Bonus Tab-Clearing Post: Physics and AI

Article over at Quanta.  Impossible to describe, go read the whole thing.

And yes, Demers' take is pretty close to mine too.

Book Mention: Take Me to Your Leader

Continuing this week's alien theme, I finally finished reading through Neal deGrasse Tyson's latest book, Take Me to Your Leader last night.  It's a fun little romp through all of the pop culture, credible sworn testimony, conspiracy theories, and blurry pictures of the last eighty or so years, along with reasonable musings on the Fermi Paradox, what forms extraterrestrial life (and their perceptions and senses) might take, etc. etc.  With all that, we are left with one stark, undeniable observation:

There is not one whit of actual evidence of alien visitation.  No artifacts, no hard data, and, despite upwards of seven billion digital cameras on this planet, no unambiguous pictures.

With that uncomfortable truth in mind, the author plows forward with his usual good-natured skeptical, slightly sarcastic way, considering all sides of the issues and philosophical conundrums.  It is very clear that he's open to any evidence, even an outright personal abduction.  However, to date, there just isn't any there there, and he's quick to gently point this out.

Recommended, but only if you generally like this sort of thing.

Finally, if you do manage to get yourself abducted, please take Seth Shostak's advice and pocket an alien artifact on your way out the door/hatch/portal/whatever.  We've all had more than enough of heartfelt testimony on these matters.  It's time for a metamaterial multitool, or whatever else they're using these days.  Plonk it down on the table in front of skeptics, say a hearty "nyeh-nyeh-nyeh," donate it to the Smithsonian Air & Space Museum, write a bestseller, and enjoy the good life.

Call Went to Voicemail

This week's episode of Big Picture Science is titled Calling All Aliens, which is naturally enough, about recent SETI and METI efforts.  By chance, I was asked by a friend earlier this week about my opinion on the general topic.  The three thoughts I had on the matter were:

1. 
   Imagine our galaxy as a pizza – a really big pizza – one meter across and one centimeter thick.  Now imagine a bubble of space radiating out from Earth in this pizza-galaxy.  A bubble big enough for our first powerful radio signals to have reached another civilization, and for them to have sent a reply back.  At the one meter pizza scale, that bubble would be about the size of a sesame seed.  Would we really expect to have reached anyone in such a small sample?
2. More quantitatively, take a current credible guesstimate at Drake's Number – about 86,400 contactable civilizations in our galaxy.  Seems a little high in comparison with numbers I've previously seen, but let's roll with it.  Given the speed of light, expect about a 50-50 chance of that bubble in Item 1 to encompass another civilization in about 330 years.  What's more, these Drake Equation results are all over the place, and numbers as low as 100 are reasonable.  That puts the expected time to a 50-50 chance of contact out around 3200 years.  Frankly,* it's just too soon to reasonably expect any return replies.
3. However, we should still keep listening, especially given the continuing and amazing increases in signal processing power.  Don't let those two-way contact numbers in the previous item discourage you.  Unlikely as it seems, a random alien signal could come crashing in any given day – and we'd best be listening.
4. Finally, if aliens ever do find us, they'll probably view us about the same way we view the Sentinelese (see inset picture): present, but violent and not worth contacting.  Best left alone.

On that note, time for a bike ride.  Anyway, the above-linked podcast is well worth an hour's listen, and one of the more noteworthy recent episodes  at Big Picture Science.

----------

*10 Internet Points if you caught that one.

A Very Interesting Step Toward Synthetic Life

For the First Time, a Cell Built From Scratch Grows and Divides

No, it is not life.  But it is an interesting conglomeration that pulls together a bunch of previous research into a neat bundle, and makes one of the main features of a living cell sort of work when it is spoon-fed some complicated key ingredients.  From the article:

“I wasn’t allowing myself to believe it for a while,” she said. “It was like, ‘Holy shit, did I actually make a dividing cell?’ … At some point, you’ve been checking enough that [you think], ‘OK, now it’s real.’”

and:

“The modern cell is like a Dreamliner,” Adamala said, referring to the Boeing 787 airplane. “We built a Wright flyer… the first bike frame with wings that flies 100 feet.”

Pretty amazing.

And no, it's not going to get away and take over the world.  Real life would rip it to shreds and chow down on the innards within milliseconds.

Because I've Been Asked About This

Sabine Hossenfelder digs into the somewhat recent, somewhat mysterious disappearance of scientists and a few scientist-adjacent people.  (8 minutes for the analysis, followed by a 2 minute ad)  TLDW: Mostly explainable, but the overall rate is about four times what one might expect in the general population, which is statistically slightly alarming.

Here's a short summary of my thoughts on the matter:

1. There's a significant subset of scientists who have a taste for outdoor adventure, so expect a higher disappearance rate there.
2. What's more, there's also another subset of scientists who will dive into adventures somewhat blindly, so expect trouble there as well.  Usually they morph over time into the previously mentioned subset – if they survive the learning curve.
3. Personally, over the years I've stumbled across a handful of things that could have been trouble: a probable pot farm, a fresh murder scene, a few awkwardly placed rattlesnakes, a couple of unoccupied hobo camps, etc.  Any of these could have been trouble had my GTFO circuits not overridden my curiosity.  I know some people, especially in Subset #2, who do not have such highly attuned GTFO circuits.
4. The relatively low number of missing persons here can account for large statistical fluctuations.  Along with numerous cases of weak GTFO wiring, this is probably enough to account for the observed 4x disappearance rate.
5. But yeah, Hossenfelder's about right on the matter.  The numbers are a little high but not badly out of line with general expectations.

Beyond that and on a personal note, I don't expect to disappear anytime soon.  Some may be comforted by this, others may be a tad disappointed.  Comment below.

PHM Side-Note on Accidents

Following up on last week's review of Project Hail Mary, one of the things that struck me both while reading and watching were the parallels between PHM and the Manhattan Project.  Yes, both were multinational big-budget science and engineering projects carried out in a rush, but that explosion 

boom

put a punctuation mark on things.  And while in the story the explosion did add a dramatic element, it was also the catalyst behind the major plot twist about which the entire story revolves.  Moreover, the incident very much echoes two criticality accidents at Los Alamos that happened just as and just after WWII ended.

Reenactment with inert components.  Don't try this one at home, kids.

Maybe it's my background, but while reading PHM this parallel jumped out immediately and it gave the whole book – and later movie – a ring of authenticity.  In case you want to learn more on the Manhattan Project (and perhaps get the feel for the behind the scenes action in PHM), look no farther than Richard Rhodes' fine 1986 history The Making of the Atomic Bomb.

Yeah, just go read the book.  Both of 'em, I mean.  And see the movie.

I Knew That Stuff Was No Good

Erythritol crosses blood-brain barrier, linked to strokes etc.  Article at Inc, though I'm sure you can find other sources.

I tried that stuff a couple of times in various Monster drinks.  Nasty.  You can just taste that it's no damed good.

A Look at the Science Behind 'Project Hail Mary'

By no coincidence at all, it's this week's topic at Big Picture Science podcast.  Go & download & listen.

I'd put in an image, but just click through and you'll get all the image I'd put here.

Really looking forward to this movie.

Kessler... Maybe?

A couple of years ago I posted a little piece on the Kessler Syndrome here (includes a brief discussion one the topic at hand, if you need a reminder), and the bottom line was "yeah, don't worry about it – yet."

Now Kessler and a co-author have recently revisited the topic, and their results are not so sunny.  You can read the abstract and download the paper here, or listen to Sabine Hossenfelder's take on the matter here.  It's going to be interesting to see this one batted around over the next few years.  With such wildly divergent results, there's going to be a good academic catfight.

My advice: Enjoy your Starlink while you can, but brush up on your HF skills too.  And if it comes to it, enjoy the light show.

Protip: Never exceed any limit labeled "Runaway Threshold."  Don't, just don't.

On the State of Particle Physics

Quanta online asks Is Particle Physics Dead, Dying, or Just Hard?

It's a real quandary, and the author explores all corners of the problem in an even-handed manner.  TLDR: On one hand, (a) the Higgs boson has been found and all the slots in the Standard Model's chart have been nicely filled, (b) the super-symmetric particles that were sort-of predicted at LHC energies simply weren't found, and (c) there are no firm theoretical predictions that we'll actually find anything else of interest at higher energies.  On the second hand, damn the torpedos, build the next generation accelerator and go look at what's out there at higher energies, because you can't know until you look.  On the third hand (see, I've just finished re-reading Project Hail Mary and one of the main characters has five hands; I could keep this up for a little while longer), that $22 billion has to come from somewhere, and we all know without even guessing that that money's going to come out of the rest of science.  Do we shut down other more promising programs just to go have a look "because you never know"?

Feel free to comment, but only after you've read the above-linked article, and then read Lost in Math for a chaser.

Finally this popped up yesterday as well (coincidence?  probably not):

Nasty, but also funny and on the nose.

Can't Wait Until Next Christmas...

... when this book comes out.  Here are a couple of teasers:

• A New Christmas Fable
• Joyous Noel!

Because this is how you get new scientists, engineers, and innovators.

Just Some Random Stuff

Random numbers have figured highly in my work for decades, but I've never had a really great random number generator at hand.  Oh, the ones out of Numerical Recipes have usually been more than adequate, but not provably adequate.  Now however, there's this...

• CURBy – the University of Colorado Randomness Beacon project
• Abstract on Bell-certified quantum random number generation

Probably more random than you bargained for!

Holman on Deep Tech – Pretty Wild Stuff

Back in September Jim Rutt posted an interview with Pablos Holman on the topic – and his new book – on what he terms deep tech.  Holman's central thesis is that we've gone too far down the software road, and while it's been a good road, we need to concentrate more on hardware and, you know, actually building cool stuff.  Autonomous shipping, mosquito-zapping lasers, safe nuclear reactors, high-speed rail, the list goes on.

And he does have a valid point.  In a 100 trillion dollar world economy, less than 5% is in the computer tech sector.  However it somehow vacuums up the lion's share of talent, venture capital, and press.  [Side note: I think a lot of reason tech has thrived over the last half-century is that it's largely immune to regulation and ignored by regulators.  Try to build a new computer operating system and The Man don't care; try to build a new type of trebuchet or nuclear reactor or choo-choo train and The Man will definitely inquire.  Close rant & back to the subject at hand.]

Some parts of both the interview and book come off as eccentric and out of touch with real developments in the real world these past few decades.  Take shipping for example.  Yes, cargo ships are still largely operated by humans.  However the number of crew required to run a ship is a fraction of what it was 75 years ago.  More importantly, containerized shipping has radically slashed port handling costs.  Overall cargo shipping costs have dropped by something like 90% since 1950.  Yes, there are always optimizations remaining, but come on, apply a generalized version of Amdhal's Law to the situation and give the industry another decade to catch its breath.  Dropping the remaining crew of ten to just one is insignificant when compared to what has just happened.

On the other hand, the guy was there at the start of Blue Origin, which just racked up a major success.  Furthermore, rocket boosters dropping out of the sky and landing on robotic barges has become something of an everyday occurrence.  Clearly it pays to think big, and especially in terms of making a profit.

Well, listen to the podcast interview (link above), read the guy's book, flip through the online footnotes for the book, then make up your own mind.  My take here is that, despite some weird twists and turns, the guy's onto something.

Sentience, Self-Awareness, and All That

Last month I listened to Nicholas Humphrey's interview on the Jim Rutt Podcast, and it was a real eye-opener.  So much so that I ran out and got a copy of Humphrey's book Sentience, read it, then listened to the podcast episode again.

Initially what captured my interest was the phenomenon of blindsight, where someone has lost the conscious ability to see due to brain damage, yet retains some vestige of vision at an unconscious level.  Here's a short video clip of a test subject, sans visual cortex due to stroke or injury, successfully navigating down an obstacle-filled hallway.  After such tests subjects often insist that they simply walked down the hall, having no conscious memory of dodging around objects.

This and other similar observations have led Humphrey to surmise that humans along several other candidate species have the capacity for conscious sensation layered onto a more primitive unconscious perception.  This follows naturally into some ideas about the meaning of self, others, and what is commonly termed theory of mind.  I won't further belabor the ideas, but simply refer you to the above-linked podcast for a much better brief introduction.

BTW & because someone's gonna ask: LLMs, conscious or not?  No, none, not at all.  Barely touched on, not even under consideration here.

One rough patch in this material is the lack of a brief glossary.  Humphrey and Rutt bemoan in the podcast how most conferences on these topics spend the first half simply trying to define the terms they're using.   Rutt mentions that he's now on the board of the California Institute for Machine Consciousness; go dig around at that site, maybe you'll find something glossary-ish.  Or perhaps keep a notepad at hand while reading in order to assemble a DIY glossary.  When I get around to a re-read, I'll do just that.

Anyway, they're making good progress, getting much farther than the old Turing Test, Chinese Room scenario, and those sorts of things.  Interesting stuff.  I won't even begin to publicly speculate if or how this bears on numerous observer-centric problems in physics, such as the measurement problem.  However, I wouldn't be surprised if Humphrey's lines of research ultimately converge there as well.

Punctuated Equilibrium Seems to be the Way of Things

Here's a short article about punctuated equilibrium in biological systems, proteins, and (surprise!) languages at Quanta.

I was very surprised to see PE arising in some genetic algorithm work I was doing several decades ago on systems that were entirely non-biological.  I simply remarked on it in the resulting paper and moved on with the main thrust of that research.  Still, the waves of evolution in this system did show significant regularity and could be useful in determining when some point of diminishing return has been reached, and hence when to stop burning computer cycles.

Looking around on the web using an AI search quickly showed that this has been observed many times before:

Give me a list of research showing punctuated equilibrium in genetic algorithm simulations of non-biological systems.

Here is a list of relevant research papers that demonstrate or analyze punctuated equilibrium (periods of stasis followed by rapid changes) in genetic algorithm (GA) simulations applied to non-biological systems, such as abstract optimization problems, engineering tasks, or artificial life models. I've focused on studies where the systems are clearly non-biological (e.g., not direct simulations of real biological processes like speciation or genetic drift in organisms). Each entry includes the title, authors, year, a brief description of the GA and system, and how punctuated equilibrium is shown. The list is chronological.

• Punctuated Equilibria in Genetic Search by Michael D. Vose and Gunar E. Liepins (1991). This paper formalizes a simple GA with selection, mutation, and one-point crossover, applied to non-biological search spaces modeled as high-dimensional dynamical systems. Simulations and mathematical analysis show punctuated equilibrium through alternating periods of relative stability (attractors) and sudden rapid evolution when populations shift between basins, supported by eigenvalue computations and trajectory diagrams.
• Genetic Algorithms and Punctuated Equilibria in VLSI by J. P. Cohoon, W. N. Martin, and D. S. Richards (1991). A distributed GA with subpopulation structures is used for the K-partition problem in VLSI (Very Large Scale Integration) chip design, a non-biological engineering optimization task. The approach is inspired by punctuated equilibrium theory, with empirical simulations demonstrating performance advantages from "ecological opportunities" that enable rapid evolutionary bursts after periods of stasis.
• Population-Based Incremental Learning: A Study on Genetic Algorithms and Simulated Annealing by Shumeet Baluja (1994). This work uses Population-Based Incremental Learning (an extension of equilibrium GAs) compared to standard GAs on non-biological optimization problems like numerical functions (e.g., De Jong’s, Griewangk’s), NP-complete tasks (e.g., jobshop scheduling, traveling salesman), and deceptive problems. It demonstrates punctuated equilibrium in parallel GA variants with subpopulations, where stasis in isolated groups is interrupted by migrations leading to rapid improvements, shown via a contrived example where single-population GAs fail but multi-population ones succeed.
• Punctuated Equilibria in Simple Genetic Algorithms for Functions of Unitation by Sangyeop Oh and Hyunsoo Yoon (2000). A simple GA with roulette wheel selection, mutation, and crossover is simulated on non-biological unitation functions (bit-string optimization in bistable potential landscapes). Theoretical analysis via diffusion equations and simulations reveal punctuated equilibrium, with long metastable periods at local optima followed by exponential-duration sudden jumps to global optima, visualized in population mean trajectories and parameter-dependent duration plots.
• Punctuation Equilibrium and Optimization: An A-Life Model by Ravi Jonnal and Anthony Chemero (2000). An evolutionary algorithm evolves artificial neural network weights for controlling a virtual creature in a 2D grid-based simulated environment with resources and obstacles (non-biological artificial life setup). Simulations compare standard and punctuated mutation rates, showing that introducing rare bursts of higher mutation leads to punctuated equilibrium patterns—long stasis interrupted by rapid fitness gains—resulting in significantly higher overall fitness scores.
• Punctuated Equilibrium and Neutral Networks in Genetic Algorithms by David Shorten and Geoff Nitschke (2022). A simple GA is applied to non-biological benchmark optimization functions (11 numerical ones plus Royal Road and Trap functions). Empirical simulations demonstrate punctuated equilibrium as populations explore neutral networks (genotype spaces with stable phenotypes), leading to periods of phenotypic stasis followed by rapid changes when escaping to fitter regions, analyzed via consensus sequences and genotype-phenotype

Huh.  Effect noted, explained, harnessed.  I wish search engines had been this effective in 2007.  Back to and onward with my real work.

How Generative AI Works

With so much gobbledygook and technobabble being thrown around regarding generative AI, it is refreshing to have someone who knows what the hell he's talking about give a brief explanation on how this stuff actually works.  Here's such a talk by William Press, who you may recognize as the lead co-author of Numerical Recipes:

https://www.slideshare.net/slideshow/how-does-generative-ai-actually-work-a-quick-semitechnical-introduction-to-large-language-models/257932856#1

Fourteen pages total, though not counting the title and that-is-all slides, it drops to eleven.  What's more, the real meat is all on slides 6 & 8, though you'll want to read through the other nine.  I mean, we're talking what, a half-hour tops?  Then you'll be a freekin' genius in comparison to most of the babbleheads out there in the j-school and VC worlds going on about this stuff.

Life is Short and Time is Tight. Read the Abstracts.

Over at Books, Bikes, & Boomsticks Tam writes on the problems associated with reading a summary.  Here, it's short, go read the whole thing: Soulless Drones

<insert schmaltzy interlude music here>

OK, we're all back.  For great art, sure, I'll agree that one has to read / listen / watch the entire piece as presented.  However, if you just want to get caught up on the n movies you missed in the Avengers movie series over the past m summers so that you can go to an afternoon matinee and enjoy it with friends – without having to spend a week of catch-up in front of the tube – nah, go ahead and skip to the Wikipedia plot summaries.  You won't have missed much anyway.  Long-series TV shows that start off promising but take a dip in the second year?  Sure, skip ahead with the summaries and see if the third year picks up to determine if it's worth your time to work through the series.  That politics-tainted speculative fiction (i.e., really bad near-future sci-fi) potboiler that everyone's talking about?  Definitely.  You'll want to know a little about what Cousin Hal or Aunt Margaret are going on about at the next family gathering, but you probably don't want to take that 500 page slog.  Again, when it comes to real art don't take any shortcuts, but when it comes to lesser works, go for the bottom line and get on with life.

Taking this a step further, who doesn't cull through the titles, then skim through the half-dozen abstracts, then really read only one or two papers each month in their favorite journals?  There just isn't time enough to keep up in a field any other way.  

"Oppenheimer" (the movie) Dissected

When Oppenheimer came to theaters, I was looking forward to a glorious movie and a glorious theater experience.  In short, the shiny new theater I picked presented such a customer-hostile environment that I walked out before even buying a ticket, and bought a copy of the disc a few months later.  Then I watched it, and was appalled by the lack of character development on the interesting historical figures, while the director instead chose to linger for a third of the film's runtime over a dull, neurotic girlfriend.  In short, this film was a beautifully made complete waste of time and right now that disc is wasting my shelf space, but only because I haven't yet thrown it out.  I haven't written a review until now either, because the film's lack of cohesive content made commentary difficult.  There really wasn't much to say about it beyond a simple "don't bother" delivered months after it had left theaters, so I didn't bother.  Giving this thing any kind of n-stars movie rating would be akin to reviewing one of those semi-funny AI generated Star Wars / Miami Vice mash-up videos.  Sometimes interesting visuals, but really nothing beyond that.

Anyway, here's a writer's take on the film, and it's not pretty (link, just under one hour).  If the hour length  gives you a case of the TLDWs, here's her summary:

• Unnecessary Complexity
• Inconsistent Pacing
• Inadequate  Context
• Failure to Establish Thematic Framing

Whew.  The poor woman actually watched the damn thing three times (that's nine hours!) to bring us this analysis.  I wish there was an obvious link from her review video for some sort of coffee fund donation, because that kind of dedication deserves more recognition.

Hubble at 35

Yesterday was the 35th anniversary of the launch of the Hubble Space Telescope, and it's still doing remarkably well.  Here's a good  summary article at space.com

Something for an Easy Day

Of course, I have engineer friends who would further approximate a cow-sphere with a cube.  Hey, whatever works man.  Just don't press it too far.