The legacy of string theory, a generation of physicists content to play with math

Spintronics will support much faster electronics using less power

YouTube:

Spintronics is short for “spin electronics,” and refers to the study of the spin of the electron. In electronic devices, spintronics leverages the spin of electrons to process and store data with extreme efficiency – this technology is just a few years from reaching the consumer market, and will make your devices faster and more efficient. For a price, of course. Let’s take a look at how spintronics got here and where it’s going.

Energy Abundance, Genetic Engineering, Super Intelligence: The world is changing dramatically

The technology is coming, but we can't foresee the social and political consequences. Hossenfelder, however, is skeptical about the survival of current democratic systems, which she discusses about 7 minutes into the video. She doesn't think the European welfare system will survive.

Will AI Save Physics? Probably not.

Hossenfelder is skeptical. 1) It's unlikely that AI will achieve breakthroughs by analyzing data. Why? because we don't have much data on the phenomena of greatest interest. 2) Perhaps there's something to be done in analyzing the published literature, but she's skeptical about that. Why? Because the current literature, though huge, consists mostly of recycling the same ideas (the ones that don't work). 3) Then there's theory development. Current systems aren't good at it because they need to be trained on something. You mean the current crop of "junk" theories? And future AI? Perhaps. But we need something better than logical AI.

A possible axiomatic derivation for the “arrow of time”

YouTube:

David Hilbert’s Sixth Problem is 125 years old and asks for an axiomatic foundation of physics. A good place to start with this, said Hilbert, would be fluid dynamics – physicists should be able to prove that our fluid dynamics equations are rooted in how we understand atoms behave when they bump into each other. This would also explain the origin of irreversibility in our lives, or the “arrow of time” as physicists like to say. Over a century later, mathematicians have made a major breakthrough in this arena. Let’s take a look.

That is, we may have an explicit way of showing how irreversible behavior can emerge from underlying reversible laws.

Sabrina Hossenfelder: The AI Revolution Hiding in Obscure Research

Hossenfelder believes that current LLM-based AI is slowly colliding with that wall Gary Marcus has been talking about. I agree with her. We need new paradigms. Starting at c. 3.55:

The new paradigm will come from models that learn by interacting with the world and that can continue to keep learning after training. These are the so-called “world models”. One step towards those good is DeepMind's Genie 2 which they announced in December. Genie which was trained on a large video dataset and generates interactive 3D environments. They can then place AI agents into these virtual worlds so that they can learn to learn. DeepMind isn’t the only one who has announced progress with world models. In January NVIDIA introduced the Cosmos platform, which also generates 3-dimension models with the laws of physics built in. For one thing, such models can be used to generate videos in which objects don’t appear out of nowhere and where perspective is consistent. But more importantly, they can be used to train other models so that they learn how reality works. This makes sense to me because it’s how human intelligence evolved. We’re interacting with the physical world and creating our own mental models of it. Deepmind calls it a foundation world model, and such world-models are almost certainly going to play a big role in the next big AI revolution. The next big steps will be systematic upgrades to reasoning capabilities and not just more training. And they’re working on it. Basically, the future is here already, it’s just stuck in obscure academic journals.

Well, not quite. I don't think the future's made it to the journals just yet. Give it a couple of years.

Sabrina Hossenfelder looks at Claude 3.5 Haiku and discovers that it lacks self-awareness

In this video Hossenfelder examines a paper recently published by Anthropic: On the Biology of a Large Language Model. She has a number of interesting observations. After reiterating that the underlying LLM is in the business of next token prediction, she reviews an example of arithmetic reasoning starting at about 1:54. After following the example more or less step by step she observes (c. 2:45):

It’s basically a heuristic text-based approximation. It’s doing maths by free-associating numbers until the right one just sort of vibes into place. But here is the kicker. If you ask Claude how it arrived at that result it says “I added the ones (6+9=15), carried the 1, then added the tens (3+5+1=9), resulting in 95.” Which is not what it did, not even remotely. It answers this question separately, giving you again, a text prediction for the answer. And I think that this shows very clearly that Claude has no self-awareness. It doesn’t know what it's thinking about. What it tells you it’s doing is completely disconnected from what it’s actually doing.

That makes sense to me. Claude really "doesn’t know what it's thinking about." Why would we think otherwise? Well, I suppose, because it tells us what it did and we treat that statement as equivalent to what humans do when they report how they accomplished some task. The thing is, humans don't necessarily know what they're doing either.

Consider and example from Piaget's 1976 book, The Grasp of Consciousness. Here's how I reported that experiment in a paper I published some years ago, First Person: Neuro-Cognitive Notes on the Self in Life and in Fiction (2000):

Let us begin with an experiment conducted by Jean Piaget as part of an investigation into consciousness. In this experiment children were asked to crawl for about 10 meters and then to describe what they had just done (Piaget, 1976, pp. 1 ff.). Four-year olds generally said either that they first moved one arm, then the other, then one leg, then the other, or legs first and then arms. Piaget called this a Z pattern. That is not, in fact, how any of them actually crawled. What they actually did was either to first move one arm, then the opposite leg, then the other arm, then the opposite leg, or the same pattern beginning with a leg. Piaget called this an X pattern. It isn't until children are seven or older that they can describe this X pattern. [...]

What is striking is that the younger children's verbal account of such a basic act is simply wrong. In order to execute the crawl there must be some brain tissue devoted to schemas regulating the appropriate actions; for crawling isn't a spinal reflex. But those regulating schemas must in some way be distinct from the schemas underlying the younger children's verbal accounts, otherwise those accounts would be more accurate. My first point is simply that we are here dealing with two different neural schemas for the same action and that one of them is grossly simplified and thus incapable of actually regulating the behavior it represents.

There are many ways in which the behavior of these four-year olds is quite different from Claude's, but I see no need to list them all. They're obvious enough. The basic point, as I say in that second paragraph, is that the young children got it wrong. And, while older children do get it right, there is a separate body of research that shows that the kind of behavior those four-year olds exhibit is common in various domains.

That research is about what is called the introspection illusion, which Wikipedia characterizes as follows:

The introspection illusion is a cognitive bias in which people wrongly think they have direct insight into the origins of their mental states, while treating others' introspections as unreliable. The illusion has been examined in psychological experiments, and suggested as a basis for biases in how people compare themselves to others. These experiments have been interpreted as suggesting that, rather than offering direct access to the processes underlying mental states, introspection is a process of construction and inference, much as people indirectly infer others' mental states from their behaviour.

Notice that last statement, “introspection is a process of construction and inference.” Isn't that what Claude was doing? It provides a plausible account of its activity, not on the basis of some examination of that activity, but rather, based on what it knows about how arithmetic is (supposed to be) done. Claude doesn't have introspective awareness of what it is going, but then neither do humans, at least not in some wide variety of cases.

Let’s return to Hossenfelder, who goes on to assert: “I’d say that self-awareness is a precondition for consciousness.[1] So this model is nowhere near conscious.” I agree with her that Claude is not conscious, but not for the reason she gives. Those four-year olds in Piaget's experiments were certainly conscious, but they lacked (a certain kind of) self-awareness. I'm inclined to think that self-awareness and consciousness are distinct mental phenomena. On consciousness, I favor the view expressed by William Powers in his 1973 book, Behavior: The Control of Perception. Powers’s account is subtle, more than I can explain here (I do explain it in a post from 2022). Suffice it to say that Powers’s account is grounded in the behavior architecture of the brain. LLMs simply don't that the required architecture. They may well talk as though they’re conscious, but they’re just faking it. It’s empty talk. 

* * * * *

[1] FWIW, I suspect that this mistaken belief is widespread.

Those huge structures recently "discovered' beneath the pyramids are probably imaginary

Sabine doubts those amazing findings of huge structures supposedly beneath the pyramids.

Sabrina's drones, they're everywhere!

if I was trying to spy on foreign countries my drones would look like this

(Photo: Tom Koerner/USFWS/Wikipedia) pic.twitter.com/Ps2D9RXPgP

— Sabine Hossenfelder (@skdh) February 2, 2025

Sabina Hossenfelder on DeepSeek and its implications

4:08: As we discussed previously, some experts including Yann LeCun and Gary Marcus, doubt that Large language models will ever get us to general intelligence, and I agree with them. The market will only react to this once it becomes clear that the existing LLMs can’t be recued with further updates. At that point a lot of money will evaporate like this. But at the moment, the stargate project is just pumping more money into an existing bubble. Building power plants, extending the grid, and improving data infrastructure generally seems like a good idea, and all these are part of the stargate project. 

But to me the Stargate Project is as crazy as if Americans had taken the first semi-conductors and spent $500 billion on factories to produce them, rather than letting markets do their thing and wait for technological developments to make microchips smaller and cheaper. Ie, to wait for them to make economic sense. It’s like the dot-com bubble, except instead of getting free T-shirts from Pets dot com, we get hallucinating chatbots and 17-hour debates about whether sentience can be monetized.

5:17: That said, the arrival of DeepSeek drives home an important message: you can save a lot of money if you let Americans do the heavy lifting and then build on that knowledge. And that goes well with the European approach, which is basically to wait and see what goes wrong in America. I’m not usually a fan of the European risk-aversity. It reminds me of how my younger brother was waiting for me to touch the electric fence. But in this case “wait and see” might indeed work out to our advantage. And if not, we’ll always be here to give Americans lectures about responsibility, sustainability, and how our regulation-heavy bureaucracy prevents us from having fun.

Hossenfelder: Wake up, people! The private companies that dominate AI are going to rule the world.

Hossenfelder:

Most politicians totally misunderstand the trouble that artificial intelligence is going to bring. This isn’t a race for profit, it’s a race for power. And that power will be in the hands of a few very rich people. Does that sound like a good future?

From the video:

2:36: The mistake that all these politicians make is to think of AI as a race for profit or prosperity. But this is much rawer. It’s a race for power. Whoever will first be in possession of AI with superhuman intelligence will rule the world. And at that point it won’t matter where the company is registered, because they’ll have multiple backups elsewhere to avoid a forced nationalization. Because once governments realize that they are no longer in control, that’s what they will try. But by then it’ll be too late. [...]

3:55: A good way to think about the “frontier models” is as a new operating system. Technically it’s not what they are, but practically, it’s how we will use them. You will sign up to one of them and use that AI to do everything on your devices. You'll use AI to write your emails, pay your bills, and procrastinate better than ever before. And for governments and other companies, that will become indispensable quickly. If they don’t sign up, they won’t be able to compete. They will need access for their financial management, military strategies, policy evaluations, everything. Not having AI access in 5 years will be like giving up on the internet today. And who will control that access? The people who own the companies. [...]

5:47: And this is what the AI race is about. It’s about world-domination. Why is Elon suing OpenAI and building his own AI if he doesn’t need the money and has enough on his hands already? It’s because he wants to rule the world. What sense does it make that Sam Altman, the CEO of OpenAI, declares both that superintelligent AI is going to be super dangerous and that the only way to find out how dangerous is to actually build it. If it’s so dangerous, why doesn’t he worry? Because he’ll be the one in power. The rest of us will be living in the metaverse, mining bitcoin to pay his energy bills. [...]

6:53: If I was the Queen of Europe, I would take the money which was earmarked for that bigger particle collider, a few dozen billion dollars or so, and instead pour it into a publicly owned frontier model like yesterday. Because otherwise, Europe will be in even bigger trouble 5 years down the line than it is now.

AIs Predict Research Results Without Doing Research

From the YouTube page:

Scientific literature is growing rapidly, meaning scientists are increasingly unable to keep up with all of the latest developments in research. AI large language models, though, can read and “digest” information much more quickly than their human counterparts, making them the perfect tools to conduct massive literature reviews. Recent research shows they’re also very accurate at predicting the results of studies that they’ve never read before. Let’s take a look.

Hossenfelder references this article:

Luo, X., Rechardt, A., Sun, G. et al. Large language models surpass human experts in predicting neuroscience results. Nat Hum Behav (2024). https://doi.org/10.1038/s41562-024-02046-9

Abstract: Scientific discoveries often hinge on synthesizing decades of research, a task that potentially outstrips human information processing capacities. Large language models (LLMs) offer a solution. LLMs trained on the vast scientific literature could potentially integrate noisy yet interrelated findings to forecast novel results better than human experts. Here, to evaluate this possibility, we created BrainBench, a forward-looking benchmark for predicting neuroscience results. We find that LLMs surpass experts in predicting experimental outcomes. BrainGPT, an LLM we tuned on the neuroscience literature, performed better yet. Like human experts, when LLMs indicated high confidence in their predictions, their responses were more likely to be correct, which presages a future where LLMs assist humans in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours.

Hossenfelder on stagnation: “Science is in trouble and it worries me”

Show notes:

Innovation is slowing, research productivity is declining, scientific work is becoming more [less] disruptive. In this video I summarize what we know about the problem and what possible causes have been proposed. I also explain why this matters so much to me.

00:00 Intro
00:33 Numbers [1]
06:33 Causes
10:32 Speculations [2]
16:25 Bullshit Research
22:06 Epilogue

I've given a fair amount of attention to this topic under the label, stagnation. I've also written a working paper: Stagnation and Beyond: Economic growth and the cost of knowledge in a complex world, July 25, 2019.

[1] Here she opens by citing results from Bloom et al. Are Ideas Getting Harder to Find? (2017), and other more recent papers. 

[2] Mentions Fast Grants, with clips showing Patrick Collison and Tyler Cowen.

Hossenfelder: Wolfram's research program seems healthy (after all). Perhaps it can work.

From the webpage:

Mathematician and Computer Scientist Stephen Wolfram wants to do no less than revolutionizing physics. He wants to do it with computer code that gives rise to all the fundamental laws of nature that we know and like -- and maybe more. Unfortunately, Einstein’s theories of general relativity inherently clash with how computers work. And yet, he and his team might have found a clever way around this problem.

Hossenfelder: The 2024 Nobel Prize in Physics Did Not Go To Physics -- This Physicist is very surprised

Hossenfelder comments: "A quick comment on the 2024 Nobel Prize in physics which was awarded for the basis of neural networks and artificial intelligence. Well deserved, but is it physics?" She also wonders if this does't reflect (what she regards as) the dismal state of current work in the foundations of physics and points out that physicists have been using neural networks for years as tools.

Herbert Simon won the Nobel Prize in Economics in 1978 for "for his pioneering research into the decision-making process within economic organizations." The Nobel Committee did not mention his work in Artificial Intelligence. But would he have gotten the prize without that work? 

* * * * *

Gary Marcus has an interesting post on the physics Nobel, for AI, and the chemistry as well: Two Nobel Prizes for AI, and Two Paths Forward:

Let’s start with Hinton’s award, which has led a bunch of people to scratch their head. He has absolutely been a leading figure in the machine learning field for decades, original, and, to his credit, persistent even when his line of research was out of favor. Nobody could doubt that he has made major contributions. But the citation seems to indicate that he won it for inventing back-propagation, but, well, he didn’t.

He goes on spell out a more detailed history of early work in neural nets, citing remarks by Steven Grossberg and Jürgen Schmidhuber.

Why is scientific progress slowing down? Science is making fewer and fewer breakthrough discoveries.[Hossenfelder]

What's going on? 

1. Nothing at all. Things are just fine. Most scientists prefer this.
2. There's nothing left to discover.
3. Current arrangements award productivity over usefulness.

Sabrina Hossenfelder on why the academy is broken [+ Mark Changizi]

"...unlike academic research, this is an honest trade. You get some of my knowledge. I get some of your attention. I like the simplicity of that" (12:53).

For my own version of Hossenfelder's story, see this post: The changing terms of my Socratic bargain with the American Academy [and the larger search for truth]. I've included that on pp. 19-25 of a working paper, Crisis Among HUMANITIES DISCIPLINES in the Twenty-First Century. 

Here's what Mark Changizi wrote when he left academia in 2010:

In academia grant-getting is paramount. Universities are a business. Not a business of student education, and not a business of fundamental intellectual research. Universities are in the business of securing grant funds. That’s how they survive. And because grants are the university’s bread and butter, grants become the academic professor’s bread and butter.

Getting grants is the principal key to individual success in academia today. They get you more space, more money, more monikers, more status, and more invitations to lunch with the president.

To ensure one is in the good graces of one’s university, the young creative aspiring assistant professor must immediately begin applying for grants in earnest, at the expense of spending energies on uncertain theoretical innovation.

In order to have the best chance at being funded, one’s proposed work will often be a close cousin of one’s doctoral or post-doctoral work. And the proposed work – in order to be proposed at all – must be incremental, and consequently applied in some way, to experiments or to the construction of a device of some kind.

So a theorist in academics must set aside his or her theoretical work, and propose to do experimental or applied work, where his or her talents do not lie.

But if you’re good at theory, you really ought to be doing theory, not its application.

There's more at the link.

Physics for little girls

Give her physics! Physics for all the little girls!

From "The Silver Chair" by C.S. Lewis, 1953 pic.twitter.com/ezENmu7Uf9

— Sabine Hossenfelder (@skdh) January 4, 2024

 

Tiby Kantrowitz's booth at Maker Faire 2014 in Queens

Sabine Hossenfelder: AI is systemic change, a huge one at that.

Her tweet:

We need to stop obsessing about the perceived shortcomings or virtues of individuals. The relevant changes in human society are systemic. Individuals rise or fall in that system. If it's not one of them, it'll be another.

The real stories aren't Trump or Thunberg or Musk, the real story is the system that made those people rise to popularity and power.

This is why AI is such an important development -- it's a systemic change, and a huge one at that.

Trump too should be seen as a symptom of a disease and not a cause. It's partly a problem specific to the US American financial and political system, but partly he is symptomatic of a general problem with democracies that they're too slow and unresponsive for modern times.

Thunberg was a condensation seed at the right place at the right time for a community that was looking for a focus. If it hadn't been her, it'd have been someone else.

It makes no sense to psychoanalyse and complain about people in those positions, it'll not make any difference in the grand scheme of things.

Who's an expert on climate change?

40% of U.S. adults think they know more about climate change than climate scientists.

Source: https://t.co/C1W5CwgeJM pic.twitter.com/Q75guIxLqw

— Sabine Hossenfelder (@skdh) November 22, 2023