Jim Keller talks about processor design

Dr. Ian Cutress, An AnandTech Interview with Jim Keller: 'The Laziest Person at Tesla',
6.16.21.

I've spoken about Jim Keller many times on AnandTech. In the world of semiconductor design, his name draws attention, simply by the number of large successful projects he has worked on, or led, that have created billions of dollars of revenue for those respective companies. His career spans DEC, AMD, SiByte, Broadcom, PA Semi, Apple, AMD (again), Tesla, Intel, and now he is at Tenstorrent as CTO, developing the next generation of scalable AI hardware. Jim's work ethic has often been described as 'enjoying a challenge', and over the years when I've spoken to him, he always wants to make sure that what he is doing is both that challenge, but also important for who he is working for. More recently that means working on the most exciting semiconductor direction of the day, either high-performance compute, self-driving, or AI.

Note: This interview is intended for an audience with technical expertise in chip design. If, like me, you lack such expertise, you just have to let if flow and be content with a mere flavor for what's going on.

Matrices, graphs, and vectors

IC: I think you said before that going beyond the sort of matrix, you end up with massive graph structures, especially for AI and ML, and the whole point about Tenstorrent, it’s a graph compiler and a graph compute engine, not just a simple matrix multiply.

JK: From old math, and I'm not a mathematician, so mathematicians are going to cringe a little bit, but there was scalar math, like A = B + C x D. When you had a small number of transistors, that's the math you could do. Now we have more transistors you could say ‘I can do a vector of those’, like an equation properly in a step. Then we got more transistors, we could do a matrix multiply. Then as we got more transistors, you wanted to take those big operations and break them up, because if you make your matrix multiplier too big, the power of just getting across the unit is a waste of energy.

So you find you want to build this optimal size block that’s not too small, like a thread in a GPU, but it's not too big, like covering the whole chip with one matrix multiplier. That would be a really dumb idea from a power perspective. So then you get this array of medium size processors, where medium is something like four TOPs. That is still hilarious to me, because I remember when that was a really big number. Once you break that up, now you have to take the big operations and map them to the array of processors and AI looks like a graph of very big operations. It’s still a graph, and then the big operations are factored down into smaller graphs. Now you have to lay that out on a chip with lots of processors, and have the data flow around it.

This is a very different kind of computing than running a vector or a matrix program. So we sometimes call it a scalar vector matrix. Raja used to call it spatial compute, which would probably be a better word.

IC: Alongside the Tensix cores, Tenstorrent is also adding in vector engines into your cores for the next generation? How does that fit in?

JK: Remember the general-purpose CPUs that have vector engines on them – it turns out that when you're running AI programs, there is some general-purpose computing you just want to have. There are also some times in the graph where you want to run a C program on the result of an AI operation, and so having that compute be tightly coupled is nice. [By keeping] it on the same chip, the latency is super low, and the power to get back and forth is reasonable. So yeah, we're working on an interesting roadmap for that. That's a little computer architectural research area, like, what's the right mix with accelerated computing and total purpose computing and how are people using it. Then how do you build it in a way programmers can actually use it? That's the trick, which we're working on. [...]

CPU Instruction Sets: Arm vs x86 vs RISC-V

IC: You’ve spoken about CPU instruction sets in the past, and one of the biggest requests for this interview I got was around your opinion about CPU instruction sets. Specifically questions came in about how we should deal with fundamental limits on them, how we pivot to better ones, and what your skin in the game is in terms of ARM versus x86 versus RISC V. I think at one point, you said most compute happens on a couple of dozen op-codes. Am I remembering that correctly?

JK: [Arguing about instruction sets] is a very sad story. It's not even a couple of dozen [op-codes] - 80% of core execution is only six instructions - you know, load, store, add, subtract, compare and branch. With those you have pretty much covered it. If you're writing in Perl or something, maybe call and return are more important than compare and branch. But instruction sets only matter a little bit - you can lose 10%, or 20%, [of performance] because you're missing instructions.

For a while we thought variable-length instructions were really hard to decode. But we keep figuring out how to do that. You basically predict where all the instructions are in tables, and once you have good predictors, you can predict that stuff well enough. So fixed-length instructions seem really nice when you're building little baby computers, but if you're building a really big computer, to predict or to figure out where all the instructions are, it isn't dominating the die. So it doesn't matter that much.

When RISC first came out, x86 was half microcode. So if you look at the die, half the chip is a ROM, or maybe a third or something. And the RISC guys could say that there is no ROM on a RISC chip, so we get more performance. But now the ROM is so small, you can't find it. Actually, the adder is so small, you can hardly find it? What limits computer performance today is predictability, and the two big ones are instruction/branch predictability, and data locality.

Now the new predictors are really good at that. They're big - two predictors are way bigger than the adder. That's where you get into the CPU versus GPU (or AI engine) debate. The GPU guys will say ‘look there's no branch predictor because we do everything in parallel’. So the chip has way more adders and subtractors, and that's true if that's the problem you have. But they're crap at running C programs.

GPUs were built to run shader programs on pixels, so if you're given 8 million pixels, and the big GPUs now have 6000 threads, you can cover all the pixels with each one of them running 1000 programs per frame. But it's sort of like an army of ants carrying around grains of sand, whereas big AI computers, they have really big matrix multipliers. They like a much smaller number of threads that do a lot more math because the problem is inherently big. Whereas the shader problem was that the problems were inherently small because there are so many pixels.

There are genuinely three different kinds of computers: CPUs, GPUs, and AI. NVIDIA is kind of doing the ‘inbetweener’ thing where they're using a GPU to run AI, and they're trying to enhance it. Some of that is obviously working pretty well, and some of it is obviously fairly complicated. What's interesting, and this happens a lot, is that general-purpose CPUs when they saw the vector performance of GPUs, added vector units. Sometimes that was great, because you only had a little bit of vector computing to do, but if you had a lot, a GPU might be a better solution. [...]

What I did during the Covid pandemic: Built a miniature Venetia palazzo

Photograph by Jenna Bascom

Sian Ballen and Jeff Hirsh, The Fisher Dollhouse: A Venetian Palazzo in Miniature, New York Social Diary, June 17, 2021.

We were completely mesmerized by the dollhouse created by art collector and patron, Joanna Fisher. Stuck in bed and unable to move during the height of Covid, Joanna discovered the dollhouse, a replica of a Venetian Palazzo, online. Immediately she knew she just “had to have it” but sadly it was sold. She sought out the designer of the original dollhouse, set designer, Holly Jo Beck, and commissioned a replica of the one made forty years ago. Joanna then proceeded to furnish the house with original artworks by artists and artisans all over the world. “The project was my savior, it cost me more than furnishing one of my own homes, but it was worth every penny.”

Fortunately we can experience this extraordinary palazzo in miniature first hand by visiting the exhibition, The Fisher Dollhouse: A Venetian Palazzo in Miniature, on display at The Museum of Arts and Design, through September 26th, 2021.

From the interview:

It’s an obtainable fantasy. Can you walk me through the process? How and when did you come up with the idea?

Well, I’ve loved miniatures since I was about three years old. My mother had a friend in Westchester, where I grew up. And on each wall in her octagonal dining room — instead of art or mirrors — she had a beautiful miniature — like a picture box — of a different room each in a different style. And they were back-lit. I was totally fascinated by them. My mother’s friend had four boys, so they never looked at the picture boxes. They didn’t care. I insisted on going every day after nursery school and she was so thrilled to have me come over. And I would stare at these things and obsess over them. And you know, the minute I was old enough to have my own dollhouse, I had one. My mother and I made one together in my room when I was little.

Did you actually make your first dollhouse?

We made it together with my dad. We made the furniture and the curtains and everything. My mother made paintings and even needle pointed rugs for it. And then when my daughter was born, the first thing I did was set about making her dollhouse, which I still have. It’s in her room in Chappaqua and it’s a beautiful one. Then I became an interior designer for about 30 years doing projects for mostly people that I knew. I did it for fun and enjoyment and I did it while my kids were at school. But Kips Bay asked me to make a miniature one time for their showhouse.

Oh, did they really? When was that?

It had to be about 30 years ago. And I still have it. It’s in my daughter’s room and it’s built into the wall. Fast forward 30 years — not only was I home-bound because of COVID, but suddenly I couldn’t walk. I couldn’t take a step and I couldn’t get any medical attention for it. Everything was being turned over to COVID. It was the height of it.

That’s scary.

I was hobbled, you know, I could not walk. So one night I was searching the internet and I came across this beautiful, old vintage replica of the Gritti Palace in a dollhouse form. And with the help of my friend Amy Kristy — she’s a miniature expert – we tracked it down. And I started to cry. Because it was very expensive and I thought, you know, “I’m not going to buy that. That’s crazy.” And then as time wore on, and the weeks passed, I still couldn’t walk and I still couldn’t do anything. I finally decided I had to have this thing. I said I don’t need any jewelry. I don’t need any clothing. I can’t even buy a pair of shoes. I’m buying this thing! We found the woman (Holly Jo Beck), who’s a set designer in England. She had used it at the Hyde Park Dollhouse Fair, like 40 years ago, as the opening thing for her booth. And then she said to me, “I sold that one, but I can make you a new one.”

Much later:

I know you don’t want to name prices or anything, but give me an example of one thing and how much it costs.

Well, the green chandelier that hangs in the center was $17,000.

There's more, photos and interview, at the link.

Speculative Engineering as Philosophy [Models of the Mind]

I am in the process of preparing a working paper based on my review of Grace Lindsey’s recent book, Models of the Mind. This is draft material that will go in the working paper just before that review.

* * * * *

Reality is not perceived, it is enacted – in a universe of great, perhaps unbounded, complexity.[1]

This started, well, I don’t know, maybe way back when I was six or seven and thought the world was actually a movie projected on a giant screen for the enjoyment of the Baby Jesus, or perhaps when I was a bit older and wondered what there was before the universe came into existence, maybe still order when I thought, as a poem about poetry, “Kubla Khan” held the secret to literary criticism, but really, forget all that. It’s there to be sure, but has no direct bearing. This particular project started with a book, Grace Lindsay’s Models of the Mind, which I reviewed for 3 Quarks Daily. That review constitutes the next section of this working paper. The purpose of this section is to provide a context for that review.

Philosophical beginnings

I saw Lindsay’s book, and to some extent reviewed it, as a work of philosophy, though not philosophy as it exists in philosophy departments. I’m using the word in a different sense, one that I did in fact pick up from a philosopher, Peter Godfrey-Smith.[2] In this view philosophy is a way of making sense of the world in the broadest possible conspectus. That is what philosophy was in the ancient world, but as we developed and accumulated knowledge, philosophers became specialists of various kinds, some became social and behavioral scientists, others became natural scientists, while still others practiced a humanities discipline. Philosophy itself became a humanities discipline, and, as such, became narrowly focused.

But we still need to be able to make sense of the world, all of it, in some way or another. And so in the last several decades we have seen intellectual specialists of one sort or another write books for a general audience – Richard Dawkins, E.O. Wilson, Stephen Mithen, Jared Diamond, Stephen Hawking, and Murray Gell-Mann come to mind, but there are many others (see my post on John Brockman’s Third Culture [3]). While these books may be directed at the general audience, I suspect that they are written to serve their authors’ need to see how things fit together. That is to say, they are written out of philosophical hunger, if you will. As such, they are works of philosophy in this extended sense.[4]

It is in that sense that Models of the Mind, is a work of philosophy. I might even hazard the assertion that it betokens a new philosophy of mind, but that might confuse it with the philosophy of mind that exists in philosophy departments. If I did that I’m afraid I’d be asking the little word “new” to do an awful lot of work. Maybe Lindsay took a course in the philosophy of mind at some point, maybe she even reads around in it, but this book certainly didn’t come out of the questions raised in that discipline. Where does this book come from? Look at the subtitle, How Physics, Engineering, and Mathematics Have Shaped Our Understanding of the Brain. That’s where it comes from. That is to say, it doesn’t come from any one, two, or three, or even five or eight academic disciplines. It comes from many and none. I see this book as part of a larger intellectual development, one not well-defined (which is probably a good thing), that will replace the traditional philosophy of mind, and a few other disciplines as well, with a more adequate approach to understanding the mind and the brain.

The most fruitful conversations about mind and brain have been those between students of neural wetware, on the one hand, and software and hardware (digital and analog) on the other. In particular, it seems to me that those conversations have been far more consequential that philosophical discussions of whether or not computer can think or the brain is a computer (which I take up later in this document). It is time to liberate those conversations from constraints imposed by our Cartesian legacy. That, in effect, is what Lindsay proposes. And that is how I framed my review.

Speculative Engineering and the problem of design

I coined the term “speculative engineering” in the preface to my book on music, Beethoven’s Anvil: Music in Mind and Culture. Here is what I said (p. xiii):

Engineering is about design and construction: How does the nervous system design and construct music? It is speculative because it must be. The purpose of speculation is to clarify thought. If the speculation itself is clear and well-founded, it will achieve its end even when it is wrong, and many of my speculations must surely be wrong. If I then ask you to consider them, not knowing how to separate the prescient speculations from the mistaken ones, it is because I am confident that we have the means to sort these matters out empirically. My aim is to produce ideas interesting, significant, and clear enough to justify the hard work of investigation, both through empirical studies and through computer simulation.

That book is speculative a way that Lindsay’s is not. I aimed for an account of how music works, from the nervous system, in performance, to the social group, from human origins to the present.

Lindsay makes no pretense of presenting a theory about the mind or brain. Rather she offers a review of models, many models, that have been and are being used in studying the brain. But as her subtitle indicates, How Physics, Engineering, and Mathematics Have Shaped Our Understanding of the Brain, she understands that an engineering perspective is important.

As I said, engineering, unlike physics, but perhaps not so unlike mathematics, is about design and construction. If we want to understand how the mind works, we must understand it from an engineering perspective. We want to see how things function, what the parts are and how they fit together to achieve a particular end. Consider this passage where Newell and Simon talk about computer science:

Computer science is an empirical discipline. We would have called it an experimental science, but like astronomy, economics, and geology, some of its unique forms of observation and experience do not fit a narrow stereotype of the experimental method. None the less, they are experiments. Each new machine that is built is an experiment. Actually constructing the machine poses a question to nature; and we listen for the answer by observing the machine in operation and analyzing it by all analytical and measurement means available. Each new program that is built is an experiment. It poses a question to nature, and its behavior offers clues to an answer. Neither machines nor programs are black boxes; they are artifacts that have been designed, both hardware and software, and we can open them up and look inside. We can relate their structure to their behavior and draw many lessons from a single experiment.[5]

“They are artifacts that have been designed,” that’s a crucial statement. The human brain has been designed as well, but not by engineers. Rather it was ‘designed’ and ‘constructed’ in a process of biological evolution taking place over hundreds of millions of years of years. But we must think like engineers if we are to understand how it works.

In my review of Models of the Mind I pick out two models that are particularly important in understanding how the mind and brain have been engineered. One is the 1943 model that McCulloch and Pitts proposed for the function of neurons: think of them as logic elements in an electronic circuit. The other is the Perceptron that Frank Rosenblatt constructed in the late 1950s.

As you may know, McCulloch and Pitts proposed that neurons were logical units in brain-based electrochemical circuits (see p. 10). Their proposal was all but ignored by neuroscientists, found some favor among philosophers – Daniel Dennett was quite struck by it [6] – but researchers in the emerging study of artificial intelligence loved it, for it squared with their preconceptions about the nature of higher mental functioning. Those researchers thought of the mind as processing symbols. This resulted in a great deal of research, at least one Nobel Prize – yes, Herbert Simon’s 1978 prize was awarded in economics but, really, everyone knows he got it for his work in psychology and AI – a lot of hope for useful systems, but not much practical success. The research enterprise, now known as GOFAI (good old-fashioned artificial intelligence) collapsed in the mid-1980s.

What happened? The systems were brittle, failing catastrophically without warning, and required many hours of meticulously crafted knowledge representation (KR), as it came to be known, knowledge generally grounded in some version of logic – many versions were tried. Unlike human or, for that matter, animal minds, GOFAI systems were explicitly designed by humans.

Rosenblatt’s legacy was different (see p. 10). Perceptrons were learning machines. Their initial promise was high, but they failed to deliver. However, Rosenblatt’s ideas were revived and supplemented in the 1980s, as GOFAI was going down, under the guise of connectionism and eventually gave rise to the very powerful systems we see today, which are based on artificial neural nets (ANN). Researchers design a learning architecture ¬– there must be at least as many such architectures as there were schemes for KR during the GOFAI era, programmers implement it, and it computes over some database, of text, images, speech, what have you, and ‘learns’ the structure of objects in the database. Just what it learns, and how, that is somewhat obscure, but it works, after a fashion, but generally much better than the GOFAI systems.

The important point is that these systems teach themselves. AI researcher Yann LeCun has made some remarks that seem relevant to me. This is from a podcast quoted by Kenneth Church and Mark Liberman in a recent article, The Future of Computational Linguistics: On Beyond Alchemy:

All of AI relies on representations. The question is where do those representations come from? So, uh, the classical way to build a pattern recognition system was . . . to build what’s called a feature extractor . . . a whole lot of papers on what features you should extract if you want to recognize, uh, written digits and other features you should extract if you want to recognize like a chair from the table or something or detect...

If you can train the entire thing end to end—that means the system learns its own features. You don’t have to engineer the features anymore, you know, they just emerge from the learning process. So that, that, that’s what was really appealing to me.[7]

That second paragraph is the important one. The system that performs the task at hand, the performing system – whether it is classifying images, translating from one language to another, playing chess, whatever – is not designed by humans. Rather it is designed by an architecture and that architecture is in turn designed by humans.

None of those architectures so far look much like the human brain. That is one thing, and very important. Just as important, however, is the fact that some of them produce very powerful systems, some of practical value. They do so because the design of the performing system has be displaced from human designers and onto a machine.

I take it then, that the problem of design is central to speculative engineering. And it is central, not just as an abstract issue, but as an inquiry into how the design process is implemented in a physical device, whether organic or artificial. How, specifically, does a system fit itself to, learn from, the environment in which it must function?

A timely book

Lindsay didn’t foreground the problem of design in Models of the Mind, but it is there, along with many other issues, or problematics as many humanists like to say. She lays them out for you, one after the other, chapter after chapter. I’ve been reading and thinking about mind and brain for half a century, but I’ve never seen a book quite like this.

It is a timely book, a book we need. It feels foundational. Certainly, students of neuroscience need to encounter it early in their education. But so should students of psychology, of artificial intelligence, and of the mind more generally, even literary critics. That's where I started half a century ago, with questions about the structure of a poem “Kubla Khan,” a structure that somehow smelled of computation.[8] I still haven’t figured out that structure. Possibly I never will. But I am content to leave that task to others, confident that that puzzle is more likely to be resolved in the conceptual universe implied by Models of the Mind than in the universe available to me back in the Jurassic Era. I call that progress.

References

[1] That statement is taken from William Benzon and David G. Hays, A Note on Why Natural Selection Leads to Complexity, Journal of Social and Biological Structures 13: 33-40, 1990, https://www.academia.edu/8488872/A_Note_on_Why_Natural_Selection_Leads_to_Complexity.

[2] See my post “What is Philosophy?” New Savanna, May 17, 2013, https://new-savanna.blogspot.com/2013/05/what-is-philosophy.html.

[3] “Brockman’s Third Culture and the emergence of a new philosophical regime,” New Savanna, May 4, 2021, https://new-savanna.blogspot.com/2021/05/brockmans-third-culture-and-emergence.html.

[4] I have used the label “philosophy new” to tag relevant posts at New Savanna. This is a link to those posts, https://new-savanna.blogspot.com/search/label/philosophy%20new.

[5] Allen Newell and Herbert A. Simon, Computer Science as Empirical Inquiry, Communications of the ACM, 19(3), p. 114.

[6] See his informal remarks, “The Normal Well-Tempered Mind,” Edge, May, 2013, where he also speculates about memes coming to tame neurons that have become “a little bit feral,” https://www.edge.org/conversation/daniel_c_dennett-the-normal-well-tempered-mind.

[7] Front. Artif. Intell., 19 April 2021 | https://doi.org/10.3389/frai.2021.625341.

[8] I tell that story in Touchstones • Strange Encounters • Strange Poems • the beginning of an intellectual life, November 2015, https://www.academia.edu/9814276/Touchstones_Strange_Encounters_Strange_Poems_the_beginning_of_an_intellectual_life.

Description and the Teleome, Part 2

To see why I'm bumping this seven year old post to the top of the queue, read its companion, linked in the first paragraph below.

* * * * *

In a previous post, Deep Learning, the Teleome, and Description, I argued the rich descriptions of art objects—literary texts, musical compositions and performances, paintings, and so forth—are reasonable proxies for what theoretical psychologist Mark Changizi has called the teleome:

the ultimate catalog of an animal’s what-it-does-es. The teleome is something along the lines of the set of all the capabilities our brains and bodies were selected to carry out. It is our set of powers, or the set of things we can do, or our function list.

I now want to say a bit more about why I think the careful study of art is central to arriving at an understanding of the human teleome.

Perception and Cognition in Context

I note first that Changizi has defined the teleome with respect to all animals and that, in his work he has considered a wide range of animal species. Thus, for example, in his work on binary vision (The Vision Revolution, chapter 2: “X-Ray Vision”), he surveyed information on eye position and orientation (side-facing or front-facing) in a wide range of species to reach his conclusion that binary vision isn’t about depth perception, it’s about seeing ‘through’ occlusions (grass, leaves) in front of the face. That is, he considered the visual system in the context of its use. It’s not merely that the eye-brain system evolved to see, but that it evolved to see in particular environments and so has become adapted to the visual affordances of those environments.

What then is the context for human artistic expression? Other people, that’s what. This is so obvious that it amounts to a truism. But in the context of 20th century academia, which, of course, is hanging on in this 21st century, that truism has a bit of a punch. In the case of literary criticism, my “home” discipline, the intellectual culture has been molded by an implicit core belief that criticism is an intimate “conversation” between the critic and the text in which the critic explores the recesses of his or her soul. The critic would then, of course, publish his or her reading and so enter into scholarly communion with other critics using literary texts to explore the recesses of their souls.

Now, few critics, living or dead, would actually subscribe to those words; most, I suspect, would object to them, and vigorously. After all, isn’t a lot of criticism about how texts form and mis-inform the minds of whole friggin’ societies? Yes, but, in my view, such moves are like the cycles upon cycles Ptolemaic astronomers had to employ to resolve the gap between their observations of planetary movement and their model, which put the earth at the center of the system. The center of academic literary criticism is the Cartesian subject, the lone mind in search of the external world and of other minds.

Given that starting point, the business of figuring out what the text means—a text, after all, that is external to the critic—is deeply problematic. It is from that difficulty that academic criticism developed a rich and elaborate array of conceptual epicycles.

Art is for Groups

When I set out to do a book on music, Beethoven’s Anvil, I simply tossed that Cartesian starting point to the winds. I didn’t even bother to mount much of an argument that against it. I simply declared that it’s not going to get us anywhere and dropped it.

I replaced it with the assertion that, when a group of people are making music, the sounds that they all perceive in common serve to couple their minds into a single dynamical system, one distributed over several physical bodies. At that time I also made the obvious generalization that all art is like that. But I did so only in my mind and my notes, not in the text of Beethoven’s Anvil. Why not go for the generalization in public? Because my specific argument about music was cut to fit the physical circumstances of music making. Any generalization would require work to refit the model to the different circumstances of, say, story telling, reading a book, looking at a painting, and so forth.

I still haven’t done that work nor, as far as I know, has anyone else. One issue, is that (films and the like excepted) visual arts are not arrayed in time. One, of course, views them in time, but the work itself doesn’t dictate the course of the viewing in the way that a musical performance does. Another problem is that, once a work has been inscribed in some more or less permanent medium (written text for verbal art, recordings for music, visual art, of course, is created in a more or less permanent medium) people can view it at different times and places. That being the case, how can we say that the work couples all those people into a single dynamical system? But, for the purposes of this post, I’ll overlook these issues.

You might, however, want to read the account of oral story telling I offer in, e.g. this post: Seven Sacred Words: An Open Letter to Steven Pinker.

Thus, as I’ve said above, the context of artistic expression is other people. The art work, whether it’s a musical performance, a poem, a story, a play, a painting, a drawing, a piece of sculpture, whatever, a work of art couples people’s minds and bodies into a single dynamical system. Given that, the question we must ask, then, is: What features must the work have in order to facilitate that coupling?

My interest in describing works of art follows directly from that question. I want to craft descriptions that set forth the various features, the affordances if you will, that allow the art work to be a medium for coupling between physically distributed minds. In the case of music the single most important feature, but certainly not the only one, is music—which I explicated at some length in Beethoven’s Anvil.

Deep Learning, the Teleome, and Description

From seven years ago, I'm bumping this to the top because the topic is on my mind these days: Why do the human sciences need rich, and objective, descriptions of aesthetic objects (poems, novels, paintings, musical performances, and so forth)? I sketch and answer to that in this post and in its companion, Description and the Teleome, Part 2.

* * * * *

Just the other day The New York Times’ John Markoff published an article on recent and apparently dramatic success in artificial intelligence: Scientists See Promise in Deep-Learning Programs. I skimmed the article—which said almost nothing about the techniques of this deep-learning—and wondered whether or not we’re being set-up for another fall. Markoff mentions the so-called “AI winter” of the mid-1980s that froze out entrepreneurial dreams of riches through practical AI when the practical proved impossible. He doesn’t mention that something similar happened to machine translation and computational linguistics in the early 1960s when Federal funding disappeared.

Stalking the Wild Teleome

So, with wonder in hand, I went over to Changizi and asked: “What do you think, boom or bust?” Mark pulled out his post: What Should We Unravel Next, After the Genome? Answer: The Teleome. He sets the stage:

Imagine that you find some mysterious device under your bed. What’s your next thought? It’s to wonder what the device does. Could it be a hand vacuum, a kid toy? …a bomb? Notice that your first thought does not concern how the device works. It’s premature to get to the “how it works” without having figured out the “what it does”. Obviously!

In much of the biological and brain sciences, however, there appears to be something of an inversion to this. In many scientific circles questions about “what it does” are deemed intrinsically unscientific or meaningless, and explanations in that domain are necessarily “just so” stories rather than science. Only questions concerning the biological mechanisms — i.e., concerning “how it works” — are truly kosher. And this attitude is reflected in funding priorities: ”how” funding dominates the “what it’s for” funding by a mile.

On the one hand, yes, sure, I agree. On the other hand, just a minute there. As you know I study literary texts, among other complex cultural objects such as movies, music, and graffiti. Not so long ago I found myself hip deep in nonsense about the adaptive purpose of art: it’s all about mating, no it’s about useful stories, no no no it’s about figuring out what to do with hyperactive intelligence, and so on. That is, they seem to be taking Changizi’s advice and are trying to figure out what it’s doing when the brain’s pumping out art and such. They’re looking for purpose, for what the design is supposed to accomplish.

On the other hand, though I have in fact given some thought to the question of just why, biologically, we make art (e.g. Emotion Recollected in Tranquility), I’ve been more compelled by the question How does it work? So I seem to be taking the wrong side of Changizi’s argument. Changizi argues we’re not going to figure out how it works until we know what it’s trying to do:

We need a “teleome,” the ultimate catalog of an animal’s what-it-does-es. The teleome is something along the lines of the set of all the capabilities our brains and bodies were selected to carry out. It is our set of powers, or the set of things we can do, or our function list.

Now, I rather doubt that Changizi would be enlightened if someone told him that, among the many powers of human brains and bodies, are the powers to make poems, music, and images. I mean, he already knows that, just as he knows that human brains and bodies fight wars, negotiate treaties, and even, on rare occasions, go boldly where no man has gone before. These proposals are chosen from the wrong conceptual universe; they don’t quite mesh with the conceptual materials of the cognitive, behavioral, and neurosciences.

A bit later in his post Changizi more or less addresses this issue: “it is not yet clear what [the teleome] would even look like. Surely it would not simply be a list. Instead, it would probably be a hierarchically tiered structure of some kind, with powers built out of sub-powers, and so on.”

Yep.

Hierarchically tiered structure.

Yep.

Powers built out of subpowers.

Check.

Describing, for example, a poem

So what’s this have to do with the humanities and poems and things like that?

Well, let me assert that, to a first approximation, writing a poem or telling a good story engages a wide range of human capabilities in an integrated fashion. Keith Oatley argues that literary texts are simulations, in the computer science sense of the term, of living life—and a half century ago Susan Langer (Feeling and Form) talked of the arts as providing virtual (her word) experience. Whatever the mind’s mechanisms are, a whole bunch of them are in use when making or comprehending art.

Why not computer in a can? Michael Nielsen on the varieties of material existence

I have a friend who is prone to thinking that the route from imagination to implementation isn't worth thinking about. I'd tease him with the idea of computer-in-a-can. It's an ordinary aerosol can. You spray computer-in-a-can on a convenient surface and voilà! you have a keyboard and monitor on that surface. You type whatever you need to type and the computer displays your result on the monitor. Just like that.

Michal Nielsen has written an interesting set of notes, The varieties of material existence (title swiped from William James). It's rather more serious and interesting than computer-in-a-can, but who knows? Some passages:

Using electrons, protons, and neutrons, it is possible to build: a waterfall; a superconductor; a living cell; a Bose-Einstein condensate; a conscious mind; a black hole; a tree; an iPhone; a Jupiter Brain; a working economy; a von Neumann replicator; an artificial general intellignece; a Drexlerian universal constructor (maybe); and much, much else. [...]

We usually think of all these things as separate phenomena, and we have separate bodies of knowledge for reasoning about each. Yet all are answers to the question “What can you build with electrons, protons, and neutrons?” [...]

What are the most interesting states of matter which have not yet been imagined? It’s remarkable that human consciousness, universal computing, superconductors, fractional quantum Hall systems (etc) are all pretty recent arrivals on planet Earth. Each is an amazing step, a qualitative change in what is possible with matter. What other states of matter are possible? What qualitatively new types of phenomena are possible, going beyond what we’ve yet conceived? Can we invent new states of matter as different from what came before as something like consciousness is from other states of matter? What states of matter are possible, in principle? In a sense, this is really a question about whether we can develop an overall theory of design? [...]

Much of my confusion is because the standard classification of matter into phases relies on that matter being at (or near) thermodynamic equilibrium. Parts of the human body are near thermodynamic equilibrium. But much is not. The thing that makes it all go, that makes life life – our metabolism – is all about energy flows that keep things away from equilibrium. [...]

I have two broad (and very different) frameworks for thinking about matter.

One of those frameworks is equilibrium statistical mechanics. This is the framework used by physicists to think about the different phases of matter, and (often) by chemists and materials scientists to think about what new materials are possible. It’s a powerful framework, and most stable matter in the world is of this type.

However, many of the most interesting systems – including universal computers, conscious minds, cells, economies, and others – don’t fit well into this framework. Rather, they have the three properties described above: many static components near thermodynamic equilibrium; many energy flows and dynamic components far from equilibrium; and surprising stability and resilience, often with built in self-healing or error-correction mechanisms.

Scienceandtechnology, or, Engineers Rule!

I'm thinking about method these days, and whatever method I've got is more like engineering than science, so I'm bumping this to the top of the queue.

There was a time when I thought of one entity: scienceandtechnology. This entity was all mathy and computery, and that differentiated it from, say, the humanities, which were not and are not at all mathy and computery.

In fact, I’m sure many people believe in this scienceandtechnology thingy.

So it came as something of a shock to realize that, no, scienceandtechnology is not one thing. It is two, science on the one hand, and technology on the other. Yes, they’re both mathy and computery, but they’re otherwise quite different.

And the difference is important, not simply for engineers and scientists, but for those of us in the human sciences who are trying to advance our knowledge of the human mind, human society, and human culture. Perhaps as a crude start, engineering is to rhetoric as science is to interpretation.

Science is about analyzing and describing to arrive at theories and models of how things work. The end result of a course of scientific work is an account of how some phenomenon can be explained within a given framework of laws and models. Such frameworks are likely to be elegant and compact. Newton, for example, had three laws of motion, not 57.

Engineering is quite different. Engineers use laws and models to analyze situations so that they can design a device to perform a certain task. The output of a course of engineering work is the description of that device and plans for its construction. To have any value those plans must specify something that can be constructed with known materials using known methods.

Thus when I was on the faculty at the Renssalaer Polytechnic Institute I learned that the engineering curriculum had a design stem, a series of courses required of all engineers devoted specifically to design. That is, it was not assumed that engineering graduates would somehow magically figure out how to design buildable stuff once they’d graduated and taken jobs in the “real” world. They were taught, given practice in, designing and building things.

Science isn’t like that. Scientists may design experiments, but that’s mostly a matter of logic, not of constructing something piece by piece by piece, and so forth, for 10s, 100s, or 1000s or more pieces. And yes, scientists may construct apparatus. To the extent they are doing that, they are acting as engineers. For that matter, engineers will make observations and conduct tests as part of their design work. And so they will, on occasion, act as scientists. But the overall objectives and methods, the envelope, if you will, of a scientific enterprise is different from that of an engineering enterprise.

Why does this matter to the human sciences? Because we deal with complex heterogeneous systems of many parts, systems where design and function are crucial. Thinking about such systems requires a mentality that, if anything, partakes more of engineering than of science. Perhaps that’s why Bruno Latour talks of compositionism, for composition implies design and construction. The FAQ at his website notes that he’s taught engineers for twenty years.

If we in the human sciences are going to ape our techno-savvy colleagues—not necessarily a good thing, but not necessarily bad either—perhaps we should pay more attention to engineers than scientists. Their problems are more like the problems of writers, musicians, artists, or even politicians and bureaucrats. They have more to teach us than do scientists. We don't need humanists who secretly wish they were scientists. We need humanists who openly aspire to engineering.

ADDENDUM: See this recent column by Mark Changizi.

Description, Interpretation, and Explanation in the Case of Obama’s Eulogy for Clementa Pinckney

I want to continue the discussion in my previous two posts, Yet Another Brief for Description (and Form), and, Why Ethical Criticism? or: The Fate of Interpretation in an Age of Computation. I want to take a quick and dirty look at description, interpretation, and explanation with respect to Obama’s eulogy for Clementa Pinckney as I discussed it in “Form, Event, and Text in an Age of Computation” [1]. In that (draft) article I first present a (toy) model of the computational analysis of a literary text (Shakespeare’s Sonnet 129) and then discuss form, arguing for a computational conception. Then I take a look at Obama’s performance (which is readily available in video form) followed by a demonstration that the text is a ring-composition [2].

The ring-composition analysis is fundamentally descriptive. To be sure, I have to do some low-level interpretation to divide the text into sections. For example, I assert that in paragraph 17 the topic shifts from Rev. Pinckney and his relations to the nation and the black church’s role in it. How do I know that? Because that’s the first paragraph where the word nation occurs; then paragraph 18 talks of the role of the black church in the Civil Rights Movement. This sort of thing seems obvious enough.

But what’s there to explain? I can imagine that, in principle, the kind of computational model I created for the Shakespeare sonnet could be created for the eulogy. That might help us to explain just how the eulogy works in the mind. But I can’t see creating such a model now; we just don’t know how. That is to say, the eulogy’s ring-form design is something that needs to be explained by an underlying psychological model. And computation will be an aspect of the model.

What I don’t do is invoke the special terminology of some interpretive system: deconstruction, Lacanian analysis, Foucaultian genealogy, and so forth. I don’t offer a “reading” of the eulogy. I don’t, for example, offer remarks about the underlying theology, which seems to invoke the notion of the Fortunate Fall in its central presentation of the mysterious was of God’s grace.

* * * * *

Nor do I address an observation Glenn Loury made in conversation John McWhorter [3]. Loury is remarking on the fact that Obama took on the role of a black preacher and drew on the tropes and stylistic moves of black vernacular preading. They are remarking that, of course, this was a performance. But not an inauthentic one, though Obama was not himself raised in the black church. Loury says:

A mask, a face has to be made. A way of being has to be fashioned. It’s gotta’ be practiced. You could see him standing in front of the mirror. John, we should write the novel John. […]

It just resonates in my mind so deeply. Because what does it mean for a people, I speak now of black Americans 30-40 million, to have the embodiment of their generational hopes, personified by a person who must adopt artifice, and manufacture, in order to present himself as being of them. What does it say of such a people.

No no no. I think this is historic profound. Excuse me if I, you know, I mean I’m just saying, here we are. Because think about it, think about it, OK, the stigma of race, slavery, OK, Orlando Patterson just brilliantly analyzes this, I think. Slavery has to be, you’re putting the slave down. The slave must be a dishonored person. OK so honor, honor becomes central to the whole quest for equality.

And having the Chief Executive of State, be of you, or at the very least, be a person who when in a position of choice, chose to be of you, is countering the dishonor in a very deep way. But perhaps the only way that the state’s symbolic power could be married to your quest for honor is through the President of someone who wasn’t quite fully of you. Your stigma still resonates even in the workings of history, that are intended to elevate you.

Those remarks are certainly worth elaboration, and that elaboration will necessarily be interpretive in the fullest sense of the word.

Roads Not Taken: A Study in Poetic Mechanism

Another working paper posted; links, abstract, and introduction as usual.

• Academia.edu: https://www.academia.edu/17120360/Roads_Not_Taken_A_Study_in_Poetic_Mechanism
• SSRN: http://ssrn.com/abstract=2677363

* * * * *

Abstract: Robert Frost’s “The Road Not Taken” is a ring composition three levels deep: 1 2 Ω 2’ 1’. The central section consists of lines 9 through 12 cuts across the boundary between the second (ll. 6-10) and third (ll. 11-14) stanzas. There is a subtle shift in tense in line 16 in which the poet in effect travels back into the past, at the moment of decision captured in the poem, so that he can anticipate the present moment in which the poem unfolds. Thus the end of the poem rejoins the beginning, not merely through the repetition of a line, but through a trick in time.

CONTENTS

Introduction: Another Ring Discovered 2

Preliminaries: Describing Form, and a Precedent 2

A Road Literary Critics Don’t Travel 3
A Parallel in “Kubla Khan” 3

Frost’s Text: The Road Not Taken 5

Robert Frost, Time Traveler: The Road Not Taken 5

Poetic Mechanism 6
But What About Meaning? 10

Further Thoughts 11

Alignment in “The Road Not Taken” 11
More Frostiness 14
More About the Poem 16

Introduction: Another Ring Discovered

Robert Frost’s “The Road Not Taken” is one of the best-known poems in the English language and a secondary school favorite in America. It would be a bit much to assert that every schoolchild has read the poem, but many have, for they’ve had no choice. Many critics written about the poem as well and there seems to be widespread agreement that the poem is a bit deceptive, as poems are wont to be.

Still, when, prompted by a post in 3 Quarks Daily, I set out to read it again, long after my schoolboy years, I wasn’t expecting to discover something in the poem that, apparently, other critics have missed. Yes, I know I know, the great texts yield endless riches. But what I discovered was something that was, to me anyhow, obvious, something about the poem’s form. Yes, four stanzas of five lines, rhymed ABAAC – that too is obvious. We learn how to do that kind of description in secondary school.

But that’s about it as far as formal description goes. After that the search for meaning takes over and never lets up. And so most of the discussion of this poem, as of others, is about meaning, and its formal features are either ignored or treated as ornamentation.

What was obvious about this poem, almost as soon as I’d reread it, is that it is a ring-composition, which I explain in detail a bit later. But I shouldn’t have had to discover that. Just as it is commonly known that dogs have four legs and a tail, so I should be commonly known that “The Road Not Taken” is a ring-composition. Once you see that, then you can see how the poem has three structural principles operating in parallel, I write about that as well.

Such a simple poem, such a rich text. So much that’s not been observed.

When WILL we open our eyes and ears?

Cleaning Coal: An Informal Study of Ecological Design

Slightly revised from My Father Cleaned Coal for a Living, originally published for the Truth and Traditions Party.

My father was trained as a chemical engineer. He spent his entire career with Bethlehem Mines, the mining division of the now defunct Bethlehem Steel Company. He designed coal cleaning plants, at least the system for actually cleaning the coal.

This is a story about the last plant that he designed, one designed to keep the air clean while at the same time reducing the cost of running and maintaining the plant. It was a beautiful and elegant solution to a nasty design problem. A study in systems thinking – though my father probably never used that phrase.

Cleaning Coal

Before coal can be turned into coke (for subsequent use as a fuel in steel-making) it must be cleaned of impurities, mostly sulfur. Most cleaning techniques take advantage of the fact that the rocks containing the impurities are denser than coal. So, you crush the raw coal until all the particles are less than, say, an eighth of an inch along any dimension. Then you float the crushed coal in some medium – generally, but not always, water – and take advantage of the fact that the rock sinks faster than the coal. There are several techniques you can use to do this, as I recall, but whichever technique you use, you end up with wet coal when you’re done.

Wet coal is considerably heavier than dry coal. As railroads charge by the pound, it costs more to ship wet coal than dry. Further, in the winter a hopper car filled with wet coal at the mine – where coal is generally cleaned, there's no point in shipping useless rock to the steel plant – is likely to be filled with frozen coal when you get to the plant. How do you empty that mess from the cars?

So, you need to dry the coal.

The old drying technique – drying ovens – leaves you with a lot of coal dust in the air. A lot. And coal dust is nasty stuff. You don't want it spewing out of chimneys anywhere in your neighborhood. Or near your farm.

Epic FAIL

So much for this "striking symbol of hope and prosperity..."

The original design obviously included specifications for an elevator big enough for a 20-storey building. In the process of scaling things up, however, nobody thought to redesign the elevator system—and, naturally, a 47-storey building requires more space for its lifts and motor equipment. Sadly, that space doesn't exist.

Instrument Matter in the Musician’s Mind: Part 1, Loosen Up

The “sex appeal” of the inorganic, like life, is another way to give voice to what I think of as a shimmering, potentially violent vitality intrinsic to matter.
—Jane Bennett, Vibrant Matter

We all know that B.B. King’s guitar is named Lucille. Why? No, not ‘why “Lucille”’? Why is it named at all?

Perhaps it’s a gesture of affection. The guitar, after all, is very close to him. It’s one of his voices, it is, in some sense, part of him.

It may be more than that. The name may well reflect the subtle intricacy of King’s relationship to his guitar, his instrument. To play an instrument well, one must learn to yield to its physicality, to blend with it. You can’t dominate it. Well, you can try, and you CAN succeed. But you pay a cost. You musicianship suffers.

As I’m not a guitar player, however, I can’t tell you what it means to yield to a guitar. Of sure, I can guess, I can make up stories, and you might find those stories convincing. If you’re not a guitar player. But guitar players, the thoughtful ones at least, will know that I’m faking it.

I suppose I could talk about the trumpet—I’ve been playing for half a century—but that’s just a little complex. And my point really isn’t about complexity. It’s about subtlety.

The Claves

So let’s talk about the claves. The claves are a pair of short sticks that tend to be roughly eight inches long and an inch in diameter. They’re used in Latin music, indeed, they’re central to many genres, to produce a sharp penetrating percussive sound. They’re usually made of hard dense wood. Mine are made of fiberglass:

You hold one clave in your left hand and then strike it with the other one, held in your right hand (if you’re right handed). Simple, no? Well, yes. And no.

It’s more like you cradle the one clave (it doesn’t matter which one) in your left hand. You hold your hand palm-up, lay the clave across it, and grip it only so much as needed to keep it in place. You don’t need to grip it tightly, nor do you even WANT to grip it tightly. If you do that, then your hand will dampen the vibrations and dull the sound. The ‘crack!’ will no longer be sharp and crisp.

Speculative Engineering

Continuing with the design theme, here’s a passage from the introduction to Beethoven’s Anvil (pp. xii – xiii) where I talk about my intellectual method as one that is as much about engineering and design as it is about analysis and science:

This book is thus about building blocks. When I was a child I had an extensive pile of wooden building blocks: various sizes of squares and rectangles, with the rectangles coming in several different length-to-width ratios, round rods that could serve a columns or as logs, some triangles, some arches, various relatively flat pieces, and so it. It was a miscellaneous collection, building blocks from various sets, but also odds and scraps of wood my father gave me or that I found here and there.

I loved building things with these blocks. I particularly remember building gasoline stations and ocean-going freighters. I surely must have build forts and castles, and then rockets and space ports. I certainly spent time building the tallest possible tower. In some cases the challenge was primarily imaginative: How do I make something like this? In other cases, there was surely an engineering challenge, e.g. just what is the best way to create that tall tower? No matter what I made, I used the same set of blocks.

Beethoven’s Anvil is about the building blocks and design principles, not so much of music, but of the brains and societies that create that music. My argument is simply that these building blocks—mostly neural circuits and social structures—are necessary. I have no particular reason to believe that I’ve defined the complete set; in fact, I have some some small reason to believe that the set is not yet complete, nor ever will be. The nervous system is plastic, taking the impress of its environment. As culture molds the human environment, so it molds the nervous system. I see no end to culture’s possibilities. Thus, it seems to me quite possible that our descendents a century or two from now will have nervous systems that differ from ours in small but critical ways.

Thus I like to think of this book as an exercise in speculative engineering. Engineering is about design and construction: How does the nervous system design and construct music? It is speculative because it must be. The purpose of speculation is to clarify thought. If the speculation itself is clear and well-founded, it will achieve its end even when it is wrong, and many of my speculations must surely be wrong. If I then ask you to consider them, not knowing how to separate the prescient speculations from the mistaken ones, it is because I am confident that we have the means to sort these matters out empirically. My aim is to produce ideas interesting, significant, and clear enough to justify the hard work of investigation, both through empirical studies and through computer simulation.

Designing Minds

 Tim Morton’s been blogging up a storm about design issues, mostly, I take it, about architectural design and the environment (this link takes you to one of a blizzard of posts in the neighborhood). But architecture is hardly the only locus of design. The human world is filled with designed objects and processes. Some quite concrete, but some rather abstract.

I’m thinking in particular about computer software, which has spread its tendrils throughout the civilized and semi-civilized and hanging-on-the-edge worlds in the past half-century. The thing about software is that we don’t understand it very well. Hence it’s ‘buggy.’ Some bugs are merely annoying, but some bugs bring the system crashing down in ways that destroys hours, days, or weeks or more of work.

It’s commonplace in the software world to think that, in designing software for some particular application, you must design to the user’s view of the world. How does the user think and act? What’s the user want and desire? At this level design is a species of cognitive anthropology. But how many software engineers are trained in ethnology? What do they know about human perception and cognition – which, at best, are poorly understood.

The other day I was talking to a very experienced and very senior software designer, database architect to be more exact, who gave me a typical example. Consider a package shipping firm, such as Federal Express, UPS, or DHL. Who’s the firm’s customer? Well, to the route driver, the customer is the person or organization who receives shipment of a package and who may also turn over a package for delivery. To the clerk in a package shipping store the customer is the person who delivers a package for shipping. To the account executive the customer is the person or organization who pays for shipping, and that person may not be the person who receives the package or who turns it over to the company for delivery.

So we’ve got at least three different definitions of customer. Now, you might say that that’s a matter of mere definition. And, in a sense you’re right. But getting people to agree on ‘mere’ definitions is not easy, especially when they’re being asked to accede to a definition that doesn’t accord with their view of the world, however limited it may be. The software world is filled with such problems of mere definition.

Thus, it was not at all surprising when this architect starting talking about how one might take a Kantian view of the problem, in which one attempts to design the software to a given user’s view of it. Or, he went on, one might opt for a Platonic view, in which one attempts to design the software so as to reflect the world as it is, in essence. That’s how he thinks about the problems he faces in being a database architect. But, alas, he tells me, there aren’t many folks in the computer biz who can follow him when he talks in such terms, Kant vs. Plato. (I wonder what Nietzschean software would be like?)

And yet, at this point, we seem irrevocably committed to living in a world pervaded by software. By this stuff we don’t understand, can’t make very well, is buggy. And breaks down. Over and over and over.

Cultural Evolution 7: Where Are We At?

I’ve got three more posts planned before I write the post that’ll go up at the Forum of the “On the Human” project of National Humanities Center. But this isn’t one of those three. Rather, I want to step back and take a look around. That’s first. Then I’m going to describe what I’m planning for the last three posts. And I’ll conclude with a note on design.

What’s Up? Two cultures, again

The general website for “On the Human” describes the project thus:

On the Human (OTH) is an online community of humanists and scientists dedicated to improving our understanding of persons and the quasi-persons who surround us. As persons are biological, psychological, historical, moral, and autobiographical beings, we employ modes of inquiry from the sciences and humanities. Contributors explore issues in metaphysics and biology, ethics and neuroscience, experimental philosophy and evolutionary psychology.

Thus it is one of those “hands across the two cultures divide” enterprises, of which I tend to be skeptical despite the fact that I’ve been crossing that pseudo-divide my whole career. But enough of my skepticism.

Let’s take the divide at face value. In those terms, what’ve we got?

Scientists and cultural evolution

We have a great deal of work on human cultural evolution over the past two decades or so, and most of it has been done by people who are trained as or think of themselves as scientists. For the most part these thinkers have cultural evolution without reference to 1) human psychology, including perceptual and cognitive, and 2) without out detailed descriptions of cultural objects and artifacts. On one, Colin Martindale is an exception. On two, linguistics is an exception.

Thus studied, human cultural evolution is a bit like the study of biological evolution without molecular biology and without plant and animal physiology and anatomy. Without those things, how could you study biological evolution at all? Where would Darwin have been if he didn’t have three or four centuries worth of natural history on which to build his thinking? How can one understand comparative morphology in ecological context if you don’t have detailed accounts of morphologies and lifeways? And yet that’s how the study of cultural evolution is proceeding.

Well, it’s not that bad. But as a first approximation, that will do.

As I’ve said, Martindale is an exception. His account of cultural evolution, unlike other accounts, is grounded in perceptual, cognitive, and affective psychology. His empirical work, of course, rests on characterizations of cultural artifacts – poems, musical compositions, tombstones, and the like – though they are not detailed descriptions comparable to standard descriptive work on flora and fauna. For what he wants to do, such detailed descriptions aren’t necessary.

By contrast, memetics, especially at its popularizing extremes, seems like an attempt to replace psychology entirely. Just dump the entire set of disciplines and think deep thought about the mind by talking of memes.

This won’t do, not at all. My own thinking makes perceptual and cognitive psychology essential (see Cultural Evolution 3: Performances and Memes). Memes, as I have defined them, are perceptual entities and so one must call on perceptual and cognitive psychology to understand them. Similarly, the cultural analog to a phenotype is a performance, understand as the mental events through which one apprehends cultural products. The trick is to understand these things as collective phenomena, not just individual ones.

Correlatively, one also needs detailed descriptive control of cultural artifacts. My two posts on Rhythm Changes give an indication of what’s entailed. Studies of language evolution entail that level of descriptive detail, but most work doesn’t seem to be aware of that level of description – which is, by the way, only a crude indicator of whats necessary and possible.

Thus, the contemporary study of cultural evolution, on the whole, does not seem like it arose as an attempt to solve problems that have arisen through close analysis and description of human culture. Rather, it is something that has been imported from biology and applied to human culture in ways that, for the most part, don’t require detailed understanding of cultural morphologies and processes. In some cases, the thinking is mostly theoretical speculation. While I am in no position to object to speculation on point of principle, I am a bit skeptical about the possible success of a speculative enterprise that betrays so little interest in the properties of the objects and processes which are the objects of speculation. Where would biology have gotten if those naturalists had been indifferent as to whether or not the creature had three legs or four, or, indeed, whether or not it had any legs at all?

Humanists on Cultural Evolution

That, in crude caricature, is the scientific side of the ledger. What about the humanists?

Links: Origins of Language and the Problem of Design

So, I get an email informing me of a link post at John Wilkins' Evolving Thoughts. I go there where I click through to James Winters's A Replicated Typo, which has a post, Answering Wallace's challenge: Relaxed Selection and Language Evolution. I start reading and find out that it's mostly about a recent article by Terrence Deacon in the Proceedings of the National Academy of Sciences. So I click through to the paper, A role for relaxed selection in the evolution of language capacity. And, naturally, I start reading that, not too much, just a little to get the flavor.

Then I start poking around in the immediate neighborhood, as Winters had indicated that Pinker, too, had a recent paper on language evolution. As indeed he does: The cognitive niche: Coevolution of intelligence, sociality, and language. So I read a little of that and did some more poking.

And what do I find? PNAS has a whole pile of articles on human evolution in that issue, and you can read them all online. For free, though, if you wish, you can pony up $25 and spend a week downloading PDFs to your heart's content. I might just do that. Douglas Wallace has a paper, Bioenergetics, the origins of complexity, and the ascent of man, which looks particularly interesting. It's the energetics and complexty stuff that has my attention. Here's the abstract:

Complex structures are generated and maintained through energy flux. Structures embody information, and biological information is stored in nucleic acids. The progressive increase in biological complexity over geologic time is thus the consequence of the information-generating power of energy flow plus the information-accumulating capacity of DNA, winnowed by natural selection. Consequently, the most important component of the biological environment is energy flow: the availability of calories and their use for growth, survival, and reproduction. Animals can exploit and adapt to available energy resources at three levels. They can evolve different anatomical forms through nuclear DNA (nDNA) mutations permitting exploitation of alternative energy reservoirs, resulting in new species. They can evolve modified bioenergetic physiologies within a species, primarily through the high mutation rate of mitochondrial DNA (mtDNA)–encoded bioenergetic genes, permitting adjustment to regional energetic environments. They can alter the epigenomic regulation of the thousands of dispersed bioenergetic genes via mitochondrially generated high-energy intermediates permitting individual accommodation to short-term environmental energetic fluctuations. Because medicine pertains to a single species, Homo sapiens, functional human variation often involves sequence changes in bioenergetic genes, most commonly mtDNA mutations, plus changes in the expression of bioenergetic genes mediated by the epigenome. Consequently, common nDNA polymorphisms in anatomical genes may represent only a fraction of the genetic variation associated with the common “complex” diseases, and the ascent of man has been the product of 3.5 billion years of information generation by energy flow, accumulated and preserved in DNA and edited by natural selection.

That third sentence is the one that got me: "The progressive increase in biological complexity over geologic time is thus the consequence of the information-generating power of energy flow plus the information-accumulating capacity of DNA, winnowed by natural selection." Sweet. What's especially sweet is the casual assertion of "biological complexity over geologic time." That's certainly how I see it, but I was under the impression that biologists bridle at the notion of increasing complexity over time. Has that notion become more acceptible while I was looking the other way? I surely hope so. In any event, David Hays and I wrote a little article on the subject, A Note on Why Natural Selection Leads to Complexity, which you may download here.

But let's get back Winters' post at A Replicated Typo. Here's a passage that get me thinking a bit:

The first aspect we need to appreciate is how Darwinian-like processes operate at the developmental-level. Deacon cites many instances, such as the fine-tuning of axonal connection patterns in the developing nervous system, where developmental processes are achieved through selection-like operations. Importantly, though, the logic differs from natural selection in one respect: “selection of this sort is confined to differential preservation only, not differential reproduction. In this respect, it is like one generation of the operation of natural selection”.The point he’s trying to get across is that these intraselection processes are taking place right across nature.

That fine-tuning of axonal connections, that, of course, has been known for some time, and Gerald Edelman has built his own theories of the brain around such phenomena, talking of neural Darwinism. What happens is this: At some relatively early stage in ontogenetic development a population of neurons sprouts dendrites like mad, forming random connections all over the place. These connections are then "pruned back" through use. Those connections that are used become stronger; those that are not, weaken and the dendrites die away. In this way the pattern of neural connectivity is "sculpted" to "match" the affordances (to use J.J. Gibson's term) offered by the environment.

It's the problem of design all over again. The brain's got billions of neurons, each of which has thousands of connections. All of these must be wired up just right. How's the brain to do this? That is to say, how do you cram all the necessary wiring information into the genome? You don't, because you don't have to. You just set up a process of random variation and selective retention, one that operates inside the organism.

One must keep in mind, of course, that those neurons that constitute the brain are each living entities and, as such, are trying to survive and thrive in their environment, which is filled with other neurons (not to mention the glial cells which surround them). The neurons don't "know" that they're "strapped" together in an organism such that they ALL survive or die according to the fate of that organism. And, there was a time when all life on earth consisted of single-celled organisms, each trying to survive, and variously cooperating and competing with their fellows.

And I've recently been making the same argument with respect to culture, especially in my recent post on design. What I haven't talked about, and what may prove a tricky little business, is where we get random variation among memes. Recall that I've defined memes as physical properties of cultural artifacts and processes, namely those properties that allow individuals to cooperate through those artifacts and processes. Certainly there will be variations in the way those memetic properties are expressed in individual performances. Indeed, there may be downright errors.

That's certainly the case in improvisation. You intend the line to go that way, but it doesn't. Maybe a finger slipped, causing a wrong note to come out, or one of the other musicians did something that changed the "valence" of your intended line. Whatever. Things got off track. So you've got to scramble to get them back on track. If you're successful, and your invention is particularly felicitous, you'll repeat it, and others will copy it, and before you know it, another meme is born.

But enough of this. I'm rambling. More later.

Cultural Evolution 6: The Problem of Design

To this point I have been taking it as obvious that a theory of cultural evolution would be a good thing to have. The only thing at issue is just what that theory would be like. Now, let’s step back for a minute and ask: Do we really need a theory of cultural evolution? Historians have been doing fine without such a theory, so why go to the trouble of creating one just because we can?

That is to say, what’s a theory of cultural evolution supposed to do? Saying that, well, it’s supposed to explain how culture evolves is no answer, because it simply assumes that culture does evolve. That’s what I want to question, if only rhetorically.

In comparison, when Darwin elaborated his account of biological evolution, he was trying to solve a particular problem. Living things appeared to be designed. Unless we’re going to posit the existence of a Divine Designer, how can we account for that appearance?

So, specifically, what are we trying to account for with a theory of cultural evolution? As near as I can tell, the gene-culture coevolution folks are trying to account for the rate of cultural change in human history. It’s too fast for biological inheritance mechanisms, so the mechanism must be cultural. It’s not at all clear to just what, specifically, Dawkins was up to. He wasn’t happy with other accounts of human culture, but he seems to have mostly been interested in memes as an example of another kind of replicator (a term he coined I believe). He wasn’t trying to solve any particular problem about human culture.

And for the most part, much of my own thinking about cultural evolution has proceeded without any specific problem in view. But I’m no longer willing to proceed with this enterprise simply because it is intrinsically interesting. We need a fairly specific problem that needs solved. What problem could that be?

I note that I’ve worked rather hard to produce an account of human cultural evolution that meets two criteria:

1. it’s consistent with what we know about human psychology and neuropsychology and with some body of thinking about at least one major aspect of human culture (music), and
2. it has the same logical and causal form as biological evolution, selection on phenotype elements and variation among genotype elements.

Since, in biology, the purpose of that logical and causal form is to account for design, I propose that it play the same role in the study of culture: to account for the design of human culture.

The Paradox of Human Design

The most obvious objection to this proposal is one that John Lawler raised to the third post in this series and that others have raised as well: “. . . whatever else may be true about it, any real evolution in language (and this goes pretty much for any other cultural phenomena as well), is that evolution of language (and culture) does not follow Mendelian rules, but rather Lamarckian.” Everything about human culture is designed, by human beings, not by some Divine Designer. So the idea that we need an account of human cultural evolution to account for culture’s design, that would appear to be a non-starter.

Not so fast. Let me repeat the response I made to Lawler: