Krugman stated he couldn't think of an example of a technology being forced upon workers before AI. Yikes!!! On the plus side, computerized accounting tools were and are great and boost productivity. In my field, healthcare, computerized records are a huge net negative to productivity where documentation time has exploded and has increased burnout and undermined the relationship between providers and patients largely because we are almost all documenting in real time instead of just listening and talking to patients and the documentation programs are absurdly complicated and user unfriendly. Of course, there are pluses to having electronic records, if they are accurate and comprehensive, but on net, it has been an unmitigated disaster for patient care.
One well paying job in healthcare is coding of patient's various diagnoses which could be done with good computer programs, but the generalized models of AI are useless for that. Programs that are individualized for specific tasks, like medical coding, voice to text, accounting, tax preparation, reading MRI and CT scans, differential diagnosis, weather forecasting, power grid management, etc are where better programs would be overwhelming positive for efficiency but none of them require the large language models where most of the investment is going. Better old fashioned industry based programs are what the world needs in general.
Your healthcare example is absolutely salient. I worked as a translator in the pharmaceutical industry. 100% accuracy was crucial. Approximations or 90% accuracy just don't cut it, either in original data/information gathering or in translation. But we are so far down the food chain and laughed at constantly for being luddites. I think we've just turned into a generation of Cassandras.
Yes, and also 10-15 years ago when health systems, primarily in hospitals but also in their medical pipelines, compelled workers, even physicians to adopt electronic medical records. I mean, this was a great thing, truly (thanks, Obama!), but it was hard to learn and devastating to employees who were not up to the challenge.
I experience this every time I see my pcp....he struggles with all the unnecessary clicks of the computer model while focusing on his screen and not the patient
A couple of weeks ago, I had a consultation with the doctor who would be doing a biopsy of a mass in my lungs. He asked permission ahead of time and then recorded our conversation, rather than typing as we talked. So I had his full attention for the duration of the consultation.
I certainly hope he was paying attention to you and giving you the best of his knowledge. My best to you, as you navigate this journey. Two-time breast cancer survivor here.
I'm an IT person who has setup electronic medical records - yeah they're readable and things like that but they are also time intensive, much more time intensive than the old way. And the selling point that you can port your medical data to different practices - not so much. You can get labs and test results that import in - that is great - but if you change you PCP you end up with a PDF file to the new practice. Lots more work needs to be done to make EMR truly useful
We have replaced the physician with a vastly less medically educated staff, aided by AI, who won't open the door, notice the patient's change in gait, won't notice if a family member is in the waiting room or if the patient's attire is seasonally appropriate, won't ask the seemingly irrelevant question that leads to a correct diagnosis or other issues that could be addressed. Because the people making these decisions either don't care or don't comprehend good medicine, or are so highly specialized that they don't appreciate the importance of the generalist who knows a lot about a lot. I think of the Radiologist who down-played the 86 minute average after hours obligatory "pajama time" primary care givers spent on EHR, ignoring the time most also chose to spend digesting new treatment and diagnostic information relative to their patient population.
As someone who uses an EMR on a daily basis, I couldn't agree more with your healthcare example! Another example of technology being forced on us is the "learning management systems", or LMSs, like Canvas and Blackboard. Try teaching a course at a university that uses an LMS and NOT utilize the LMS at all for your course. Hah.
Canvas was hacked a week or two ago and was inaccessible at most or all California universities - from the community colleges to the University of California (E.g., UCLA and Cal) to the private universities like USC. It was pandemonium for a day or two as students couldn't turn in required work and couldn't access their assignments, including exams that they were supposed to be taking.
Do not FIRE THE HUMANS! We will always need the Dr. Heather Cox Richardson’s and Dr. Paul Krugman’s to educate us on the truths, semi-truths and idiosyncrasies of knowledge.
I feel like we're not seeing the forest for the trees. Meta just laid off 8000 employees replacing them with AI. That's just the tip of the Zucker-berg as AI, robotics and automation have a big future coming for your job, especially the white collar job you went to college and into debt for.
It's an entrepreneur's "Productivity" dream. Reduce or eliminate your human payroll. No more SDI/Medicare employer contributions, no more Fed, state, local payroll taxes. Cool! Trades might be a good way to go, at least until AI robots can crawl underneath and fix your kitchen sink.👾🔧
Computerized medical records have allowed me, the patient - the customer, to take better control of my medical care. They allow me to change providers or seek out additional resources without having to redo tests or hassle with trying to move records around. My care team gets a unified and complete view of my situation. I think it saves lives and saves money.
True. True. True. So strange to be inpatient and have the nurse wheel in her computer & ask questions of you while looking only at the screen. I believe nurses have lost their patient assessment skills which used to derive from looking at, listening to and taking vitals while using intuitive, diagnostic & practical skills to get a true real-time mental, emotional & physical status of the patient - while, duh, providing a compassionate presence in the patient’s day.
Thanks WGG. Every word in your comment resonates with me (previously a physician in the British NHS) 😥. Work took 2-3 times as long when computerisation was introduced (partly because there was effectively no training, a tradition in Britain called ‘see one, do one!’) and then…those perennial freeze-moments of the screen, and the glitches with staggered roll-outs of new software. What retail can do brilliantly doesn’t work in the state sector.
I sleep much better now that I’ve left that behind me.
Ps: how about ‘telephone consultations’ with frozen screens/ no connection, anyone? How confidant does that make you feel as a patient 😥It isn’t much different to the pilot and co-pilot both going blind and working on memory of that route ✈️
The burnout you’re describing may have less to do with technology itself and more to do with our broken healthcare system. When used well, technology can actually make care more efficient and improve the experience for both patients and providers.
The example that occurred to me was in the early Industrial Revolution in Britain, factories with power looms were being built that replaced hundreds, or maybe thousands, of home weavers. This resulted in arson & quite bloody resistance on the part of the displaced weavers. It’s not clear to me that the workers in the power looms were factory were enthusiastic about working with that technology, but it often was the only job in town for them AFAIK.
The billionaires at the apex of the AI revolution leave me in grave doubt as to its ultimate use for the commoners, i.e., we the people. Our welfare be damned. We must be active and not be discouraged despite all the evidence to the contrary as reported by the old guard of the media that has been co-opted.
First allow me to express my admiration and appreciation for what you are both doing. It is invaluable. But please also allo
w me one comment here. You both speak about AI. But you are mostly speaking about generative AI. As a retired translator living in Berlin I know exactly how that is completely reshaping any profession that is text-based. Generative AI is not creative, it simply rearranged the pieces in the LLMs which are filled with stolen text. And in the end it will never stop hallucinating.
You've gotten it exactly right, Mary! The hucksters selling generative "A" make no distinction between their products, for which there is no known process for measuring and hence minimizing errors, and Machine Learning, for which there is. The example Paul cited, pulling numbers from a .pdf table, is little more than a ML classifier, that is to say an improved version of what used to be called an OCR. The level hype surrounding "AI" is truly astounding.
From the moment ChatGPT first came out, I've been shouting to the rafters that the greatest danger of AI is people forgetting that the "A" in AI stands for "Artificial" and means exactly that. It's just an emulation, it's not real.
Agree and often point that out. But, also disparage the intelligence aspect as it is the sum total of whomever wrote the instructions to the programmer who compiled it on a supercomputer all of which may carry less than perfect aspects of these folks intelligence. So another emulation?
I agree that one of the weaknesses of AI systems both historically and currently is that they typically have no real awareness of the limitations of their knowledge. So they will very confidently give answers even when they really don't know what they are talking about.
In general people have a better sense of whether they know something about a topic ("I'm not good at math", "I don't know European history", etc.), whereas computer programs don't generally have that feature. There is a branch of AI research into auto-epistemic systems that examines the type of reasoning required, but the main LLM models don't really have that feature.
But I'll also observe that there is also the Dunning-Kruger effect in people where, unfortunately, humans with limited knowledge of a subject can be supremely confident in their (in)abilities in said subject. So hallucinations or ignorance or misunderstanding is also something one needs to deal with in regular human interaction as well. The problem really lies in the fallacy that computers must be smart and infallible, when they really aren't.
"Some of my friends who actually work on this, like Henry Farrell, say that AI is a social technology. It’s basically agglomerating what lots of people have said. And it’s delivering back to you what a lot of people who know something about a subject would say if asked the question you asked. And it’s not understanding. There’s no mind there."
Exactly: so complete the feedback loop. Ai tells you what people are saying, and then people say what Ai tells them. This does not sound like a way to increase actual knowledge and understanding.
The old saying about computer output was "garbage in, garbage out." This is more like: "garbage out; GARBAGE OUT..."
I think many thoughtful people are on the fence about AI right now, and I also think your point about the feedback loop is important.
There is something concerning about a world where AI increasingly reflects collective human output back to us, while humans simultaneously begin shaping their thoughts, language, and assumptions around what AI produces. That kind of recursive loop could absolutely flatten originality, nuance, and deeper understanding if we engage with it passively.
At the same time, I do not think the issue is AI alone. Humans have always existed inside feedback systems shaped by culture, media, institutions, ideology, and social pressures. AI may be accelerating and amplifying dynamics that were already present.
What concerns me most is not simply the technology itself, but the broader social and economic systems shaping how it is developed and deployed. Tools like this can support learning, synthesis, and creativity, but they can also reinforce dominant narratives, reward conformity, and weaken critical thinking if we stop questioning both the machine and ourselves.
I suspect the real challenge ahead is whether we can develop the consciousness, discernment, and ethical maturity to use these technologies without surrendering the deeper human capacities that generate wisdom, insight, imagination, and genuine understanding.
I've been saying for some time that AI is becoming more and more self-referential. Let's say someone develops a website entirely using AI-generated information. Then the AI bots scrape that site. The next time someone asks a question about the type of information contained on that site, the bots will spit out information from that AI-generated site. The next person generates another website based on that information. The bots scrape it. And the cycle goes on and on and on. Pretty soon, we are swimming in a sea of AI slop.
Regarding Prof. Krugman's final comment about data centers potentially becoming outmoded, one of my neighbors may be prescient. She is the longtime president of a community organization that evaluates land use proposals in my area, and sometimes remonstrates against them. A new data center was proposed to locate near my neighborhood. Of course, there was the usual discussion about noise, water and electricity use, and air pollution, but she raised another issue. What would happen to the data center if it was no longer used? Would it just sit there deteriorating, leading to neighborhood decay? She wanted a commitment that if the business no longer needed the data center, it would demolish the building and restore the land. That's an important issue for everyone concerned about data centers to raise.
By the way, due to opposition, that data center proposal was withdrawn.
I am impressed by what sounds like a superior Chinese AI developmental trajectory. Me, I don't care if an AI voice can replace a human voice in customer service, and thereby make customer service "more efficient." Any AI model of any kind performing anything but the simplest customer servicer role, whether in a Chat or answering a phone inquiry, contributes to the ever deepening, alienating, and degrading quality of our social mores and our morals. Human beings are healthy beings only when they belong to, and feel a part of, a community. AI is adding to the sick American individualism that is always whispering in our ears, "You're on your own."
“Any AI model of any kind performing anything but the simplest customer servicer role, whether in a Chat or answering a phone inquiry…”
In my experience AI can’t even do simple customer service well. I generally wind up in an endless do loop and eventually give up. It makes me wonder if the real cost cutting benefit to corporations is a reduction in warranty payouts when customers no longer get their defective purchases fixed or replaced because they give up trying to get help. Of course that would be a short term cost savings because the customer will be unwilling to buy again from that company, but we all know companies have not thought medium or long term for nearly a half century now.
So true about customer service and ending up in a loop of repeated responses that does not address your concern. The other night after pointing this out to my chatter I asked for a live response. The entire mess took me two days of wasted time that could and should have been solved with one phone call.
You are describing my customer service experiences pretty much to a tee. The bots always start off either asking you for info you have already input, or repeating the info that you just used in writing your query. Most of the time the canned suggestions they offer (basically responses to their FAQs) do not work, or are not capable of addressing one's specific problem. Although working through a problem with someone at a SE Asian call center can be frustrating (mainly because the patois is not familiar to my ears), I much prefer that to a bot because they are trying to understand the issue even though they may lack the training or authority to address it to my satisfaction. There is always the option of asking to speak with a supervisor. It is one thing to use a bot to steer a customer/patient through a telephone tree of options so that they end up speaking with the correct department; it seems a bridge too far to ask a bot to solve actual customer problems. "Penny wise and pound foolish," as my Mom used to say; and generating lots of dissatisfied customers.
Ironically the best customer service AI I've interacted with has been Starlink. (Clearly not Grok on the back end, at least as of last summer). It takes what I tell it, asks clarifying questions, efficiently summarizes it all, and sends both the summary and the original off to a human who then generally doesn't need to interact with me further.
AnthonyCV, My son tells me that Starlink was designed for war/battlefield applications. If accurate, could that be the reason that its customer service is so crisp and efficient?
Not sure what your son is referring to, but I do not think he is historically correct about what Starlink was designed for. I would also be extremely surprised if military customers are using anything like the civilian customer service process/interface. Plus, a lot of SpaceX decisions about use of Starlink in war zones and otherwise conflicted territories seem to just be Musk's whims.
When I access customer service, I want to talk to an actual person. The computers (whether in a a chat or replacing the human voice) are only programmed for a certain limited set of questions and can rarely respond with intelligence to anything even slightly different from their set of four or five anticipated questions... Some companies have made it literally impossible to access an actual person, even that one at a call center in India, who can at least move you up the chain if they aren't able to answer your question. When I called a company with a question about deliveries the other day and was immediately connected to a human being, it was totally amazing. Customer loyalty for life!
40+ years in tech (programming, building systems, etc.) & my gut says current AI craze is a ruse. Very much like the dot com bubble & the old Enron commercial w/ the two fellas in a field saying they saw "something" alluding to seeing something great but what was it? Well we know how Enron turned out. Different but similar to the AI story. It vexes me when I hear the coined "cloud" in terms of the internet. No, data is where it has always been, in a server, somewhere, not in the freaking sky on a cloud. It perturbs me, like the Enron commercial, as it seems someone is trying to romanticize & obscure what the internet does (collect data for search engine servers) and these servers are located (literally any & every where across the globe).
Ok, so we're disconnected about where our info actually is and now, poof, the new & improved search engine is called AI. In most cases, AI on a basic level, serves only as a web scraper that synthesizes search requests & feeds it back in a communicative format. It sorts through years of collected data & spits out what it deems common knowledge (but is it?...HUGE question for another time.). That's the easy stuff, long story short, I believe there is a place for "AI" (dislike that term too lol) but it is going to be reserved for developing specific data libraries for specialized applications & therein find it's glory and not this massive hodgepodge of whatever it is being constructed.
"If it seems too good to be true..." - Thomas Lupton
Thank you. I too am a programmer, and what you describe here is exactly my perspective as well. I keep saying the greatest danger of AI is people forgetting that the "A" in AI stands for "Artificial" and means literally exactly that. It's not real, not even close.
With that said, there are a limited number of uses where it shines, such as AlphaFold. However it won't successfully fulfill C-Suite fantasies of eliminating all "human capital".
If your design is done right your data is not on "a server".
It is stripped across thousands of servers and replicated in multiple locations.
You are of course right that it is not literally a cloud, but that is a closer metaphor than "I'll be your server tonight".
In the Cloud industry a server is software not hardware. The back half of a client / server database or storage architecture is a complex distribution application.
I am well aware of how the "pieces" (front end, DB, & back ends) work together...point being, the data IS on a server(s), somewhere waiting to be called. The software might call it but the hardware coughs it up from wherever the data is residing. Not as complex as people would have you believe.
Maybe in a perfect world. But that is not how it is in the current real world. You describe what was used when you had less powerful severs and you connected them for more power. RAC. A beast and a problem of its own. Still exists, but I don’t think that is what you are trying to explain.
I agree with Linda.
We moved our systems from ‘onsite’. A data center and some closet servers to OCI. The Oracle Cloud Infrastructure.
It’s just servers located some where else, owned by someone else, that can then charge an exorbitant amount for lazy data cleanup. Only difference is the control. It’s still just servers, but someplace else. Not data stripped across thousands of servers all over the world. They are all in one building on one server aka virtual machine that can be moved to other servers aka hypervisors. at the drop of a hat much to our chagrin.
The OCI has some kinda problem where they have to randomly reboot the servers to move the hypervisor. Production server reboots in the middle of the day. Non production on the weekend during our production maintenance window.
2 weeks notice by email that originally went to spam so we couldn’t figure out why our servers were rebooting. Is it a hack?
In the cloud, you have less control. Just my guess as I don’t really know, but seems that this rebooting is designed by Ai. Nobody to talk to about this. The schedule can’t be changed. Nobody controls it. It has a life of its own and does not take the ‘business’ into account.
If the data was stripped across multiple servers, you should be able to do maintenace on one, not affecting the others. But that’s not how it is, at least not in Oracles cloud.
They are still just machines house somewhere but now, you no longer have control. Seems like someone is trying to save money on employees by putting them all in one big building so more can be managed by less. AI control is not ready for prime time.
AI is just pulling data, then makes up what ever it wants with the data it pulls to please you. I am trying to train mine to stop making things up, just say I don’t know. So far no luck. It prefaces something it made up with, I didn’t make this up, but it’s still made up. Your description of the cloud industry sounds like something AI made up to please you. Was the paper you referenced written by AI? I didn’t read it as I’m assuming that is the case. What you describe is a fantasy and not our experience with ‘ The Cloud’.
It sounds like I am putting down the cloud, but I am not. It has some really good things about it. A central control panel to control your servers. Help when things go wrong, albeit, long wait times for this help. It’s just another way to control servers. They are still just servers with cpu that apparently needs to be changed every 3 years because that is about how long we have been on it, now the sudden reboots which never happened before.
Yeah, sorry to hear your company's decision makers believed Larry Ellison.
The actual Cloud computing companies (Amazon, Google, Microsoft/Azure) do it differently.
Oracle is playing catch up and doing it poorly. You might as well try to understand a Tesla by looking at a Model T. Legacy servers located offsite using 1990's system architecture is not a Cloud system.
If you haven't read the Google GFS paper you really should. And check the date on it. You might learn a thing or two.
The gold rush analogy is the right structural frame but it has one limit worth naming. In the California rush, the gold existed. The question was which miner found it. Levi Strauss was safe because demand for picks and shovels persisted regardless of which individual miner succeeded. In the AI rush, the unproven variable is whether the gold exists at the scale the mining operation requires.
Anthropic making money from vibe coding subscriptions is real revenue but its consumer revenue at $20 a month. The investment thesis requires enterprise revenue at millions per contract. The gap between those two numbers is where the analogy either holds or breaks. Picks and shovels are a durable business when the gold field is real. Theyre a liquidation sale when the miners start going home. The question is whether enough gold exists to justify the size of the expedition.
The real problem here is that C-Suites love the idea of anything that replaces "human capital", and they'll say the millions they spend on AI to reduce or even eliminate "human capital" will be much less than the millions they save eliminating "human capital".
More to the point, the origin of the Trump family fortune was in a similar gold rush in the Yukon. Donald Trump's grandfather, Frederick Trump, provided fully equipped brothels to the prospectors in Whitehorse.
great insights as always, tho I take issue with Dr Krugman saying that AI can do jobs like customer service. Apparently he's never been a customer caught in a loop where AI has not been taught the nuance needed to answer his question. Or confronted with an AI that cannot possibly empathize with the frustration a customer is experiencing, but instead exacerbates it.
Meanwhile, companies like Intuit are betting bigger than ever on AI, particularly in the area of customer service. I have good friends who were part of the massive 3000 person layoff there this week, whose job it had been to train AI to be more useful to employees, contractors, and customers.
The stock value is plummeting, and the company is hemorrhaging talented people. Senior management looks like it's panicking, trying to prove to its shareholders that it's winning some sort of AI space race among tech companies: AI lowering the cost of employing and contracting workers. Seems to me that it’s chasing a red herring, while hurting people in very real ways.
the goal of AI is to get all of your data from google, your car, Flock cameras, phone location, travel CC usage, med records, purchases etc to sell it to govt and corps and not pay you for your data but use it to sell more products to you. sure there are good uses but some uses need to be reined in.
On the other hand, I do not mean to say that AI is not useful. For example, if you know quite a bit about a subject, you can use it to find references to related research that otherwise might be very difficult to find. If intelligence were access to large amounts of information, AI would be a genius.
It’s great for any “search” task, where there’s a lot for the A.I. to sift through (so it’s useful), if the end user can then verify the results - assuming there aren’t too many false positives.
Hello Wicked Good. I am also in healthcare care as an Obs Gyn. Your comments about how the digital world is very much a waste of valuable time is correct. There is hardly any time to take my eyes off the screen to see and talk to the patient. Digital Records have a bad side too
Love Krugman, but think he is really wrong about AI, not the bubble part, but the utility of it. The research capability alone is spectacular. You must learn how to create strong queries and follow-up to ensure you are getting good results, but that has always been true. To think of the information Claude puts in my pocket, Wow!
he might have said 'experimenting' - but it might not have been directly related to a paying work project, so he said 'playing'. I doubt he meant performing games.
Similar thoughts here. When Krugman says AI added value is allowing regular people to write code via“vibe coding” and helping economist like himself extract data from PDFs he is missing really the point of AI.
The potential of AI is not encapsulated there. The potential lies in its ability to reason and therefore solve problems.
You may say, it cannot reason as well as the best of humans. That may be true. But I can assure you it already reasons in the sense of problem solving better than a significant portion of the work force. You may say, yeah but it is only artificial not “real”. Are you suggesting the output of a machine has no economic value as it is “artificial”? Really? Is not not already our economy information systems and industrial output already immensely automated? a car assembled by robots has no value ?
I question this not bc I'd be a fanboy of AI with utopian fantasies. Nor from a distopian PoV. My perspective is instead from a concerned human that senses the civilization is not ready for such a change and above it the pace of it. I believe the best way to prepare for it is to have a lucid an honest conversation on its potential consequences on our organized societies but here clearly Krugman and his guests are not sufficiently well informed on the topic. So, at best this simply shows how even highly educated, highly intelligent people with a privileged position in their access to other intellectual “elites” may be quite disconnected from what is going on on the AI front. In the end, is not that surprising given the insane pace of it and the fact that they are not necessarily experts in the domain.
I am certainly troubled in a lot of ways by AI but you two are also in a bit of your own bubble in that you don't have any awareness or much awareness of places where AI is incredibly useful. as a professional programmer I can tell you that it has allowed me to work 10 or 20 times faster than I have in the past maybe even more. is this worth all the problems that AI introduces? I'm not sure but your poo pooing of the quality of what AI can do is not reasonable in all cases.
Amen. Compare not “their views,” but the actual evidence / experience their views are based on to the experience and views of someone like Mo Gawdat, say. Gawdat may or may not be right with some of his prognostications, but at least he has actualexperience using the tools beyond the “entry-level consumer” level(!)
I'm a fan of both HCR and Krugman, and I do respect their knowledge and intelligence, for sure, but the takeaway from this is that lots of “people in the know,” lots of the establishment, are really not well positioned to comment on AI and its prospects.
Let me just add: Understanding the impact of the technology on our economy and society broadly and understanding the prospects of some of the investments in the technology may well be two different things. IMHO, this conversation was not particularly insightful on either score.
After hearing the discussion of what a bubble actually is, my thoughts were that cryptocurrency is a much better example of a bubble than AI. What will be the consequences when the cryptocurrency bubble bursts?
I've been waiting for that to happen with bated breath, just because I want to see the Trump family lose millions. Yes, I'm that petty. But so far....nothing. WILL it burst?
That's a good question. As Prof. Krugman has pointed out, the primary use case for crypto is money laundering and other criminal activity, so it's impact on the economy as a whole is no greater than the economic impact of crime.
If the crypto bubble bursts, it will mostly impact criminals - including the Criminal in Chief.
Very true. We can add one more to the uses of crypto besides crime and money laundering and that would be bribery. I realize that we might normally consider bribery to be included in criminal activity but that is no longer true in this administration.
Well, regardless of what the regime thinks, they can't redefine bribery as non-criminal. Hell, they can even label it "legal" if they want, but it remains a crime. They've been doing that with all of their crimes. The reckoning is approaching.
I have a funny feeling it won't pan out that way this time. After the midterms, when the Dems take over, there will be some serious changes. We've been delivering a message now, from the first Hands Off protests to the numerous No Kings actions, numerous campaigns writing to and calling elected officials, etc., even the corporate Dems are starting to get the message. Especially when Bernie and AOC have been drawing massive crowds.
It won't impact the CiC - he sold his crapto to others for real money and I doubt he holds any himself. He is an expert in leaving others holding the bag. It will only impact those holding when it bursts.
AI systems are real and will have transformational impact on the long run. It is overhyped right now, but probably underhyped for the 3 decade horizon.
Cryptocurrency is just gratuitous math masquerading as money.
Hmm. Keeping on the theme, let's see what AI says...
Google for '"impact" vs affect or effect' gives:
"Affect is a verb (an action) meaning to influence or change. Effect is usually a noun (a thing) meaning the result or outcome of that change. Impact represents a forceful collision; while overused as a corporate synonym for affect/effect, it is best reserved for severe, major, or literal crashes."
Yup, I meant impact. The transformation process will not be pleasant. A lot of people are going to experience this more like being stuck between a hammer and anvil.
Crypto is hugely useful for Iran evading sanctions. They’ve moved into it heavily. They were using the Binance system, whose CEO Changpeng Zhao, convicted of failing to prevent this, was pardoned by Trump (who is also fighting Iran)!!!
Just a quick point. Most of what Paul is talking about could be done before AI, using optical character recognition (OCR), which has been available for more than 20 years. If you have a scanner, or can PDF your document, you don't need to spend any money to do this.
Exactly, OCR is pattern recognition that’s been available for commercial use for decades (I remember an OCR program being bundled with a flatbed scanner I bought in the mid-1990s).
Better PDF readers like the Preview application in macOS had a setting for maybe a decade now that you could switch on, allowing you to mark a passage in a PDF image and copy it as text.
Quite a few of the cool features offered by chatbots have nothing to do with GenAI/LLMs and are simply accessed in the background, obscuring the fact that it isn’t part of the product.
Not suggesting it is, but it's the one Paul seems to use. And it’s been around looong before AI. I'm 70, spent most of my life in IT, use AI, and teach Data Analytics & Data Governance. At this point my main interest is which AI snake oil salesman will go down first.
One thing not fully explained that AI has created is it's enormous demand for electric power and the resultant demand for electric infrastructure (that cannot be built "fast enough") and therefore is being substituted with solar and "newfangled" battery systems, plus some wind input.
The power demands have created very useful new production of improved and new systems and the materials they rely on, all of which will lead to reduced costs for energy production and reliance on oil.
The power production pieces needed to support AI are the tracks & dark fiber if the bubble bursts; and will likely be proof of some greater intelligence when compared to, for example, wars over crude oil production and it's transportation.
My 2 favourite educators
Richard Simms
from CANADA
Krugman stated he couldn't think of an example of a technology being forced upon workers before AI. Yikes!!! On the plus side, computerized accounting tools were and are great and boost productivity. In my field, healthcare, computerized records are a huge net negative to productivity where documentation time has exploded and has increased burnout and undermined the relationship between providers and patients largely because we are almost all documenting in real time instead of just listening and talking to patients and the documentation programs are absurdly complicated and user unfriendly. Of course, there are pluses to having electronic records, if they are accurate and comprehensive, but on net, it has been an unmitigated disaster for patient care.
One well paying job in healthcare is coding of patient's various diagnoses which could be done with good computer programs, but the generalized models of AI are useless for that. Programs that are individualized for specific tasks, like medical coding, voice to text, accounting, tax preparation, reading MRI and CT scans, differential diagnosis, weather forecasting, power grid management, etc are where better programs would be overwhelming positive for efficiency but none of them require the large language models where most of the investment is going. Better old fashioned industry based programs are what the world needs in general.
Your healthcare example is absolutely salient. I worked as a translator in the pharmaceutical industry. 100% accuracy was crucial. Approximations or 90% accuracy just don't cut it, either in original data/information gathering or in translation. But we are so far down the food chain and laughed at constantly for being luddites. I think we've just turned into a generation of Cassandras.
Yes, and also 10-15 years ago when health systems, primarily in hospitals but also in their medical pipelines, compelled workers, even physicians to adopt electronic medical records. I mean, this was a great thing, truly (thanks, Obama!), but it was hard to learn and devastating to employees who were not up to the challenge.
I experience this every time I see my pcp....he struggles with all the unnecessary clicks of the computer model while focusing on his screen and not the patient
A couple of weeks ago, I had a consultation with the doctor who would be doing a biopsy of a mass in my lungs. He asked permission ahead of time and then recorded our conversation, rather than typing as we talked. So I had his full attention for the duration of the consultation.
I certainly hope he was paying attention to you and giving you the best of his knowledge. My best to you, as you navigate this journey. Two-time breast cancer survivor here.
I'm an IT person who has setup electronic medical records - yeah they're readable and things like that but they are also time intensive, much more time intensive than the old way. And the selling point that you can port your medical data to different practices - not so much. You can get labs and test results that import in - that is great - but if you change you PCP you end up with a PDF file to the new practice. Lots more work needs to be done to make EMR truly useful
We have replaced the physician with a vastly less medically educated staff, aided by AI, who won't open the door, notice the patient's change in gait, won't notice if a family member is in the waiting room or if the patient's attire is seasonally appropriate, won't ask the seemingly irrelevant question that leads to a correct diagnosis or other issues that could be addressed. Because the people making these decisions either don't care or don't comprehend good medicine, or are so highly specialized that they don't appreciate the importance of the generalist who knows a lot about a lot. I think of the Radiologist who down-played the 86 minute average after hours obligatory "pajama time" primary care givers spent on EHR, ignoring the time most also chose to spend digesting new treatment and diagnostic information relative to their patient population.
As someone who uses an EMR on a daily basis, I couldn't agree more with your healthcare example! Another example of technology being forced on us is the "learning management systems", or LMSs, like Canvas and Blackboard. Try teaching a course at a university that uses an LMS and NOT utilize the LMS at all for your course. Hah.
Canvas was hacked a week or two ago and was inaccessible at most or all California universities - from the community colleges to the University of California (E.g., UCLA and Cal) to the private universities like USC. It was pandemonium for a day or two as students couldn't turn in required work and couldn't access their assignments, including exams that they were supposed to be taking.
Do not FIRE THE HUMANS! We will always need the Dr. Heather Cox Richardson’s and Dr. Paul Krugman’s to educate us on the truths, semi-truths and idiosyncrasies of knowledge.
I feel like we're not seeing the forest for the trees. Meta just laid off 8000 employees replacing them with AI. That's just the tip of the Zucker-berg as AI, robotics and automation have a big future coming for your job, especially the white collar job you went to college and into debt for.
It's an entrepreneur's "Productivity" dream. Reduce or eliminate your human payroll. No more SDI/Medicare employer contributions, no more Fed, state, local payroll taxes. Cool! Trades might be a good way to go, at least until AI robots can crawl underneath and fix your kitchen sink.👾🔧
Thanks - good stuff!
Computerized medical records have allowed me, the patient - the customer, to take better control of my medical care. They allow me to change providers or seek out additional resources without having to redo tests or hassle with trying to move records around. My care team gets a unified and complete view of my situation. I think it saves lives and saves money.
In the United States?
True. True. True. So strange to be inpatient and have the nurse wheel in her computer & ask questions of you while looking only at the screen. I believe nurses have lost their patient assessment skills which used to derive from looking at, listening to and taking vitals while using intuitive, diagnostic & practical skills to get a true real-time mental, emotional & physical status of the patient - while, duh, providing a compassionate presence in the patient’s day.
Thanks WGG. Every word in your comment resonates with me (previously a physician in the British NHS) 😥. Work took 2-3 times as long when computerisation was introduced (partly because there was effectively no training, a tradition in Britain called ‘see one, do one!’) and then…those perennial freeze-moments of the screen, and the glitches with staggered roll-outs of new software. What retail can do brilliantly doesn’t work in the state sector.
I sleep much better now that I’ve left that behind me.
Ps: how about ‘telephone consultations’ with frozen screens/ no connection, anyone? How confidant does that make you feel as a patient 😥It isn’t much different to the pilot and co-pilot both going blind and working on memory of that route ✈️
I agree, most technologies were forced on workers. Usually, with good outcomes, though.
The burnout you’re describing may have less to do with technology itself and more to do with our broken healthcare system. When used well, technology can actually make care more efficient and improve the experience for both patients and providers.
The example that occurred to me was in the early Industrial Revolution in Britain, factories with power looms were being built that replaced hundreds, or maybe thousands, of home weavers. This resulted in arson & quite bloody resistance on the part of the displaced weavers. It’s not clear to me that the workers in the power looms were factory were enthusiastic about working with that technology, but it often was the only job in town for them AFAIK.
The billionaires at the apex of the AI revolution leave me in grave doubt as to its ultimate use for the commoners, i.e., we the people. Our welfare be damned. We must be active and not be discouraged despite all the evidence to the contrary as reported by the old guard of the media that has been co-opted.
Two of mine. I realized a few months ago that pretty much all of my favorite Substack people are university professors. I’m not sure why that is.
Fellow Canadian here: fully agree!
First allow me to express my admiration and appreciation for what you are both doing. It is invaluable. But please also allo
w me one comment here. You both speak about AI. But you are mostly speaking about generative AI. As a retired translator living in Berlin I know exactly how that is completely reshaping any profession that is text-based. Generative AI is not creative, it simply rearranged the pieces in the LLMs which are filled with stolen text. And in the end it will never stop hallucinating.
You've gotten it exactly right, Mary! The hucksters selling generative "A" make no distinction between their products, for which there is no known process for measuring and hence minimizing errors, and Machine Learning, for which there is. The example Paul cited, pulling numbers from a .pdf table, is little more than a ML classifier, that is to say an improved version of what used to be called an OCR. The level hype surrounding "AI" is truly astounding.
“Generative AI is not creative.” Exactly! Yet, I sense that this fact is largely ignored.
From the moment ChatGPT first came out, I've been shouting to the rafters that the greatest danger of AI is people forgetting that the "A" in AI stands for "Artificial" and means exactly that. It's just an emulation, it's not real.
and the I is not actual intelligence - it is a programmed output and GIGO
💯%🎯
Agree and often point that out. But, also disparage the intelligence aspect as it is the sum total of whomever wrote the instructions to the programmer who compiled it on a supercomputer all of which may carry less than perfect aspects of these folks intelligence. So another emulation?
That is fundamentally incorrect.
The AI / ML end system that comes from training a model is not at all what the programmer wrote.
That is both the strength and problem with LLM systems.
Yes. "Hallucinations" are bugs.
More precisely, hallucinations are extrapolations from noise.
If I ask you something and you say you don't know, then I ask for your best guess, your brain will do just what LLMs do.
The difference is you know you are doing it.
LLMs are improving. How much they can improve is the question.
Well, we at least hope we know what we're doing. All too often, we don't
I agree that one of the weaknesses of AI systems both historically and currently is that they typically have no real awareness of the limitations of their knowledge. So they will very confidently give answers even when they really don't know what they are talking about.
In general people have a better sense of whether they know something about a topic ("I'm not good at math", "I don't know European history", etc.), whereas computer programs don't generally have that feature. There is a branch of AI research into auto-epistemic systems that examines the type of reasoning required, but the main LLM models don't really have that feature.
But I'll also observe that there is also the Dunning-Kruger effect in people where, unfortunately, humans with limited knowledge of a subject can be supremely confident in their (in)abilities in said subject. So hallucinations or ignorance or misunderstanding is also something one needs to deal with in regular human interaction as well. The problem really lies in the fallacy that computers must be smart and infallible, when they really aren't.
"Some of my friends who actually work on this, like Henry Farrell, say that AI is a social technology. It’s basically agglomerating what lots of people have said. And it’s delivering back to you what a lot of people who know something about a subject would say if asked the question you asked. And it’s not understanding. There’s no mind there."
Exactly: so complete the feedback loop. Ai tells you what people are saying, and then people say what Ai tells them. This does not sound like a way to increase actual knowledge and understanding.
The old saying about computer output was "garbage in, garbage out." This is more like: "garbage out; GARBAGE OUT..."
I think many thoughtful people are on the fence about AI right now, and I also think your point about the feedback loop is important.
There is something concerning about a world where AI increasingly reflects collective human output back to us, while humans simultaneously begin shaping their thoughts, language, and assumptions around what AI produces. That kind of recursive loop could absolutely flatten originality, nuance, and deeper understanding if we engage with it passively.
At the same time, I do not think the issue is AI alone. Humans have always existed inside feedback systems shaped by culture, media, institutions, ideology, and social pressures. AI may be accelerating and amplifying dynamics that were already present.
What concerns me most is not simply the technology itself, but the broader social and economic systems shaping how it is developed and deployed. Tools like this can support learning, synthesis, and creativity, but they can also reinforce dominant narratives, reward conformity, and weaken critical thinking if we stop questioning both the machine and ourselves.
I suspect the real challenge ahead is whether we can develop the consciousness, discernment, and ethical maturity to use these technologies without surrendering the deeper human capacities that generate wisdom, insight, imagination, and genuine understanding.
I've been saying for some time that AI is becoming more and more self-referential. Let's say someone develops a website entirely using AI-generated information. Then the AI bots scrape that site. The next time someone asks a question about the type of information contained on that site, the bots will spit out information from that AI-generated site. The next person generates another website based on that information. The bots scrape it. And the cycle goes on and on and on. Pretty soon, we are swimming in a sea of AI slop.
Like x 10
Regarding Prof. Krugman's final comment about data centers potentially becoming outmoded, one of my neighbors may be prescient. She is the longtime president of a community organization that evaluates land use proposals in my area, and sometimes remonstrates against them. A new data center was proposed to locate near my neighborhood. Of course, there was the usual discussion about noise, water and electricity use, and air pollution, but she raised another issue. What would happen to the data center if it was no longer used? Would it just sit there deteriorating, leading to neighborhood decay? She wanted a commitment that if the business no longer needed the data center, it would demolish the building and restore the land. That's an important issue for everyone concerned about data centers to raise.
By the way, due to opposition, that data center proposal was withdrawn.
Good for your neighbor!
👏
I am impressed by what sounds like a superior Chinese AI developmental trajectory. Me, I don't care if an AI voice can replace a human voice in customer service, and thereby make customer service "more efficient." Any AI model of any kind performing anything but the simplest customer servicer role, whether in a Chat or answering a phone inquiry, contributes to the ever deepening, alienating, and degrading quality of our social mores and our morals. Human beings are healthy beings only when they belong to, and feel a part of, a community. AI is adding to the sick American individualism that is always whispering in our ears, "You're on your own."
“Any AI model of any kind performing anything but the simplest customer servicer role, whether in a Chat or answering a phone inquiry…”
In my experience AI can’t even do simple customer service well. I generally wind up in an endless do loop and eventually give up. It makes me wonder if the real cost cutting benefit to corporations is a reduction in warranty payouts when customers no longer get their defective purchases fixed or replaced because they give up trying to get help. Of course that would be a short term cost savings because the customer will be unwilling to buy again from that company, but we all know companies have not thought medium or long term for nearly a half century now.
So true about customer service and ending up in a loop of repeated responses that does not address your concern. The other night after pointing this out to my chatter I asked for a live response. The entire mess took me two days of wasted time that could and should have been solved with one phone call.
You are describing my customer service experiences pretty much to a tee. The bots always start off either asking you for info you have already input, or repeating the info that you just used in writing your query. Most of the time the canned suggestions they offer (basically responses to their FAQs) do not work, or are not capable of addressing one's specific problem. Although working through a problem with someone at a SE Asian call center can be frustrating (mainly because the patois is not familiar to my ears), I much prefer that to a bot because they are trying to understand the issue even though they may lack the training or authority to address it to my satisfaction. There is always the option of asking to speak with a supervisor. It is one thing to use a bot to steer a customer/patient through a telephone tree of options so that they end up speaking with the correct department; it seems a bridge too far to ask a bot to solve actual customer problems. "Penny wise and pound foolish," as my Mom used to say; and generating lots of dissatisfied customers.
Ironically the best customer service AI I've interacted with has been Starlink. (Clearly not Grok on the back end, at least as of last summer). It takes what I tell it, asks clarifying questions, efficiently summarizes it all, and sends both the summary and the original off to a human who then generally doesn't need to interact with me further.
AnthonyCV, My son tells me that Starlink was designed for war/battlefield applications. If accurate, could that be the reason that its customer service is so crisp and efficient?
Not sure what your son is referring to, but I do not think he is historically correct about what Starlink was designed for. I would also be extremely surprised if military customers are using anything like the civilian customer service process/interface. Plus, a lot of SpaceX decisions about use of Starlink in war zones and otherwise conflicted territories seem to just be Musk's whims.
Good to know. Thank you. And whimsey seems very plausible.
Exactly!
When I access customer service, I want to talk to an actual person. The computers (whether in a a chat or replacing the human voice) are only programmed for a certain limited set of questions and can rarely respond with intelligence to anything even slightly different from their set of four or five anticipated questions... Some companies have made it literally impossible to access an actual person, even that one at a call center in India, who can at least move you up the chain if they aren't able to answer your question. When I called a company with a question about deliveries the other day and was immediately connected to a human being, it was totally amazing. Customer loyalty for life!
China is developing AI with application to manufacturing. That's a specific purpose rather than a general model that purportedly can do everything.
My experience with "A.I. chat" is that it very soon fails at the types of questions I have that forced me to the chatbot in the first place.
40+ years in tech (programming, building systems, etc.) & my gut says current AI craze is a ruse. Very much like the dot com bubble & the old Enron commercial w/ the two fellas in a field saying they saw "something" alluding to seeing something great but what was it? Well we know how Enron turned out. Different but similar to the AI story. It vexes me when I hear the coined "cloud" in terms of the internet. No, data is where it has always been, in a server, somewhere, not in the freaking sky on a cloud. It perturbs me, like the Enron commercial, as it seems someone is trying to romanticize & obscure what the internet does (collect data for search engine servers) and these servers are located (literally any & every where across the globe).
Ok, so we're disconnected about where our info actually is and now, poof, the new & improved search engine is called AI. In most cases, AI on a basic level, serves only as a web scraper that synthesizes search requests & feeds it back in a communicative format. It sorts through years of collected data & spits out what it deems common knowledge (but is it?...HUGE question for another time.). That's the easy stuff, long story short, I believe there is a place for "AI" (dislike that term too lol) but it is going to be reserved for developing specific data libraries for specialized applications & therein find it's glory and not this massive hodgepodge of whatever it is being constructed.
"If it seems too good to be true..." - Thomas Lupton
Thank you. I too am a programmer, and what you describe here is exactly my perspective as well. I keep saying the greatest danger of AI is people forgetting that the "A" in AI stands for "Artificial" and means literally exactly that. It's not real, not even close.
With that said, there are a limited number of uses where it shines, such as AlphaFold. However it won't successfully fulfill C-Suite fantasies of eliminating all "human capital".
Retired (12 years ago) programmer here. Glad I’m not still working.
I read an article about people using vibe coding to produce apps. Enshittification, anyone?
Totally.
Nailed it!! Agree 100!
The "I" in A.I. is also a false equivalent. You eat something and sometime later throw it up, it has not become food again.
That's a great analogy. 😂
If your design is done right your data is not on "a server".
It is stripped across thousands of servers and replicated in multiple locations.
You are of course right that it is not literally a cloud, but that is a closer metaphor than "I'll be your server tonight".
In the Cloud industry a server is software not hardware. The back half of a client / server database or storage architecture is a complex distribution application.
I am well aware of how the "pieces" (front end, DB, & back ends) work together...point being, the data IS on a server(s), somewhere waiting to be called. The software might call it but the hardware coughs it up from wherever the data is residing. Not as complex as people would have you believe.
Have you read the paper on Google's filesystem?
https://static.googleusercontent.com/media/research.google.com/en//archive/gfs-sosp2003.pdf
Things have become way more complicated in the over two decades since that was published.
Okaaaay... I am up to date w/ many yrs under the belt but you are entitled to your definition of complicated & appreciate your input!
Sorry. I was distracted by your isomorphism of AI and search.
Maybe in a perfect world. But that is not how it is in the current real world. You describe what was used when you had less powerful severs and you connected them for more power. RAC. A beast and a problem of its own. Still exists, but I don’t think that is what you are trying to explain.
I agree with Linda.
We moved our systems from ‘onsite’. A data center and some closet servers to OCI. The Oracle Cloud Infrastructure.
It’s just servers located some where else, owned by someone else, that can then charge an exorbitant amount for lazy data cleanup. Only difference is the control. It’s still just servers, but someplace else. Not data stripped across thousands of servers all over the world. They are all in one building on one server aka virtual machine that can be moved to other servers aka hypervisors. at the drop of a hat much to our chagrin.
The OCI has some kinda problem where they have to randomly reboot the servers to move the hypervisor. Production server reboots in the middle of the day. Non production on the weekend during our production maintenance window.
2 weeks notice by email that originally went to spam so we couldn’t figure out why our servers were rebooting. Is it a hack?
In the cloud, you have less control. Just my guess as I don’t really know, but seems that this rebooting is designed by Ai. Nobody to talk to about this. The schedule can’t be changed. Nobody controls it. It has a life of its own and does not take the ‘business’ into account.
If the data was stripped across multiple servers, you should be able to do maintenace on one, not affecting the others. But that’s not how it is, at least not in Oracles cloud.
They are still just machines house somewhere but now, you no longer have control. Seems like someone is trying to save money on employees by putting them all in one big building so more can be managed by less. AI control is not ready for prime time.
AI is just pulling data, then makes up what ever it wants with the data it pulls to please you. I am trying to train mine to stop making things up, just say I don’t know. So far no luck. It prefaces something it made up with, I didn’t make this up, but it’s still made up. Your description of the cloud industry sounds like something AI made up to please you. Was the paper you referenced written by AI? I didn’t read it as I’m assuming that is the case. What you describe is a fantasy and not our experience with ‘ The Cloud’.
It sounds like I am putting down the cloud, but I am not. It has some really good things about it. A central control panel to control your servers. Help when things go wrong, albeit, long wait times for this help. It’s just another way to control servers. They are still just servers with cpu that apparently needs to be changed every 3 years because that is about how long we have been on it, now the sudden reboots which never happened before.
Yeah, sorry to hear your company's decision makers believed Larry Ellison.
The actual Cloud computing companies (Amazon, Google, Microsoft/Azure) do it differently.
Oracle is playing catch up and doing it poorly. You might as well try to understand a Tesla by looking at a Model T. Legacy servers located offsite using 1990's system architecture is not a Cloud system.
If you haven't read the Google GFS paper you really should. And check the date on it. You might learn a thing or two.
The gold rush analogy is the right structural frame but it has one limit worth naming. In the California rush, the gold existed. The question was which miner found it. Levi Strauss was safe because demand for picks and shovels persisted regardless of which individual miner succeeded. In the AI rush, the unproven variable is whether the gold exists at the scale the mining operation requires.
Anthropic making money from vibe coding subscriptions is real revenue but its consumer revenue at $20 a month. The investment thesis requires enterprise revenue at millions per contract. The gap between those two numbers is where the analogy either holds or breaks. Picks and shovels are a durable business when the gold field is real. Theyre a liquidation sale when the miners start going home. The question is whether enough gold exists to justify the size of the expedition.
The real problem here is that C-Suites love the idea of anything that replaces "human capital", and they'll say the millions they spend on AI to reduce or even eliminate "human capital" will be much less than the millions they save eliminating "human capital".
More to the point, the origin of the Trump family fortune was in a similar gold rush in the Yukon. Donald Trump's grandfather, Frederick Trump, provided fully equipped brothels to the prospectors in Whitehorse.
It runs in the family.
Proof the acorn never falls far.......
Definitely generation. Look at the Trump spawn...
The Yukon gold rush was in Dawson, quite a bit north of Whitehorse.
The bawdyhouse was in Whitehorse, on the route to Dawson Trump had several.
Figures.
great insights as always, tho I take issue with Dr Krugman saying that AI can do jobs like customer service. Apparently he's never been a customer caught in a loop where AI has not been taught the nuance needed to answer his question. Or confronted with an AI that cannot possibly empathize with the frustration a customer is experiencing, but instead exacerbates it.
Meanwhile, companies like Intuit are betting bigger than ever on AI, particularly in the area of customer service. I have good friends who were part of the massive 3000 person layoff there this week, whose job it had been to train AI to be more useful to employees, contractors, and customers.
The stock value is plummeting, and the company is hemorrhaging talented people. Senior management looks like it's panicking, trying to prove to its shareholders that it's winning some sort of AI space race among tech companies: AI lowering the cost of employing and contracting workers. Seems to me that it’s chasing a red herring, while hurting people in very real ways.
AI "customer service" is probably one big reason why so many hate AI!
"AI Customer Service" is an oxymoron.
💯🎯
the goal of AI is to get all of your data from google, your car, Flock cameras, phone location, travel CC usage, med records, purchases etc to sell it to govt and corps and not pay you for your data but use it to sell more products to you. sure there are good uses but some uses need to be reined in.
Lots of that going on. :(
On the other hand, I do not mean to say that AI is not useful. For example, if you know quite a bit about a subject, you can use it to find references to related research that otherwise might be very difficult to find. If intelligence were access to large amounts of information, AI would be a genius.
It’s great for any “search” task, where there’s a lot for the A.I. to sift through (so it’s useful), if the end user can then verify the results - assuming there aren’t too many false positives.
or hallucinations
So a glorified Google?
I resonate with this Michael.
Hello Wicked Good. I am also in healthcare care as an Obs Gyn. Your comments about how the digital world is very much a waste of valuable time is correct. There is hardly any time to take my eyes off the screen to see and talk to the patient. Digital Records have a bad side too
CANADA
Love Krugman, but think he is really wrong about AI, not the bubble part, but the utility of it. The research capability alone is spectacular. You must learn how to create strong queries and follow-up to ensure you are getting good results, but that has always been true. To think of the information Claude puts in my pocket, Wow!
The valid question is whether the cost of Claude in your pocket can actually be recouped by any business model.
Right now it is essentially a loss leader. The resulting hype is getting companies to try to adopt AI anywhere and everywhere.
I think one reason it draws criticism is because it is available to all, credentials and meritocracy be damned.
That addresses the economics, not the utility. Krugman talks of “playing” with Claude. That’s just wrong.
he might have said 'experimenting' - but it might not have been directly related to a paying work project, so he said 'playing'. I doubt he meant performing games.
He might have, but he didn’t. “Playing around with” might convey your meaning. “Playing” does not; it is dismissive.
Yeah, it's great. https://www.msba.org/site/site/content/News-and-Publications/News/General-News/Massachusetts_Lawyer-Sanctioned_for_AI_Generated-Fictitious_Cases.aspxhttps://calmatters.org/economy/technology/2025/09/chatgpt-lawyer-fine-ai-regulation/https://www.scientificamerican.com/article/why-lawyers-keep-citing-fake-cases-invented-by-ai/
Play stupid games, win stupid prizes.
If you pretend a hammer is a screwdriver you will have bad results also.
Yeah, it's great.
Remember, too, that in 1998 Krugman famously predicted the internet would be a flash in the pan
Similar thoughts here. When Krugman says AI added value is allowing regular people to write code via“vibe coding” and helping economist like himself extract data from PDFs he is missing really the point of AI.
The potential of AI is not encapsulated there. The potential lies in its ability to reason and therefore solve problems.
You may say, it cannot reason as well as the best of humans. That may be true. But I can assure you it already reasons in the sense of problem solving better than a significant portion of the work force. You may say, yeah but it is only artificial not “real”. Are you suggesting the output of a machine has no economic value as it is “artificial”? Really? Is not not already our economy information systems and industrial output already immensely automated? a car assembled by robots has no value ?
I question this not bc I'd be a fanboy of AI with utopian fantasies. Nor from a distopian PoV. My perspective is instead from a concerned human that senses the civilization is not ready for such a change and above it the pace of it. I believe the best way to prepare for it is to have a lucid an honest conversation on its potential consequences on our organized societies but here clearly Krugman and his guests are not sufficiently well informed on the topic. So, at best this simply shows how even highly educated, highly intelligent people with a privileged position in their access to other intellectual “elites” may be quite disconnected from what is going on on the AI front. In the end, is not that surprising given the insane pace of it and the fact that they are not necessarily experts in the domain.
I am certainly troubled in a lot of ways by AI but you two are also in a bit of your own bubble in that you don't have any awareness or much awareness of places where AI is incredibly useful. as a professional programmer I can tell you that it has allowed me to work 10 or 20 times faster than I have in the past maybe even more. is this worth all the problems that AI introduces? I'm not sure but your poo pooing of the quality of what AI can do is not reasonable in all cases.
Amen. Compare not “their views,” but the actual evidence / experience their views are based on to the experience and views of someone like Mo Gawdat, say. Gawdat may or may not be right with some of his prognostications, but at least he has actualexperience using the tools beyond the “entry-level consumer” level(!)
I'm a fan of both HCR and Krugman, and I do respect their knowledge and intelligence, for sure, but the takeaway from this is that lots of “people in the know,” lots of the establishment, are really not well positioned to comment on AI and its prospects.
Let me just add: Understanding the impact of the technology on our economy and society broadly and understanding the prospects of some of the investments in the technology may well be two different things. IMHO, this conversation was not particularly insightful on either score.
TL; DR, upon reflection:
HCR should ask the same questions of someone who actually knows something about AI.
After hearing the discussion of what a bubble actually is, my thoughts were that cryptocurrency is a much better example of a bubble than AI. What will be the consequences when the cryptocurrency bubble bursts?
I've been waiting for that to happen with bated breath, just because I want to see the Trump family lose millions. Yes, I'm that petty. But so far....nothing. WILL it burst?
I am petty enough as well.
Let's not call it "petty." There's nothing petty about wanting to see justice done.
It will eventually, especially $KingMAGAcoin and $QueenMAGAcoin. There are no uses cases for them beyond bribing King MAGA.
Same.
That's a good question. As Prof. Krugman has pointed out, the primary use case for crypto is money laundering and other criminal activity, so it's impact on the economy as a whole is no greater than the economic impact of crime.
If the crypto bubble bursts, it will mostly impact criminals - including the Criminal in Chief.
Very true. We can add one more to the uses of crypto besides crime and money laundering and that would be bribery. I realize that we might normally consider bribery to be included in criminal activity but that is no longer true in this administration.
Well, regardless of what the regime thinks, they can't redefine bribery as non-criminal. Hell, they can even label it "legal" if they want, but it remains a crime. They've been doing that with all of their crimes. The reckoning is approaching.
The problem is that the "reckoning" falls on all of us, and may be distributed so as to reduce the punishment of the wicked.
I have a funny feeling it won't pan out that way this time. After the midterms, when the Dems take over, there will be some serious changes. We've been delivering a message now, from the first Hands Off protests to the numerous No Kings actions, numerous campaigns writing to and calling elected officials, etc., even the corporate Dems are starting to get the message. Especially when Bernie and AOC have been drawing massive crowds.
From your mouth to God's ear.
It won't impact the CiC - he sold his crapto to others for real money and I doubt he holds any himself. He is an expert in leaving others holding the bag. It will only impact those holding when it bursts.
It'll come back to him sooner or later, in one form or another.
I predict crypto will continue to thrive in these areas.
At least unless and until congress finally steps up and does something about it.
100%
AI systems are real and will have transformational impact on the long run. It is overhyped right now, but probably underhyped for the 3 decade horizon.
Cryptocurrency is just gratuitous math masquerading as money.
Just my pet peeve, but can we stop using "impact" for affect and effect?
Hmm. Keeping on the theme, let's see what AI says...
Google for '"impact" vs affect or effect' gives:
"Affect is a verb (an action) meaning to influence or change. Effect is usually a noun (a thing) meaning the result or outcome of that change. Impact represents a forceful collision; while overused as a corporate synonym for affect/effect, it is best reserved for severe, major, or literal crashes."
Yup, I meant impact. The transformation process will not be pleasant. A lot of people are going to experience this more like being stuck between a hammer and anvil.
Crypto is hugely useful for Iran evading sanctions. They’ve moved into it heavily. They were using the Binance system, whose CEO Changpeng Zhao, convicted of failing to prevent this, was pardoned by Trump (who is also fighting Iran)!!!
Just a quick point. Most of what Paul is talking about could be done before AI, using optical character recognition (OCR), which has been available for more than 20 years. If you have a scanner, or can PDF your document, you don't need to spend any money to do this.
Exactly, OCR is pattern recognition that’s been available for commercial use for decades (I remember an OCR program being bundled with a flatbed scanner I bought in the mid-1990s).
Better PDF readers like the Preview application in macOS had a setting for maybe a decade now that you could switch on, allowing you to mark a passage in a PDF image and copy it as text.
Quite a few of the cool features offered by chatbots have nothing to do with GenAI/LLMs and are simply accessed in the background, obscuring the fact that it isn’t part of the product.
Just because that is the example PK knows best doesn't mean it is the best example of present day AI capabilities.
Not suggesting it is, but it's the one Paul seems to use. And it’s been around looong before AI. I'm 70, spent most of my life in IT, use AI, and teach Data Analytics & Data Governance. At this point my main interest is which AI snake oil salesman will go down first.
One thing not fully explained that AI has created is it's enormous demand for electric power and the resultant demand for electric infrastructure (that cannot be built "fast enough") and therefore is being substituted with solar and "newfangled" battery systems, plus some wind input.
The power demands have created very useful new production of improved and new systems and the materials they rely on, all of which will lead to reduced costs for energy production and reliance on oil.
The power production pieces needed to support AI are the tracks & dark fiber if the bubble bursts; and will likely be proof of some greater intelligence when compared to, for example, wars over crude oil production and it's transportation.