The Chinese Room Argument
Infants don’t learn language like neural nets do. Unlike nets, no feedback is involved, which amazingly, makes learning faster.
As is typical of research in psychology, the hard part is thinking of something clever to do, rather than actually carrying it out.
[ Proc. Natl. Acad. Sci. vol. 117 pp. 26548 – 26549 ’20 ] is a short interview with psychologist Richard N. Aslin. Here’s a link — hopefully not behind a paywall — https://www.pnas.org/content/pnas/117/43/26548.full.pdf.
He was interested in how babies pull out words from a stream of speech.
He took a commonsense argument and ran with it.
“The learning that I studied as an undergrad was reinforcement learning—that is, you’re getting a reward for responding to certain kinds of input—but it seemed that that kind of learning, in language acquisition, didn’t make any sense. The mother is not saying, “listen to this word…no, that’s the wrong word, listen to this word,” and giving them feedback. It’s all done just by being exposed to the language without any obvious reward”
So they performed an experiment whose results surprised them. They made a ‘language’ of speech sounds which weren’t words and presented them 4 per second for a few minutes, to 8 month old infants. There was an underlying statistical structure, as certain sounds were more likely to follow another one, others were less likely. That’s it. No training. No feedback. No nothin’, just a sequence of sounds. Then they presented sequences (from the same library of sounds) which the baby hadn’t heard before and the baby recognized them as different. The interview didn’t say how they knew the baby was recognizing them, but my guess is that they used the mismatch negativity brain potential which automatically arises to novel stimuli.
Had you ever heard of this? I hadn’t but the references to the author’s papers go back to 1996 ! Time for someone to replicate this work.
So our brains have an innate ability to measure statistical probability of distinct events occurring. Even better we react to the unexpected event. This may be the ‘language facility’ Chomsky was talking about half a century ago. Perhaps this innate ability is the origin of music, the most abstract of the arts.
How infants learn language is likely inherently fascinating to many, not just neurologists.
Here are links to some other posts on the subject you might be interested in.
https://luysii.wordpress.com/2013/06/03/how-infants-learn-language-iv/
https://luysii.wordpress.com/2011/10/10/how-infants-learn-language-iii/
https://luysii.wordpress.com/2010/10/03/how-infants-learn-language-ii/
https://luysii.wordpress.com/2010/09/30/how-infants-learn-language/
Phil Anderson probably never heard of Ludwig Mies Van Der Rohe, he of the Bauhaus and his famous dictum ‘less is more’, so he probably wasn’t riffing on it when he wrote “More Is Different” in August of 1970 [ Science vol. 177 pp. 393 – 396 ’72 ] — https://science.sciencemag.org/content/sci/177/4047/393.full.pdf.
I was just finishing residency and found it a very unusual paper for Science Magazine. His Nobel was 5 years away, but Anderson was of sufficient stature that Science published it. The article was a nonphilosophical attack on reductionism with lots of hard examples from solid state physics. It is definitely worth reading, if the link will let you. The philosophic repercussions are still with us.
He notes that most scientists are reductionists. He puts it this way ” The workings of our minds and bodies and of all the matter animate and inanimate of which we have any detailed knowledge, are assumed to be controlled by the same set of fundamental laws, which except under extreme conditions we feel we know pretty well.”
So many body physics/solid state physics obeys the laws of particle physics, chemistry obeys the laws of many body physics, molecular biology obeys the laws of chemistry, and onward and upward to psychology and the social sciences.
What he attacks is what appears to be a logical correlate of this, namely that understanding the fundamental laws allows you to derive from them the structure of the universe in which we live (including ourselves). Chemistry really doesn’t predict molecular biology, and cellular molecular biology doesn’t really predict the existence of multicellular organisms. This is because new phenomena arise at each level of increasing complexity, for which laws (e.g. regularities) appear which don’t have an explanation by reducing them the next fundamental level below.
Even though the last 48 years of molecular biology, biophysics have shown us a lot of new phenomena, they really weren’t predictable. So they are a triumph of reductionism, and yet —
As soon as you get into biology you become impaled on the horns of the Cartesian dualism of flesh vs. spirit. As soon as you ask what something is ‘for’ you realize that reductionism can’t help. As an example I’ll repost an old one in which reductionism tells you exactly how something happens, but is absolutely silent on what that something is ‘for’
“Everything in chemistry turns blue or explodes”, so sayeth a philosophy major roommate years ago. Chemists are used to being crapped on, because it starts so early and never lets up. However, knowing a lot of organic chemistry and molecular biology allows you to see very clearly one answer to a serious philosophical question — when and where does scientific reductionism fail?
Early on, physicists said that quantum mechanics explains all of chemistry. Well it does explain why atoms have orbitals, and it does give a few hints as to the nature of the chemical bond between simple atoms, but no one can solve the equations exactly for systems of chemical interest. Approximate the solution, yes, but this is hardly a pure reduction of chemistry to physics. So we’ve failed to reduce chemistry to physics because the equations of quantum mechanics are so hard to solve, but this is hardly a failure of reductionism.
The last post “The death of the synonymous codon – II” — https://luysii.wordpress.com/2011/05/09/the-death-of-the-synonymous-codon-ii/ –puts you exactly at the nidus of the failure of chemical reductionism to bag the biggest prey of all, an understanding of the living cell and with it of life itself. We know the chemistry of nucleotides, Watson-Crick base pairing, and enzyme kinetics quite well. We understand why less transfer RNA for a particular codon would mean slower protein synthesis. Chemists understand what a protein conformation is, although we can’t predict it 100% of the time from the amino acid sequence. So we do understand exactly why the same amino acid sequence using different codons would result in slower synthesis of gamma actin than beta actin, and why the slower synthesis would allow a more leisurely exploration of conformational space allowing gamma actin to find a conformation which would be modified by linking it to another protein (ubiquitin) leading to its destruction. Not bad. Not bad at all.
Now ask yourself, why the cell would want to have less gamma actin around than beta actin. There is no conceivable explanation for this in terms of chemistry. A better understanding of protein structure won’t give it to you. Certainly, beta and gamma actin differ slightly in amino acid sequence (4/375) so their structure won’t be exactly the same. Studying this till the cows come home won’t answer the question, as it’s on an entirely different level than chemistry.
Cellular and organismal molecular biology is full of questions like that, but gamma and beta actin are the closest chemists have come to explaining the disparity in the abundance of two closely related proteins on a purely chemical basis.
So there you have it. Physicality has gone as far as it can go in explaining the mechanism of the effect, but has nothing to say whatsoever about why the effect is present. It’s the Cartesian dualism between physicality and the realm of ideas, and you’ve just seen the junction between the two live and in color, happening right now in just about every cell inside you. So the effect is not some trivial toy model someone made up.
Whether philosophers have the intellectual cojones to master all this chemistry and molecular biology is unclear. Probably no one has tried (please correct me if I’m wrong). They are certainly capable of mounting intellectual effort — they write book after book about Godel’s proof and the mathematical logic behind it. My guess is that they are attracted to such things because logic and math are so definitive, general and nonparticular.
Chemistry and molecular biology aren’t general this way. We study a very arbitrary collection of molecules, which must simply be learned and dealt with. Amino acids are of one chirality. The alpha helix turns one way and not the other. Our bodies use 20 particular amino acids not any of the zillions of possible amino acids chemists can make. This sort of thing may turn off the philosophical mind which has a taste for the abstract and general (at least my roommates majoring in it were this way).
If you’re interested in how far reductionism can take us have a look at http://wavefunction.fieldofscience.com/2011/04/dirac-bernstein-weinberg-and.html
Were my two philosopher roommates still alive, they might come up with something like “That’s how it works in practice, but how does it work in theory? “
One of the problems with being over 80 is that you watch your friends get sick. In the past month, one classmate developed ALS and another has cardiac amyloidosis complete with implantable defibrillator. The 40 year old daughter of a friend who we watched since infancy has serious breast cancer and is undergoing surgery radiation and chemo. While I don’t have survivor’s guilt (yet), it isn’t fun.
Reading and thinking about molecular biology has been a form of psychotherapy for me (for why, see the reprint of an old post on this point at the end).
Consider ALS (Amyotrophic Lateral Sclerosis, Lou Gehrig disease). What needs explaining is not why my classmate got it, but why we all don’t have it. As you know human neurons don’t replace themselves (forget the work in animals — it doesn’t apply to us). Just think what the neurons which die in ALS have to do. They have to send a single axon several feet (not nanoMeters, microMeters, milliMeters — but the better part of a meter) from their cell bodies in the spinal cord to the muscle the innervate (which could be in your foot).
Supplying the end of the axon with proteins and other molecules by simple diffusion would never work. So molecular highways (called microtubules) inside the axon are constructed, along which trucks (molecular motors such as kinesin and dynein) drag cargos of proteins, and mRNAs to make more proteins.
We know a lot about microtubules, and Cell vol. 179 pp. 909 – 922 ’19 gives incredible detail about them (even better with lots of great pictures). Start with the basic building block — the tubulin heterodimer — about 40 Angstroms wide and 80 Angstroms high. The repeating unit of the microtubule is 960 Angstroms long, so 12 heterodimers are lined up end to end in each repeating unit — this is the protofilament of the microtubule, and our microtubules have 13 of them, so that’s 156 heterodimers per microtubule repeat length which is 960 Angstroms or 96 nanoMeters (96 billionths of a meter). So a microtubule (or a bunch of microtubules extending a meter has 10^7 such repeats or about 1 billion heterodimers. But the axon of a motor neuron has a bunch of microtubules in it (between 10 and 100), so the motor neuron firing to the muscle moving my finger has probably made billions and billions of heterodimers. Moreover it’s been doing this for 80 plus years.
This is why, what needs explaining is not ALS, but why we don’t all have it.
Here’s the old post
Neurology is fascinating because it deals with illnesses affecting what makes us human. Unfortunately for nearly all of my medical career in neurology ’62 – ’00 neurologic therapy was lousy and death was no stranger. In a coverage group with 4 other neurologists taking weekend call (we covered our own practices during the week) about 1/4 of the patients seen on call weekend #1 had died by on call weekend #2 five weeks later.
Most of the deaths were in the elderly with strokes, tumors, cancer etc, but not all. I also ran a muscular dystrophy clinic and one of the hardest cases I saw was an infant with Werdnig Hoffman disease — similar to what Steven Hawking has, but much, much faster — she died at 1 year. Initially, I found the suffering of such patients and their families impossible to accept or understand, particularly when they affected the young, or even young adults in the graduate student age.
As noted earlier, I started med school in ’62, a time when the genetic code was first being cracked, and with the background then that many of you have presently understanding molecular biology as it was being unravelled wasn’t difficult. Usually when you know something you tend to regard it as simple or unimpressive. Not so the cell and life. The more you know, the more impressive it becomes.
Think of the 3.2 gigaBases of DNA in each cell. At 3 or so Angstroms aromatic ring thickness — this comes out to a meter or so stretched out — but it isn’t, rather compressed so it fits into a nucleus 5 – 10 millionths of a meter in diameter. Then since DNA is a helix with one complete turn every 10 bases, the genome in each cell contains 320,000,000 twists which must be unwound to copy it into RNA. The machinery which copies it into messenger RNA (RNA polymerase II) is huge — but the fun doesn’t stop there — in the eukaryotic cell to turn on a gene at the right time something called the mediator complex must bind to another site in the DNA and the RNA polymerase — the whole mess contains over 100 proteins and has a molecular mass of over 2 megaDaltons (with our friend carbon containing only 12 Daltons). This monster must somehow find and unwind just the right stretch of DNA in the extremely cramped confines of the nucleus. That’s just transcription of DNA into RNA. Translation of the messenger RNA (mRNA) into protein involves another monster — the ribosome. Most of our mRNA must be processed lopping out irrelevant pieces before it gets out to the cytoplasm — this calls for the spliceosome — a complex of over 100 proteins plus some RNAs — a completely different molecular machine with a mass in the megaDaltons. There’s tons more that we know now, equally complex.
So what.
Gradually I came to realize that what needs explaining is not the poor child dying of Werdnig Hoffman disease but that we exist at all and for fairly prolonged periods of time and in relatively good shape (like my father who was actively engaged in the law and a mortgage operation until 6 months before his death at age100). Such is the solace of molecular biology. It ain’t much, but it’s all I’ve got (the religious have a lot more). You guys have the chemical background and the intellectual horsepower to understand molecular biology — and even perhaps to extend it.
“The power of language is its ambiguity” sayeth I. This came up because my nephew married a wonderful Russian expat a few weeks ago. Plucky fellow that he is, he’s learning to speak Russian. Like my wife’s friend of 50+ years ago he is amazed at how many words the language has. Russian apparently has a word for everything so there is little ambiguity, which must make the language hard to pun in.
Someone Googled the number of words in Russian and English and they’re about the same.
Perhaps the lack of ambiguity makes Russian hard to learn (and use). Computer languages (basic, C, pascal) are completely unambiguous. Every reserved word and operator means exactly one thing, no more no less.
Most people find programming far from intuitive. It’s hard to express our sloppy ideas in unambiguous computer language. Given it’s difficulty giving concrete form to your ideas, computer languages aren’t as powerful (in the sense of being easy to use) as your sloppy sentences.
Why should language be so ambiguous? My guess is, that it has to be this way given the way we perceive the world (and the way the world probably actually is — ontology if you want to impress your friends).
We don’t live in Plato’s world of perfect forms, but in a world of objects that only partially and rather poorly instantiate them. This is as true of science as anything else — even supposedly well defined terms change their meaning — are the giant viruses really viruses? What do we really mean by a gene? It used to be a part of DNA coding for a protein, but what about the DNA that controls when and where a protein is made. Mutations here can cause disease, so are they genes?
Language, to be useful, must express our imperfect ways of rigidly classifying the world (perhaps because such a classification is impossible).
Socially, I never thought of our family as inhibited, but the Russians I met seemed more alive and vibrant than our lot (this without them living up to their reputation of hard drinking).
What do Richard Feynman and Charles Darwin have in common? Both have written books which show a brilliant mind at work. I’ve started reading the New Millennium Edition of Feynman’s Lectures on Physics (which is the edition you should get as all 1165 errata found over the years have been corrected), and like Darwin his thought processes and their power are laid out for all to see. Feynman’s books are far from F = ma. They are basically polished versions of lectures, so it reads as if Feynman is directly talking to you. Example: “We have already discussed the difference between knowing the rules of the game of chess and being able to play.” Another: talking about Zeno “The Greeks were somewhat confused by such problems, being helped, of course, by some very confusing Greeks.”
He’s always thinking about the larger implications of what we know. Example: “Newton’s law has the peculiar property that if it is right on a certain small scale, then it will be right on a larger scale”
He then takes this idea and runs with it. “Newton’s laws are the ‘tail end’ of the atomic laws extrapolated to a very large size” The fact that they are extrapolatable and the fact that way down below are the atoms producing them means, that extrapolatable laws are the only type of physical law which could be discovered by us (until we could get down to the atomic level). Marvelous. Then he notes that the fundamental atomic laws (e.g. quantum mechanics) are NOTHING like what we see in the large scale environment in which we live.
If you like this sort of thing, you’ll love the books. I don’t think they would be a good way to learn physics for the first time however. No problems, etc. etc. But once you’ve had exposure to some physics “it is good to sit at the feet of the master” — Bill Gates.
Most of the readership is probably fully engaged with work, family career and doesn’t have time to actually read “The Origin of Species”. In retirement, I did,and the power of Darwin’s mind is simply staggering. He did so much with what little information he had. There was no clear idea of how heredity worked and at several points he’s a Lamarckian — inheritance of acquired characteristics. If you do have the time I suggest that you read the 1859 book chapter by chapter along with a very interesting book — Darwin’s Ghost by Steve Jones (published in 1999) which update’s Darwin’s book to contemporary thinking chapter by chapter. Despite the advances in knowledge in 140 years, Darwin’s thinking beats Jones hands down chapter by chapter.
What can 546 dogs tell us about cancer, and STDs (sexually transmitted diseases)? An enormous amount ! [ Science vol 365 pp. 440 – 441, 464 3aau9923 1 –> 7 ’19 ]. You may have heard about the transmissible tumor that has reduced the Tasmanian Devil population from its appearance in ’96 by 80%. The animals bite each other transmitting the tumor. Only 10 – 100 cells are transferred, but death occurs within a year. The cells survive because Tasmanian devels have low genetic diversity.
The work concerns a much older transmissible tumor (Canine Transmissible Venereal Tumor — aka CTVT) which appeared in Asia an estimated 6,000 year ago, and began dispersing worldwide 2,000 years ago. Unlike the Tasmanian devil tumor, the tumor is usually cleared by the immune system.
The Science paper has 80+ authors from all over the world, who sequenced the protein coding part of the dog genome (the exome) to a > 100fold depth. The exome contains 43.6 megabases. The tumor is transmitted by sex, and the authors note that this mode of transmission nearly requires a rather indolent clinical course, as the animal must survive long enough to transmit the organism again. This fits with syphilis, AIDs, gonorrhea. Contrast this with anthrax, cholera, plague which spread differently and kill much faster.
So what does CTVT tell us about cancer? Quite a bit. First some background. The Cancer Genome Atlas (CGA) was criticized as being a boondoggle, but it at least gave us an idea of how many mutations are present in various cancers– around 100 in colon and breast cancers.
Viewed across all dogs, the CTVT genome is riddled with somatic mutations (as compared to the genome of the dog carrying the tumor) –148,030 single nucleotide variants (3.4/1000 !) 12,177 insertion/deletions. Of the 20,000 dog genes only 2,000 didn’t contain a mutation. This implies that most genes in the mammalian genome aren’t needed by the cancer cells. The CTVTs also show no signs of the high rates of chromosomal instability seen in human tumors.
The work provides evidence that cancer isn’t inherently progressive. This gives hope that some relatively indolent human cancers (say cancer of the prostate) can be controlled. This calls for ‘adaptive therapy’ — something that limits tumor growth rather than trying to kill every cancer cell with curative therapy which, if it fails, essentially selects for more aggressive cancer cells.
Some 14,412 genes have 1 mutation changing the amino acid sequence (nonSynonymous) and 5,704 have protein truncating mutations. The ratio of synonymous to non synonymous mutations is about 3 implying that the mutations which have arisen haven’t been selected for (after all the triplet code for 20 amino acids and 1 stop codon has 64 possibilities), so the average amino acid has 3 codons for it. This is called neutral genetic drift.
They also found 5 mutated genes present in all 541 tumors — these are the driver mutations, 3 are well known, MYC, PTEN, and retinoblastoma1.
Tons to think about here. I’ll be away for a few weeks traveling and playing music, but this work should keep you busy thinking about its implications.