Tag Archives: artificial intelligence

Edmund Phelps vs. Breitbart

Leave a reply

Edmund Phelps, a Nobel prize-winning economist, is not alive to defend himself against this Breitbart article claiming to be based on his ideas:

This is where incoming Fed chairman Kevin Warsh’s argument about artificial intelligence and monetary policy fits naturally. Warsh has argued that AI represents a genuine supply-side productivity shock, not merely demand-pull inflation dressed up as growth. If that’s right, then accommodating those gains with lower interest rates isn’t a concession to inflation. It’s a recognition that the economy’s supply capacity is improving.

Under a Phelpsian framework, the policy goal should be to let productivity gains flow through the economy without forcing an unnecessary contraction. Rate cuts in this setting aren’t simply “loose money.” They can be the mechanism that prevents firms from overreacting to realized productivity gains by pulling back too sharply on hiring and investment.

This also explains why the tight labor market may be protective against recession rather than a precursor to it. In the standard story, a productivity acceleration can be dangerous because it follows a period of overexpansion. Firms hired in anticipation of the boom, and when productivity finally arrives, they discover they no longer need as much labor. Layoffs ensue, consumer confidence cracks, spending falls, and the risk of a recession becomes large.

This sounds kind of smart-ish, but I was skeptical because of the source. Phelps’s main ideas were about expectations of the future, true, but it seemed to me that this was likely to be a twisting of his ideas to meet a desired political end, rather than a serious analysis.

Since Phelps is not around to defend his ideas, I asked Gemini to assume his persona and write a response. Here is what she said:

It is flattering, I suppose, to see my theories invoked, but this argument entirely misapprehends the mechanics of both expectations and capital. If artificial intelligence is truly about to unleash a massive, structural leap in total factor productivity, the natural rate of interest must rise, not fall. When an economy anticipates a surge of genuine, grassroots dynamism—moving beyond the mechanical, exogenous growth of standard Solow models—firms aggressively seek capital today to capture those future gains. If a central bank artificially suppresses interest rates in the face of that surging demand, it does not foster innovation; it merely ignites asset bubbles and distorts the very market signals that tinkerers and entrepreneurs rely upon.

Furthermore, the author has forgotten the most crucial lesson of the expectations-augmented Phillips curve. You cannot placate the labor market with cheap money today by promising them a technological supply miracle tomorrow. If the Federal Reserve cuts rates in an already tight market, businesses and workers will instantly observe the monetary expansion and update their inflation expectations, driving up prices and wages immediately. The inflation takes root long before the first AI efficiency is ever deployed. True mass flourishing requires the discipline of real capital costs, not the illusion of a monetary free lunch.

Let’s imagine a case where we were actually observing increased productivity in real time because the AI technology is progressing and rolling out much faster that previous technologies like the “electrification, automobiles, radio, modern manufacturing” – which Breitbart mentions. In this case, we should also be seeing inflation expectations starting to shift toward deflation, and it would then make sense to lower interest rates and try to get more cash flowing through the system. But this can be done in real time in response to the signal. The argument that it can or should be done in advance would seem to be invalid. I don’t think Breitbart “overlooked” anything here, I just think this is an un-serious twisting of serious ideas, meant to fool people into supporting a preferred political agenda.

AI as pro-social media

Leave a reply

This is an idea I hadn’t thought of – because AIs are inherently rational and objective (more or less, most of the time?), they tend to nudge people towards rational evidence-based views (which tend to be centrist politically). Whereas social media tends to herd people toward extreme views, wild conspiracy theories, and political fringe-ism.

My provisional theory is that LLMs, as a consumer product, will push people’s senses of reality closer together in a sort of mirror image of the way social media has fractured them. They are not algorithms meant to custom-tailor content (including facts) to you, and what you will find infuriating or motivating. They are centralized systems that, until you prompt them or give them context, behave basically the same way for everyone. As the philosopher Dan Williams put it, “Whereas social media democratised information, LLMs technocratise it.”

A couple points he makes is that even though AIs can demonstrably be trained to favor one part of the political spectrum (see the analysis of Grok below posted by the blog Urbanomics), even then they are much more centrist than social media. So is it possible AI could nudge our entire society in a more rational, evidence-backed direction?

There is no absolute guarantee they are going to stay this way, of course. If I wanted to train an agent right now to go out and look at the day’s news headlines, and filter them for me in the style of Breitbart (which generally provides factually accurate information, but cherry-picked and presented with biased language), I could do that.

https://gulzar05.blogspot.com/2026/04/weekend-reading-links_12.html

David Chalmers, the “hard” and “soft” problems of consciousness, and could I be a zombie and not know it?

Leave a reply

This article is by a person who disagrees with David Chalmers, but it happens to have a good summary of what David Chalmers had to say.

The consciousness debate is often formulated in terms used in an influential talk given by a young David Chalmers in Tucson in 1994. Chalmers, a philosopher, distinguished two separate “problems of consciousness.” The first is the very hard problem described above: understanding the processes in the brain that give rise to the many aspects of our visible behavior and our inner behavior that we can report about. Chalmers christened this hard problem as the “easy” problem of consciousness. 

Then he declared that there is another distinct problem — why the brain’s behavior is accompanied by experience at all — which he christened the “hard” problem of consciousness. Today, this so-called “hard problem” is mentioned in all debates on consciousness. According to many, it unveils the very limits of current scientific understanding. Chalmers claimed that even after hypothetically accounting for our entire behavior, and for all our reports about our inner life, there would still be an “explanatory gap” between brain processes and experience…

Chalmers asks us to contemplate what he calls a “philosophical zombie.” This is a hypothetical entity that looks and behaves like a human in all respects, including reporting emotions, feelings, dreams and experience, yet it has no consciousness. As Chalmers puts it, “There is nobody home.” This is a rhetorical trick that induces us to distinguish between behavior and a hypothetical reality accessible only by introspection. The very fact that a philosophical zombie could be conceived, Chalmers argues, shows that inner experience is intrinsically distinct from observable natural phenomena. 

The article has a link to a video of Chalmers’s original lecture, for people who have time to watch videos. Such people must either not have jobs, houses or children, or else their jobs, houses and children must be less demanding of time and attention than mine, or maybe they are wealthy enough to pay other people to deal with their houses and children while they do their jobs (but this raises some questions of priorities, in my view.

April 2026 in Review

Leave a reply

In fast-moving current events as I write (Saturday, May 2), active so-called “kinetic” warfare seems to have subsided in and around Iran. Let’s hope the trend continues in this hopeful direction. Human rights violations elsewhere and global economic impacts persist.

Most frightening and/or depressing story: We have heard horror stories about U.S. government debt over the decades, many not grounded strictly in evidence. But this time really seems to be different, where the absolute size of the debt at the moment means higher than normal interest rate payments as a fraction of the economy and tax revenue. At the risk of stating the obvious, this means the government has less money for things other than interest payments. Meanwhile the trends are increasing debt level, increasing interest rates, and potentially lackluster economic and tax revenue growth, all pointing toward a runaway train. Hoping for a pickup in economic growth seems to be the main strategy being pursued to counteract this feedback loop.

Most hopeful story: AI science seems to have been a theme of mine in April. We can constraint an AI scientist to actually respect the laws of physics, potentially accelerating scientific and technological progress. AI should also be good at synthesizing past research to form a basis for future progress, and organizing data in an accessible way so that others (human and/or AI) can confirm findings or make new discoveries from that same data. I know some very nice people who work in today’s academic publishing industry, but this may not be an area of rapid future growth. The future of engineering and scientific modeling will probably consist of giving an AI a very detailed specification for what you want it to accomplish, then reviewing/validating the result when it comes back.

Most interesting story, that was not particularly frightening or hopeful, or perhaps was a mixture of both: Augmented (aka mixed) reality glasses are getting pretty common in China, and slowly catching on elsewhere. Early adopters include cheating students, of course.

Academic publishing beyond the current citation regime

Leave a reply

This blog post from someone named David Oks pines for a time before citations became a central pillar in the effort to advance scientific publishing. Thank you Mr. Oks for sharing your views, which got me thinking. I am not sure I agree 100%, because an important part of civilization and science is our ability to document the state of knowledge so that each successive generation can build on it. But the point is taken that the extreme focus on gaming this one metric (number of citations, citation scores of individual academics and journals) has become an end in itself, rather than a means for advancing civilization and science.

I do think AI can be very good at improving the state of the “literature review”. Every scientific article starts with a summary of literature on the topic, which the authors then typically build on (although, some articles are just literature review). A human author can spend years sifting through a vast amount of literature relevant to a topic, particularly a novel or interdisciplinary topic, trying to find those few needles in a haystack that are really the most, say, 100 highly relevant papers, and synthesize them into a foundation that can be built on it. I have done this, and it is actually a very fun thing to do (for my personality type, I suppose, not for everyone, but there are many like me…) Through this process, you gradually build and refine your own unique mental model on a subject which then can become the foundation for your personal unique contribution to human progress. However, even spending years, you can’t come close to looking at everything possibly relevant, and you can miss some of those needles. An AI should be able to look at literally the entire haystack, find and synthesize the needles, in a matter of minutes or at most hours, compared to months or years for a person. The only sad thing here is that the best mental models in the brains of the smartest humans might not be built in the same way. Overall though, I suspect the rate of progress can be increased significantly.

What is an alternative to the citation regime according to Mr. Oks?

So I suspect that we’ll have to fundamentally rethink the institutions of scientific life for the age of strong AI. Perhaps, as AI makes it possible to do much more science much more quickly, the culture of science will become more like the culture of engineering—faster, more collaborative, less interested in priority claims. In such a world, the most efficient unit of scientific contribution might be a living document, perhaps even just a GitHub repo: something with data, code, analysis, and a thin narrative layer that AI scientists could read, regenerate, or update as needed. And citations, in this world, could ultimately become obsolete. Journal articles would survive, though perhaps they’d become something closer to definitive pronouncements on major breakthroughs or on the state of knowledge in a given domain—a bit like what scientific books were before the rise of journals. In a world where science is much more productive than it is today, legitimacy will be the scarce factor in the production of useful scientific knowledge.

I am not sure I have experienced the culture of engineering described here in my own engineering career, but this may be referring to something more like “product development” rather than the public infrastructure and environmental planning type work I do. Anyway, my idea of AI-backed knowledge synthesis plus this idea of open data and code may be on to something.

instructions for the AI scientist

Leave a reply

Here’s an example of how AI science could work. If you ask machine learning to review a data set from a system and predict the future behavior of the system, it can often do a good job, but the mathematical approach it is using to do that is mostly opaque. It is also divorced from any sense of the system structure, its interaction with the physical universe, and the physical laws governing it. But you can ask the computer to constrain itself within those known physical laws, and then it may be able to provide you insights on the structure and physical processes inside the system. So at that point, you and/or the computer should be able to form scientific hypotheses and test them against the data. This example from Water Resources Research is about soil moisture.

Coupling Intuitive Physics Into Deep Learning for Soil Moisture Flow Processes Learning

Soil water flow processes in the unsaturated zone support ecosystems and regulate water, energy, and biogeochemical cycles. Recently, deep learning (DL) approaches have significantly advanced soil moisture (SM) prediction tasks yet still challenging to interpret. It’s difficult to peer into the internal reasoning procedures of algorithms, let alone associate them with specific physical processes. Thus, DL alone is unlikely to satisfy soil hydrological modeling needs and cannot advance process understanding. Here, we present DPL-S (deep process learning for SM dynamics) approach, which couples intuitive physics into deep learning architecture as structural guidance to facilitate comprehensive surrogate modeling of soil water flow. DPL-S discretizes the SM state evolution into multiple sub-process effects (e.g., gravity, matric potential) at the intuitive physics level and abstracts them into format-specific and learnable tensors. By cascading state-action matrices in a differentiable end-to-end framework and enforcing penalties for physical inconsistencies, DPL-S enables a profound understanding of physical functions and scenes of soil water flow. Comprehensive numerical experiments including layered soil conditions and tests with in situ observations, demonstrate that it achieves reliable SM profile reconstruction with predictive performance comparable to the state-of-the-art DL model on supervised items. The internal inference of DPL-S is fully transparent and the tensor representations achieve strict physical realism under limited water content supervision, thus enabling continuous predictions like physical models during the testing period. The model’s flexibility, generalization, noise resistance, and large-sample diverse data synergies are also evaluated. This work represents a solid step toward learning hydrophysical processes from large data sets.

the future of scientific and engineering modeling

Leave a reply

I’ve been using AI to assist me with coding (R, in my case) since shortly after ChatGPT came out. In engineering, we tend to run off-the-shelf models that were of course written in some kind of code. Sometimes these are open access but often they are proprietary. The brutality of market discipline pretty much requires specialized off-the-shelf solutions in industry because customers are not going to be willing to pay for custom coding. The proprietary ones are even often preferred for legal/liability reasons. Anyway, the future of modeling appears to be humans providing a detailed specification to an AI agent which then follows it to do the coding, debug the coding, run the model, process and present the results. The humans have to be able to detect whether the results are BS, of course, at this point in history. One can imagine using a different agent or a more specialized agent to assist the humans in the bullshit-detection stage, that agent getting more independent over time, and so in a cycle. I wonder if we will be using agents to set up, run, and post-process the specialized models, or if things will trend toward just letting the agents write more fundamental code over time. Or maybe the specification will be what future scientists, engineers, and business people focus their efforts on, with translating that into 0s and 1s being basically a commodity done on the fly by AIs. This makes sense to me – the most crystal clear function of AI so far, in my view, is making it easier and easier for humans to communicate with computers in more abstract language, logic, and mathematical symbols.

Anyway, this example used something called Roo Code which included a couple versions of Claude Code along with some other agents, to run a fishery-related model. There is a peer reviewed article, but I also like this blog post and this example of a specification given to the agents.

MEMOP, MOP, and SMOP

Leave a reply

These are some more decision-making frameworks I hadn’t heard of, at least by name. I am hopeful that AI can help techniques like this make the transition from gown to town.

Developing Robust Management Pathways for Nutrient Pollution in Watersheds Under Climate Uncertainty

Nutrient management represents an enduring effort toward sustainability. However, long-term management planning faces notable challenges, mainly due to substantial investments required under uncertainty of forthcoming climate. To address these challenges, this paper proposes and tests a Multi-climate-scenario (MCS) Multi-epoch Multi-objective Planning (MEMOP) framework (in combination MCS-MEMOP). This framework divides the long-term planning horizon into multiple epochs, allowing nutrient mitigation measures (e.g., fertilization management, filter strip) to be initiated at any epoch, each with its own water quality and investment constraints. To tackle climate uncertainty, it incorporates principles of Robust Decision-Making. MCS-MEMOP generates solution pathways outlining the timeline and progression of management measures, tested here for a case of a small, agriculture-dominated watershed. Considering a single climate scenario, the MEMOP method was compared with Multi-objective Planning (MOP) and Stepwise MOP (SMOP) for a 25-year nutrient management horizon, using the SWAT model to evaluate the test case water quality effects of solution pathways. Results show that MEMOP’s multi-epoch approach generates a larger and more diverse set of solutions than MOP and SMOP, offering greater flexibility to select optimal trade-offs among objectives. Additionally, MEMOP solutions exhibit superior cost-effectiveness compared to MOP and SMOP solutions. Applied separately to different climate scenarios, the MEMOP results show that changed climate conditions may significantly alter the Pareto front. In contrast, MCS-MEMOP yields robust solutions that can consistently satisfy 72%∼89% of epoch-specific constraints under new climate conditions in the test case, with a cost increase of 12% that reflects the price of addressing climate uncertainty in this case.

March 2026 in Review

Leave a reply

In fast-moving current events as I write on April 4, 2026…I have nothing left to say about the stoopid war in Iran and our stooopid war criminal “leaders” who chose this path. I tell my children “stupid” is a bad word that nice people don’t use, and I don’t use it lightly here. I just hope when I am reviewing April 2026 a month from now at least the part where human beings are dying daily from shooting and blowing up is over.

Most frightening and/or depressing story: The idea of the lone psychopath developing a bioweapon in their garage with AI assistance is very scary. I outlined some proposals out there for how to deal with this issue, but none are really completely satisfying. Of course, nuclear proliferation is always a close runner-up.

Most hopeful story: I took my first Waymo rides in the Phoenix area in March, and I observed Waymo being tested in Philadelphia. I would like to live in a society where transportation is oriented around walking, cycling and other very light personal vehicles, and public transportation. But given that the U.S. is unequivocally not headed in that direction, I think autonomous vehicles are going to be a win for safety, mobility, and the environment in most U.S. cities.

Most interesting story, that was not particularly frightening or hopeful, or perhaps was a mixture of both: I finally got around to reading The Singularity is Nearer. Kurzweil is very big on cultured meat, vertical farming, 3D printing, and generally using computer simulation to super-charge scientific and technical progress in many areas. Then there are his weird ideas about nanobots in our brains allowing us to upload our brains to the internet sometime in the 2030s – sounds crazy at first, but I could sit down and name a few things that sounded crazy a decade ago and are now commonplace. I mused about when the robots are coming now that we seemingly have their AI brains ready for transplant. I mused about the seeming paradox that AI is increasing demand for dirty fossil fuel energy and its attendant impacts while also representing some possibility of a longer-term solution to those problems. It seems like slowing down the deployment of AI is not on the table, so the important question becomes how long is “longer-term” – if measured in single digit years, we just may pull through, but if measured in multiple decades, we may be sunk. Anyway, I brainstormed a list of specific areas of research AI may be able to boost: incremental improvement and deployment of today’s solar, wind, battery, electrification, and electric grid technology; fusion power; safer, more cost-effective and scalable fission power; space-based solar technology; cutting edge materials science and energy storage technology; and fundamental research into the mysteries of the universe, which also comes with attendant risks.

teacher vs. chatbot?

Leave a reply

This article is about an English teacher trying to figure out how to cope with AI in and out of the classroom. The main ideas seem to be just in-class writing assignments and discussions. This is high school (I think), and one thing that might surprise an older reader is the idea that the teacher spends a significant amount of time reading out loud to students, and also giving them time to read themselves in class. I am not surprised because I see the teachers of my own children in middle school doing this.

I’ve thought about trying my hand at teaching, at some point. I may or may not ever do that, but if so I will have to think about how to use AI in an engineering classroom. I don’t have all the answers of course, but this is a bold new frontier so neither does anyone else. Here is my first attempt at a brainstorming list:

  • In-class writing assignments,
  • Paper and pencil tests (or portions of a test – maybe if there is a 2-3 hour chunk of time to take a test, you have the first hour to do the paper and pencil part, hand that in, and get the second part where you are allowed to at least use a spreadsheet program. Or if the class period is only an hour, you could give a test over two days)
  • Journaling – I’ve always thought that reading and writing needs to be a part of what students do, even in STEM fields. In my own classes, teachers have tended to spend almost all the time focused on doing math problems. The problems are critical and I plan to come back to them. But I learn by reading and writing first and foremost, and there are others like me. So my idea is to have short reading assignments and ask students to summarize key points in a few bullets or sentences to show that they did the reading. Then, crucially, I would ask them to relate the reading to a relevant outside reference – kudos for a peer reviewed article, but could also be a relevant news or magazine article, or even a Wikipedia page. Now, obviously students are going to use AI for this, and that is okay. There would have to be classroom discussion related to it, and points given for being able to discuss what they wrote intelligently.
  • Okay, now to those problems, homework assignments, and/or take-home exams. These have always been the beating heart of engineering education. This is the struggle and the hard work of relating theory to problem solving, resulting in deep understanding. This is also where students benefit from individualized feedback from teachers and collaboration with peers. They are going to interact with AI here, and I see no real alternative other than classifying AI as not cheating. In fact, this is where you may want to actually encourage them to experiment and openly share what effective techniques they found that they think enhanced their learning. You have to give some points for the homework, otherwise all but the most-disciplined students will not make the significant effort needed for learning. But most of the points should probably be given for doing supervised in-class assignments that are similar to the homework.
  • There’s another problem here though – distance learning has become pretty common, particularly in graduate school. And that throws a monkey wrench into the whole in-person thing. It is potentially unfair if some students are in-person and some are virtual – are you going to surveil students on-camera the whole time and try to ascertain if they are cheating? This is a tough one and I don’t have all the answers.
  • Then there is the traditional “term paper” or “research paper”. This is a traditional cornerstone, but students are just going to use AI for it. Asking them to present their results and asking them personalized questions is part of the answer here, but that doesn’t really help them with feedback on their writing. The in-class writing assignments have to be part of the answer here. But again, in terms of out-of-class-writing, I think you have to classify the use of AI as not cheating and actually encourage them to experiment and share what they thought was effective.

Well, that was just some stream-of-consciousness drivel, wasn’t it? For me, writing and thinking are inseparable. I had some thoughts on this topic in the back of my mind, but now that I have tried to articulate them (without the use of AI), they are more fully formed than they were before I did that. If I were going to think about this more right now, I might ask AI to help me find some more articles and blogs on the subject.