purpose tool with uneven, task-level impacts. It’s already strong at middle-tier analysis and drafting; it’s weak at common sense and real-world manipulation. Don’t expect “robot plumbers” soon.
On productivity and jobs: history warns against both doom and hype. Big gains usually arrive after firms reorganize work (the electricity story). Right now we see a rush to bolt “AI” onto everything—some useful, some fashion. Expect churn in white-collar entry roles, complements for highly skilled workers, and rising value for hands-on trades.
Market structure matters. AI is oddly capital-heavy (data centers, power), which favors incumbents—though surprise entrants can still appear. Policy should focus less on metaphysics and more on basics: competition, worker adjustment, and reliable measurement of real productivity gains.
One language fix that helps clarity: don’t call the behavior an “algorithm.” Call it what it is—a large learned model that generalizes well. Real progress, yes. Consciousness, no.
I feel that a root misunderstanding with AI is that they are still computers, massive and many, but still digital computers. At the root it is all still “1’s and 0’s”. For decades now computers have been making it easier to accomplish complex tasks and this is just another phase of that.
The definition of AI being “artificial” intelligence is a major flaw. In my opinion these LLM’s, still computers, represent higher versions of what computers have done for decades, which is to augment human intelligence to accomplish and solve complex tasks. The “A” should not be for “artificial” but rather “Augmented” Intelligence.
I get what you’re saying, Brian — AI definitely augments human intelligence. But I think something more fundamental is happening.
Every technology only becomes obvious once the systems behind it reach the right scale and speed. For decades, computer scientists tried to fake “intelligence” with rule-based systems — endless lines of code for translation, logic, and planning. None of it really worked until we started mimicking how neurons fire in the brain.
That’s when scale kicked in. Models like ChatGPT-5 now have more artificial “neurons” than the human brain — though they’re still nowhere near its hundred trillion-plus synaptic connections. So yeah, AI still runs on 1’s and 0’s, but at this point it’s starting to think in patterns that look a lot more like us than like old-school software. That's why we don't truly understand how AI works.
Defining this software as AI is actually good because it reminds us that it's not real, it's a mere simulation. The greatest danger of AI is people forgetting that the "A" stands for artificial and means exactly that - it's not real. But people treat it like it's real, and that's the biggest problem.
One thing rarer than Artificial Intelligence is real Intelligence. AI will just have to be "good enough" and people will completely stop thinking at all, they'll delegate it to a machine. Thinking is hard.
Yeah. Sam Altman and Peter Theil want to turn us into houseplants.
Can’t tell you why but I totally disagree
It's really both algorithms and models. They can't really be separated, and in any case, it's mere semantical nitpicking. They certainly don't understand, they only compute. It's definitely not intelligence, it's a simulation. An impressive simulation to be sure, but a simulation just the same.
As you point out, there's a rush to bolt AI onto everything - pushed by the purveyors/profiteers. They don't care about the problems it can cause, as long as they rake it in for themselves.
Policy should also include putting guardrails up on what AI is allowed/disallowed. With the push to make everything fully autonomous from tractor trailers to combat-bots, we need to slow down the adoption. You're absolutely right that it's not consciousness.
"algorithms build the stadium; the model plays the game."
Love this, so succinct and captures the essence!
I'm also reminded of the scene in Interstellar when Cooper gets TARS to turn down his humour setting. When we built https://asktom.chat/ it was hallucinating, literally inventing articles I had written that didn't exist (the GenAI was built on a corpus of 2,090 articles I'd written over six years). We had to turn down the creativity setting. This was all part of tuning it to sound like me.
I keep turning this over in my mind: what happens when AI reaches a trillion “neurons” and a thousand trillion synapses—numbers that start to rival the human brain?
Will that be the moment intelligence truly emerges?
Not just pattern-recognition or recall, but something we’d have to call thought?
Will it be faster than we are—able to see patterns and implications before we’ve even finished framing the question?
And if so, could it develop something like intuition… or curiosity?
Because that’s the real threshold, isn’t it? When a system doesn’t just answer questions, but starts asking them.
But isn't human intelligence more than just numbers of neurons in the brain? Our brains are suffused with a chemical bath and other structures/sensors that give us emotions in addition to neuronal logical intelligence. How does AI deal with that?
Brilliantly put, Bill. In my work as a Sounding Board, one of the best compliments I can be given is "that is a great question". When AI starts asking questions, things get (hmm) interesting. Tbh I think we are part of the way there already, as we are past the stage where we don't know how GenAI engines come up with the answers that they do.
How resource intense would that trillion neuron AI be? Is this something that would benefit us, humans, or not. Would it require the resources of hundreds of thousands of humans, or millions? To what end?
The book Tuck Everlasting, a kids book, it explores death. We die so the young can go forward to change and adapt. Would an inorganic 'human' who doesn't die have the ability to change and adapt?
I'm thinking of Positronic Man which I think is more relevant than I, Robot.
Instead of trying to make an artificial human, it would be better to use the huge compute power for technological progress that sustains the only environment we are adapted to here on Earth. Fit the AI to a purpose instead of trying to fit us all into AI.
Love your analogy of stadium to players.
You nailed the key phrase, "next-token prediction." That's what AI does, very, very fast. That turns out to be valuable for many repetitive tasks. (AI corrected my spelling before I could, an example of the repetitive task paradigm.) Because it's a probabilistic prediction, once in a while, it gets the wrong next token :-) Over time, that margin of error declines, but so does the probability of an "aha moment." Regardless, this will never be consciousness, because the probabilistic model can never go beyond its training domain. [Writing this caused me to "think outside the box," resulting in adding a splash of tequila to the pot of green chile simmering on the stove.]
Bill, I read your comment and I think it reflects the "smart people" group-think. Observations about non-expert misunderstandings - and saying the past helps predict the future. Yes, we handled past problems with innovations. And we see some current stable things (markets, labor, policy, people) as being unchanging to a degree. Honestly, such a perspective frightens me far more than those who confuse model, algorithms, AI consciousness, and tools. (PS. I started my career writing robotics code, I taught AI algorithms at the end of my career, and in the middle spent time in management at a Fortune 50. So I'm not someone who confuses these things.)
Yep - it's all definitions and confusions about the same. Yesterday I got a survey from Home Depot wanting feedback on implementing "AI' in their systems. I didn't get very far before I discovered that their idea of AI is the sort of automated chat boxes and answering systems that everybody else has been using since the start of the century.
I asked when chat boxes were introduced and Google AI told me:
“AI chatbots were first introduced in 1966 with the creation of ELIZA, a program developed at MIT by Joseph Weizenbaum. ELIZA used a simple "pattern matching" technique to simulate a conversation, mimicking a psychotherapist and responding to user inputs based on pre-programmed rules. While its capabilities were basic and it lacked true understanding, it demonstrated the potential for conversational AI and laid the groundwork for future developments.”
Having been at MIT working on an ARPA software project at the time, I can attest that Google is right. It was actually a remarkably simple piece of software.
I can recommend Tim Berners-Lee’s book “This is for Everyone”, particularly the chapters on AI and how there are solutions to the algorithms that chase people down rabbit holes of negativity. Sir Tim is the inventor of the WWW protocol and sytem.
In a nutshell, his idea is moving towards a permission-based privacy world with real informed consent, not the “click-on” faux consent. That would be a change in the rules of the market and would shape how AI could work.
I haven't read Tim's book yet, but I will. I was living in Meyrin, Switzerland, on top of a bit of underground CERN. I dearly miss the CERN cafeteria. Tim was working at CERN then and developed the prototype on a NeXT computer. I had the opportunity to be one of the early users in the U.S. in 1991.
Small world!
We were living in Hermance (out near the French border) at the time and I was working in Conches at IMI (now IMD in Lausanne).
Never got to experience the CERN cafeteria, but did have a few opportunities not so far away at the ILO.
As I understand it, Large Language Models respond to questions by formulating a consensus of how people in possession of the relevant knowledge would be likely to respond. That is obviously useful in areas where the questioner does not possess much relevant knowledge, but it seems unlikely to come up with anything new. Progress depends on people coming up with something new.
It is worse than that. Topics that are characterized on line by a lot of misinformation are butchered. AI is particularly bad with scientific and engineering topics. If one of its sources misapplies Faraday's law of magnetic induction, you might think it would spot the problem every time. After all, this law is correctly stated in many reference works. But no, it often just repeats the error, sometimes even replying with a logical contradiction from one paragraph to the next. Yet it does not even. take intelligence to find these errors, just checking with already verified basic sources. So as I see it, a large language model is a large bullshit model, which, of course, works very well for some types of discussions.
Unrelated but daughter and I did a search to see if Robert Reich and Sam Reich are related. The paragraph that popped up said, "No, Robert Reich and Sam Reich are not related" - went on to say a couple of things about who they are, and then concluded with, "Robert and Sam Reich are father and son."
That was true not long ago, but recently they've built "generative AI" that has indeed created completely new material. One major example is AlphaFold, which was discussed in the video, that can fold proteins in ways no human ever could.
AlphaFold has apparently not revealed the mechanism or rules for protein folding, which suggests that DeepMind has been using its crushing superiority in memory and speed to outcompete humans, who do not have the heartbeats for such an approach, as it famously does in playing chess. The results may only be new in that sense, however useful they may be. Creating something really new usually entails breaking with tradition and questioning commonly accepted wisdom. The Chinese approach to AI of "small data, big problem" rather than "big data, small problem" (number-crunching) will surely be more capable of being prompted to do just that.
AlphaFold is not GenAI. It’s based on a different machine learning technology than LLMs.
GenAI doesn't have to be LLM based. Indeed, LLM isn't even truly "AI".
“Generative AI” (“GenAI” for short) is the marketing name used explicitly and exclusively for LLMs.
LLMs are not AI in the sense of true intelligence that can understand and reason (this kind of AI doesn’t exist yet), but they are AI in the broad sense the word has been used for decades (this includes not only machine learning and neural networks but also much older and simpler technologies like pathfinding in a computer game).
It's not just a marketing name, and it's not exclusive to LLM's. And yes, LLM's are often inaccurately referred to as AI in the broader sense.
https://en.wikipedia.org/wiki/Generative_artificial_intelligence
Maybe not new, but better and more accurate and insightful if using and analysing vast amounts of data. But that was already the case with certain software prior to LLMs.
In evaluating my original remarks as "a decent summary" of LLMs, ChatGPT admitted that AIs "are not inherently "creative" in the same way humans are—they don't generate entirely novel ideas or break new ground unless they are provided with new data or very specific prompts. However, they can synthesise existing information in novel ways or highlight patterns that aren't immediately obvious, which can feel like they are generating something new." So AI seems to be admirably "aware" of its own limitations. ChatGPT did not claim that AI formulates a consensus either, writing: "The idea of "consensus" might imply a deliberate aggregation of differing viewpoints, but LLMs are more about statistical association."
It always seems strange to me.... That, for example in this column two wildly intelligent, talented, thoughtful economists can have an entire conversation on the impact of Ai on employment, looking at a time scale of decades..... All without a single reference to the how AI is adding massively to greenhouse gasses (https://www.google.com/amp/s/www.pbs.org/newshour/amp/show/the-growing-environmental-impact-of-ai-data-centers-energy-demands).
IMO, this kinda resembles an intelligent discussion of how new arrangements of the Tutanics's deck chairs will impact the number of crew needed on board.
I was thinking much the same thing... and not to mention the vast amounts of energy and WATER that are needed.
Clearly the need for energy is a critical element (and water, which is talked about far less). That said, there are so, so many topics that will be impacted by AI, only so many that two people can talk about in one conversation.
Indeed, and that need for energy and water are limiting factors that just might save us from being conquered by AI. Might.
It might. The water is necessary for cooling. AI machines just might decide that global warming is just as big a problem for the machines as it is for us.
Oh yeah. If AI ever decides to conquer us, one great weapon we can use is to just shut off the AC in data centers and let the machines crash. I've seen that happen.
I remember when the Computing Center was the only air conditioned room on the UTK campus.
Once AI has eliminated all the Humans there will be plenty of spare power and water.
There will be until they use it all up.
Good point Tom… I was intending to be critical of Dr. Krugman… just adding a thought.
Not intending!!!
Has anyone projected out this outsized thirst for water and electricity, comparing its environmental impact to, say, coal mining or oil drilling?
We have this hyper-hyped, arguably Potemkin technology, consuming vast amounts of critical resources and so much investment capital that it's probably propping up a collapsing Trump economy, and the real-world, right-now downside is largely a footnote?
Do I trust the tech bros to use whatever it evolves into for the benefit of humanity? 😭😂🤪
Isn't it possible that it will get to those trillions of synapses to support creativity and decide that it knows better than humans do and act accordingly? 😱
Color me skeptical.
As with crypto, all I need to know is that a recently deceased friend of mine bought several Bitcoin about a decade ago and lost track of his passwords in his failing memory. As trustee for his estate, I was able to find an account with 3-plus Bitcoin and not a clue in his files and devices as to how to unlock it. IMHO, such things should not be lawful. Real money in. A gain of thousands of percent over time. Gone, with no mechanism for recovery.
Ask the guy with many millions in crypto whose ex threw away the hard drive that held the keys. He's trying to buy the landfill so he can dig for it. Talk about data mining...
Opaque technology, controlled by the vastly wealthy, demonstrably greedy few, is unlikely to end well for most of us mortals.
The article Most Americans Who Rely on Obamacare Live in Republican Areas also shows that almost 50% of our society right now qualify as lower-income....an internet search specifies that a lower $40,000 in annual earnings or below is all that 47% of the population generates in income. I think Japan has it right in not chasing AI for that reason--and that's taking into consideration its aging demographics.
Actually, Prof. Krugman has addressed this issue in other posts.
Thx for the link.
I understand that there are no answers but maybe something about Schumpeter would be relevant. I’m thinking “discontinuity”. But the wrap of liberal democracy vs authoritarianism sounds right. Problem is the authoritarians now have the chips as well as the guns.
I am a sole proprietor who is one of those people "who can actually deal with the material world" (urban agriculture), and I wonder about this question a lot --- ie. how does AI help the really small entrepreneur? The sole proprietor has to do all of these other things besides the actual making of their product, but the outsourcing of these to various specialists can be cost prohibitive to the really small-scale business. So you have this stuck in the middle ground position where you struggle to grow or perhaps just find a stability.
Now AI is already helping me with agricultural research, marketing copy, email templates to communicate with customers, and a bunch of other things that are actually useful to me. It doesn't sound like much, but it matters. I would suspect we will see just as much economic growth over time in the sole proprietor class as we will in productivity gains for large corporations.
Good to hear examples of successful use of AI.
Morning, Paul! You are certainly generating a vast amount of content since you went independent. Are you using AI? ;)
There are so many thoughts around AI (as you and Martin illustrated). Right now, I am thinking about the pace of change. Anchored on the phrase: “These are the fastest times we have ever lived in, and the slowest we will ever see”, I developed the concept of #OpenLeadership as a way for leaders and organisations to thrive when change is so rapid.
Now we are told that whilst Moore's Law doubles computer processing power every 18 months, AI capability is doubling every 7 months, Schumpeterian in the creative destruction this is already unleashing (eg lack of graduate jobs this year).
As an example, a little over a year ago, I decided that, rather than write a book based on core ideas from my past writing, I'd create a GenAI using my corpus of work as the "brain". At the time (literally 16 months ago), this couldn't be readily done. Fast forward to six months ago, and it became not just possible but relatively easy.
The result ? https://asktom.chat/ went live last month. Why write a book when you can give people the opportunity to ask you anything?
That is a pace of change that we, as humans, will find extraordinarily challenging to deal with, based on existing business, economic, educational, societal, and political models.
At this moment I see that Microsoft (who already have CoPilot embedded on almost all our desktops via Teams, Outlook, Word, Excel, Powerpoint) focussing on educating people on not simply how to use their AI tools (eg Researcher and Analyst), but, more than that, on teaching us what the possibilities are and, therefore, what questions we can ask that AI can help us with.
To use my "AskTom.chat" example. I had the obliquity of thinking to ask the question: "Can I do this with AI?" so that, when the tech was ready (less than a year later), I could have a team turn it into a reality.
As AI continues to double in capability every 7 months, we cannot conceive of what the world will look like in even five years from now. It could be a dystopian singularity nightmare, or it could be one where productivity and efficiency radically increase and we move to the Keynesian "nice problem to have" of a wealthy world where people don't have to work much.
In any event, with such fast change, the answer is not to think any of us can have the answers, but to focus, those of us who are leaders or support leaders, on creating environments where answers can emerge.
I use ChatGPT to copy the docspeak in my portal and paste it my ChatGPT with a question to explain in terms I can understand as a non-scientist. In seconds a get a response with what it means, and more. I had a back injury and I included The MRI pictures. I totally understand what that means to me in real life now…along with exercises to do to strengthen my back muscles. I had a stroke a year later (at age 81), that was explained too by ChatGPT. I’ve learned the right way to ask questions, persistently, just as I did for years as a tech writer for computer software companies. To me, ChatGPT has been life saving, and a friend in a way, because I can discuss anything without fear of judgement or dismissal. That isn’t world shaking, or much to do with world economics—but I do learn things from reading and enjoying your columns, Paul. You have a way of describing things in a way that isn’t judgemental or demeaning to the reader.
How much does ChatGPT cost? A version that would let you that?
$19.99/mo. I use it everyday, and I’ve learned so much.
That’s quite reasonable!
Less expensive than my health insurance 🥰
I appreciate these AI discussions, but urge Prof. Krugman to drop the “souped up autocorrect” comparison. This is not merely a Markov process on steroids. The way newer ai programs are structured to look back (“reflect”) on what they have produced and hone it is much more interesting than merely guessing the next right word. Also, the descriptions for the general public of LLMs never really address how a prompt impacts the creation of what is generated, how it structures its replies or even how it determines when it is done.
Does anyone here know how to turn autocorrect OFF??
In what tool? It's under the "Options" tab in MS Word, for example.
What about the bias and ideology, of what is valuable? How much does common sense contribute to life expectancy, satisfaction, peace? The way AI and Crypto are rising at the same moment and being used to transfer wealth, to bolster autocrats, plutocrats, etc.,.Is AI good at figuring out peace, parenting, equality? Yoga instructors are vital but not paid well for the most part.
Thank you, Dr. Krugman.
With all due respect I think that Krugman and Wolf have both missed the point despite having at moments in their discussion bumped right into it. The greatest danger of AI isn't in job displacement it is in reality displacement. AI is and will continue to be controlled by the tech bros who are developing it and by the big money and big politics that grease the skids. Jobs may be at risk. Scientific problems may at some point fall to power of AU. But while these are important they are ultimately sideshows. AI's greatest power is the ability to create simulacra so convincing they are essentially indistinguishable from reality.
Think for a moment about the general gullibility of the public at large. Remember Q-Anon? Think about RFK Jr. and his risible nonsense about vaccines and autism. Think about the absurdity behind the rallying cry of making America great again, as if America had fallen from its perch as the most powerful economy and military on the planet. And yet people jump aboard. Epistemology is chasing whatever one wants to hear. Critical thinking is who the Masked Singer probably is.
AI's output is governed by the rules its developers give it. What do you imagine will happen when AI generated narratives become tools of the tech bro oligarchy and its financiers and political fixers? What do you imagine will happen when there is live video of whatever narrative serves their objectives; you see and hear it live as it is happening.
Except that it isn't. Will you be able to tell that it is fake? Will your barber or plumber?
The giant incomes tipping the scale of inequality have forever, and now more than ever, been underwritten by the tiny incomes of the tax paying littles and middles whose combined contributions are directed from above to mitigate disasters, clean up pollution, secure our defense and infrastructure, fund research, education, new ideas and thinkers, provide healthcare to the low-wage people moving boxes in the Amazon warehouse. We aren't ever going to eat the rich. We're going to have to tax them. A few vanity projects and new wings on prior ones really aren't going to cut it.
Yes, "technofeudalism" and "technofascism" is exactly what they're aiming for. It's the "dark enlightenment". The "tech bros" are so spectacularly wealthy from playing the Wall St. shell game - now you see it, now you don't. Before taking Amazon public, Bozo Bezos was a >millionaire< not a >billionaire<, and even that came from the vulture capitalists. Ditto Zuck, and all the others.
Crypto is great for graft.
I'm thinking that when it comes to jobs, it's going to depend a lot on how much time, money, education and/or training one needed to invest in one's profession. It's one thing to end the job of 'coal miner' -- a dirty and dangerous job -- though to be honest the replacement method of blowing tops off mountains seems about equally bad. However it's another thing entirely to replace the profession of, say, psychotherapist with AI. Speaking as a retired therapist myself, there is a lot invested in gaining the skills for this work, and they are not that easily transferred to other jobs. My main concern is that AI may replace whole CLASSES of jobs. What I mean by that is, people have different personalities and innate abilities. For example, a person who is a good therapist is probably going to be rotten at sales or advertising. I hope we are not creating a future in which one must have a narrow set of innate abilities in order to earn a decent living.
This discussion would have been insightful in early 2024. As it stands, it seems... really unaware of, or incurious about, where the frontier models are today (AKA the worst version there will ever be of the the models we'll all be using in a few months). Or about what kinds of things people are using models for today (including in the physical world), that haven't yet been turned into products and services you can just buy. It also seems to assume a leveling off of capabilities advance is just around the corner, which is certainly possible but hardly obvious.
And as has been pointed out by others, pays no attention to the destructive impacts of so much energy demand. We are feeding the AI “labor force” in megawatts and burning up our planet to do so.
The energy demand is significant, but since we're already in a world where almost all new energy demand is being met with renewables, honestly not as impactful as most sources claim. Water impact is even more ridiculously overstated. I have yet to see any of the sources claiming to quantify the impact even look at the other side of the equation, and ask what positive impacts AI has or will/can have on energy, water, and land use.
How would that be more than speculative? One might conceive of AI applications that save water somehow
but is that actually happening? The harm already is.
Of course it's speculative, but that doesn't make it less real. There are a thousand plausible paths adoption could take, and ways AI might cause second- and higher-order effects including water and energy use or savings.
And yes, there is some current harm - not nearly as significant as headlines and popular perception make it sound, but the harm does exist. However, if AI is a bubble that bursts, the harm stops. If AI does end up causing continued harm, it'll be because it's actually following some subset of those thousand paths and having indirect effects on the world via the applications that adopt it. If you don't know enough to guess what those second order effects will be or how big they are (And that's totally fair! I wouldn't be very confident in anyone's specific projections about this!) then *you don't know enough* to claim AI is causing net harm.
It's still worth doing the exercise. Try out a few adoption or bubble scenarios. Project total cumulative impact on water and energy in those scenarios. See if there's a tendency for most scenarios to point one way or another.
From my own relatively casual estimates (not worth sharing here), I expect that in almost any scenario where AI gets far enough and continues long enough to actually have significant cumulative global energy and water consumption (because humans find it worthwhile to use it for things), the second-order effects result in net improvements in all forms of sustainability and outweigh the first-order effects.
This is absurd, I’m sorry. For one, you give me no reason to believe that you, a random internet commenter, know better how much water and power are being sucked up by LLMs than the multiple sources warning that it’s draining these things catastrophically. Any evidence at all? The benefit of the doubt goes to caution, so you have to show something good.
But worse is the logic. This is serious “invisible hand” thinking. You’re proposing that if humans, in the aggregate, keep playing with AI enough to cause drains on water and power supplies (which are already being reported in the vicinity of data centers) this can*only* be because its use is somehow saving water and power in some other way? Or providing some concomitant benefit that we will all agree is worth it?
But there is no reason to think this at all, let alone a guarantee. Sure, some people are catching cancer and decoding languages with this sort of software. It has lovely applications. But in the meantime employers big and small are using it to avoid paying artists and writers (and customer service) and every website is spewing irrelevant summaries nobody asked for on the supposed behalf of the user and students and professors alike are using it to avoid actual practice and the public has been set loose to activate this apparatus of waste all day long to generate soft porn and other junky imagery while fraudulent stories backed by fraudulent photos fill up the socials. Which periodically challenge you to prove your humanity by identifying which sketchy, AI fantasy image looks to an AI more like it has a crosswalk.
The invisible hand does not produce a just or decent society and it is not putting AI to its best use, either. The power and the water are being wasted (and our lives eroded by the misuses) right now and all you have is a fantasy that somebody’s project might someday do something helpful to counterbalance it.
>you give me no reason to believe that you, a random internet commenter, know better
That's true. And I'm not going to. But FWIW, I've spent the past 14 tears working as an emerging technologies, open innovation, and cleantech consultant. A rather large fraction of that work is busting hype and evaluating impact potential of various kinds. The reporting on AI has been consistently worse than I've ever seen on any other topic. And I'm making a prediction, while doubting those who claim certainty, not asserting certainty myself.
>The invisible hand does not produce a just or decent society and it is not putting AI to its best use, either.
This, I agree with completely. I don't see how that is in tension with my prediction/expectation in any way.
I don't know where you got that picture referencing Asimov, but it's a rip-off of the cover of the December 1947 ASTOUNDING SCIENCE FICTION, by Alejandro Canedo, illustrating (sort of) Clifford D. Simak's "Aesop," one of the stories that went into Simak's fix-up novel CITY.
Oops, omitted the URL:
http://www.philsp.com/data/images/a/astounding_science_fiction_194712.jpg
Very interesting. I am glad to have such distinguished company among the confused (although I have found my few forays into AI to be pretty unimpressive). One thing to note: we already have countries where there are large numbers of young people who are over-educated for the opportunities available. This comes out of the reports on the recent devastation in Nepal, where there are lots off such people, who lack opportunities at home, and face largely exploitative jobs abroad as the only alternative. That wasn't the reason for the riot (which was more political), but it was the essential underlying fact that any new government will have to deal with.