Saturday, May 29, 2010

No Ghost in the Machine

Back when I was a graduate student at Berkeley, I worked as a computer consultant for UC Berkeley's Computing Services department. One day a woman came in and wanted a tour of our APL graphics lab. So I showed her the machines we had, which included Tektronix 4013 and 4015 terminals, and one 4027, and drew a few things for her. But then the incomprehension set in:

"Who's doing the drawing on the screen?" she asked.

I explained that the program was doing the drawing.

"No, I mean what person is doing the drawing that we see?" she clarified.

I explained that the program was written by me and other people.

"No, I don't mean the program. I mean, who is doing the actual drawing, right now?

I explained that an electron gun inside the machine activated a zinc sulfide phosphor, and that it was directed by the program. I then showed her what a program looked like.

All to no avail. She could not comprehend that all this was taking place with no direct human control. Of course, humans wrote the program and built the machines, but that didn't console her. She was simply unable to wrap her mind around the fact that a machine could draw pictures. For her, pictures were the province of humans, and it was impossible that this province could ever be invaded by machines. I soon realized that nothing I could say could rescue this poor woman from the prison of her preconceptions. Finally, after suggesting some books about computers and science she should read, I told her I could not devote any more time to our discussion, and I sadly went back to my office. It was one of the first experiences I ever had of being unable to explain something so simple to someone.

That's the same kind of feeling I have when I read something like this post over at Telic Thoughts. Bradford, one of the more dense commentators there, quotes a famous passage of Leibniz

Suppose that there be a machine, the structure of which produces thinking, feeling, and perceiving; imagine this machine enlarged but preserving the same proportions, so that you could enter it as if it were a mill. This being supposed you might visit its inside; but what would you observe there? Nothing but parts which push and move each other, and never anything which could explain perception.

But Leibniz's argument is not much of an argument. He seems to take it for granted that understanding how the parts of a machine work can't give us understanding of how the machine functions as a whole. Even in Leibniz's day this must have seemed silly.

Bradford follows it up with the following from someone named RLC:

The machine, of course, is analogous to the brain. If we were able to walk into the brain as if it were a factory, what would we find there other than electrochemical reactions taking place along the neurons? How do these chemical and electrical phenomena map, or translate, to sensations like red or sweet? Where, exactly, are these sensations? How do chemical reactions generate things like beliefs, doubts, regrets, certainty, or purposes? How do they create understanding of a problem or appreciation of something like beauty? How does a flow of ions or the coupling of molecules impose a meaning on a page of text? How can a chemical process or an electrical potential have content or be about something?

Like my acquaintance in the graphics lab 30 years ago, poor RLC is trapped by his/her own preconceptions, I don't know what to say. How can anyone, writing a post on a blog which is entirely mediated by things like electrons in wires or magnetic disk storage, nevertheless ask "How can a chemical process or an electrical potential have content or be about something?" The irony is really mind-boggling. Does RLC ever use a phone or watch TV? For that matter, if he/she has trouble with the idea of "electrical potential" being "about something", how come he/she has no trouble with the idea of carbon atoms on a page being "about something"?

We are already beginning to understand how the brain works. We know, for example, how the eye focuses light on the retina, how the retina contains photoreceptors, how these photoreceptors react to different wavelengths of light, and how signals are sent through the optic nerve to the brain. We know that red light is handled differently from green light because different opsins absorb different wavelengths. And the more we understand, the more the brain looks like Leibniz's analogy. There is no ghost in the machine, there are simply systems relying on chemistry and physics. That's it.

To be confused like RLC means that one has to believe that all the chemical and physical apparatus of the brain, which is clearly collects data from the outside world and processes it, is just a coincidence. Sure, the apparatus is there, but somehow it's not really necessary, because there is some "mind" or "spirit" not ultimately reducible to the apparatus.

Here's an analogy. Suppose someone gives us a sophisticated robot that can navigate terrain, avoid obstacles, and report information about what it has seen. We can then take this robot apart, piece by piece. We see and study the CCD camera, the chips that process the information, and the LCD screens. Eventually we have a complete picture of how the robot works. What did we fail to understand by our reductionism?

Our understanding of how the brain works, when it is completed, will come from a complete picture of how all its systems function and interact. There's no magic to it - our sensations, feelings, understanding, appreciation of beauty - they are all outcomes of these systems. And there will still be people like RLC who will sit there, uncomprehending, and complain that we haven't explained anything, saying,

"But how can chemistry and physics be about something?"

Friday, May 28, 2010

Casey Luskin: Information Theory Expert

Well, it looks like the Discovery Institute was so unnerved by my pointing out the misunderstandings and misrepresentations in Stephen Meyer's book, Signature in the Cell, that they devoted two whole chapters to attacking me in their new book. The always-repulsive David Klinghoffer called me a "pygmy" and made fun of the name of my university (page 6). Paul Nelson called my critique a "fluffy confection" and alleged I was guilty of "sophistry". Casey Luskin said I indulged in "gratuitous invective".

The DI's responses to my arguments about Signature are about at the level of what you'd expect from them. I already replied to Paul Nelson months ago here, but of course they didn't see fit to reference that.

In their new book, they trot out lawyer Casey Luskin as their new expert on information theory. Luskin's main points are

(1) Shannon and Kolmogorov complexity are not "useful metrics of functional biological information" and
(2) eminent scientists have adopted Dembski and Meyer's notion of "functional information".

Here's my response:

(1) No measure of information is perfect. Both Shannon and Kolmogorov have proved useful in biological contexts; to claim, as Luskin does that they are "outmoded tools" is ridiculous. An exercise for Luskin, or anyone else: do a search of the scientific literature for "Shannon information" in biology, and count how many hits you get. Now do the same thing for "functional information". See the difference?

Indeed, it is the apparent incompressibility of the genome that suggests, through Kolmogorov complexity, that random mutations played a very significant role in its creation.

(2) Luskin cites a 1973 book by Orgel, where Orgel used the term "specified complexity", as evidence that creationist information is used by real scientists. However, Orgel did not give a rigorous definition of the term, and no one has since then. The term was only used in a popular book, and Orgel never published a definition in the peer-reviewed scientific literature. Dembski later claimed that Orgel's term was the same as his, and Luskin now repeats this falsehood. A lie can travel around the world, while the truth is just lacing up its sneakers.

Luskin points out that very recently, Szostak has introduced a notion of "functional information". However, Szostak's "functional information" is not a general-purpose measure of information. It certainly does not obey the axioms of information as studied by information theorists, and it does not obey Dembski's "law of conservation of information". Furthermore, it is only defined relative to a set of functions that one specifies. Change the functions, and you might get a completely different measure. So it is clear that Szostak's measure is not the same as Dembski's.

Might Szostak's idea prove useful one day? Perhaps, although the jury is still out. It has yet to receive many citations in the scientific literature; one of the papers cited by Luskin is by creationist Kirk Durston. The last time I looked, Durston's paper had essentially no impact at all, to judge by citation counts.

In any event, my claim was "Information scientists do not speak about ‘specified information’ or ‘functional information.’” Luskin offers Szostak as a counterexample. But Szostak is not an information scientist; he's a biologist. No discussion of "functional information" has yet appeared in the peer-reviewed information theory literature, which was my point. Luskin's trotting out of Szostak's paper does not refute that.

A Much-Too-Credulous Review of Signature in the Cell

John Walker is a pretty bright guy who's done some interesting work, but in this review of Stephen Meyer's Signature in the Cell, he demonstrates insufficient skepticism about Meyer's claims.

He asks, where did the information to specify the first replicator come from?, and then follows with this non-sequitur: The simplest known free living organism (although you may quibble about this, given that it's a parasite) has a genome of 582,970 base pairs, or about one megabit (assuming two bits of information for each nucleotide, of which there are four possibilities).

Of course, this is silly. Nobody thinks the first replicator was anywhere near this complicated, or even that it necessarily had a "genome" based on DNA. Even the genetic code itself may have evolved. Hypotheses like the RNA World suggest that the first replicator might have consisted of only a few hundred base-pairs.

Oddly enough for someone who has worked in artificial life, Walker shows no sign of having read Koza's 1994 paper, which shows how self-replicators can emerge spontaneously and with high probability in computer simulations.

He then goes on to claim you find that in the finite time our universe has existed, you could have produced about 500 bits of structured, functional information by random search. The only problem? The term "structured, functional information" has no definition in the scientific literature - it's just babble invented by creationists like Dembski and Meyer. There's no sign that Walker has read any of the criticism of Dembski's work.

Walker goes on to give a definition of "structured, functional information" as "information which has a meaning expressed in a separate domain than its raw components". But then there are lots of examples of such information occurring in nature, such as varves. Varves are layers of sediment which encode yearly information about the environment in which they formed. Another example is Arctic ice cores, which encode essential information about climate that is being mined by climatologists today.

Finally, the notion of "meaning" is incoherent. Disagree? Then tell me which of the following strings have "meaning" and which do not:

#1:
001001001100011011111010010111010010111000100000100000100111

#2:
010100111011001100001111101011100101110011110110010000001101

#3:
101010101010101010101010101010101010101010101010101010101010

#4:
101111101111101110101110111110101111101110101110101110101001

If that's too easy for you, let's try another. List all the binary strings of length 10 that have "meaning", and explain, for each one, what the meaning is.

Bottom line: insufficient skepticism leads to credulous acceptance of bad ideas.

Wednesday, May 26, 2010

Stephen Meyer - More Honesty Problems?

At Christianity Today, Stephen Meyer repeats the falsehood that "We know that information—whether inscribed in hieroglyphics, written in a book, or encoded in a radio signal—always comes from an intelligent source." It's simply not so - for example, in the Kolmogorov theory, any random source produces information. Even in Meyer's own idiosyncratic definition of information, natural systems produce information - such as when you stick your head out the window to see if it will rain that day. Where did you get that information? Not from any intelligent source.

And he adds some new falsehoods: "My recent book on the subject received enthusiastic endorsements from many scientists not previously known as advocates of ID, such as chemist Philip Skell, a National Academy of Sciences member..."

As is well-known to anyone who follows the creation-evolution debate, Philip Skell is a longtime evolution opponent. His anti-evolution activity dates from at least 2000, and he has been quite active since then.

Meyer also claims, "those who reject ID within the scientific community do so not because they have a better explanation of the relevant evidence, but because they affirm a definition of science that requires them to reject explanations involving intelligence—whatever the evidence shows". Scientists don't reject explanations involving "intelligence"; they simply don't find "intelligence" alone to be a useful explanation for most phenomena. No archaeologist finds a potsherd and exclaims, "Intelligence must have been involved in the creation of this pot!" To do so would be regarded as moronic. Rather, archaeologists spend their time figuring out who made an artifact, what they used it for, and how it fits into a larger understanding of the human culture it was a part of. Contrary to Meyer's bogus claim, fields like archaeology have no problem incorporating human agency into their studies. But no scientific field incorporates agency without some evidence of the agent actually existing - something Meyer has yet to provide.

If ID wants to be taken seriously, ID advocates have to distance themselves from spokesmen who are more interested in public relations than scientific truth.

Tuesday, May 25, 2010

How to Test for Syphilis

Yesterday on NPR's "All Things Considered" I heard this segment about "sparsity", which tried to link several different issues about data compression in one story. I don't think it was very successful, and the chosen term "sparsity" wasn't really representative of the content.

Nevertheless, the piece opened up with an interesting puzzle. The Army wants to do comprehensive blood tests for syphilis, a relatively rare disease, but each individual test is expensive. How can they test everyone more cheaply?

The idea is simple: mix the blood of g individuals together, and test that. A positive outcome indicates that at least one person in the group has syphilis, and a negative outcome indicates that no person in the group has it. In the event of a positive outcome, test all the members of the group.

Now let p be the probability that a randomly-chosen person has syphilis, and let N be the population size. What is the optimal size for the group? Choose g too small, and you end up doing lots of tests, because N/g (the number of groups) is large. Choose g too large, and it becomes very likely that testing the group will be positive, so you end up doing lots of tests again. Somewhere in between is the optimal choice of g.

How do we find it? There are N/g groups, and we have to test each of them. A randomly-chosen person fails to have syphilis with probability 1-p, so the probability that everyone in the group fails to have it is (1-p)g. Hence the probability that a group tests positive is 1-(1-p)g, and the expected number of positive-testing groups is (N/g)(1-(1-p)g). We have to test each person in a positive group, so this means g(N/g)(1-(1-p)g) = N(1-(1-p)g) additional tests. If the cost of a test is C, then the total cost is CN/g (the cost to test each group), plus CN(1-(1-p)g) (the cost to test everyone in the positive-testing groups). Factoring out CN, we need to minimize

1/g + 1 - (1-p)g.       (1)

Notice that this expression only depends on p, the probability.

We can, in fact, find a closed form for this minimum (or at least Maple can), but it is somewhat complicated, and depends on the little-known Lambert W-function. It's easier to compute the minimum through bisection or some other method for any particular p. A recent estimate is that syphilis occurs in about 15.4 per 100,000 people, which corresponds to p = .000154. For this value of p, the optimal g is g = 81, which gives (1) the value of .0247. Thus, for this p, we are able to test everyone by combining tests -- for less than 3% of the cost of testing everyone individually.

By the way, a simple heuristic argument suggests that the g that minimizes (1) will be about p: to minimize (1), choose g so that the two terms in the sum (1), 1/g and 1-(1-p)g, are equal. Set g = p; then 1/g = p½. On the other hand, 1-(1-p)g = 1 - (1-g-2)g. But (1-1/g)g is about 1/e for large g, so using the Taylor series for exp(x), we see that 1 - (1-g-2)g is about 1/g, too. So this choice of g will be very close to the one minimizing (1), and the value of (1) is therefore about 2/g = 2 p½.

Added: with more complicated tests (see the comments), one can do even better. But some of the proposed methods are so complicated that it seems to me the possibility of human error in carrying them out would rule them out.

Saturday, May 22, 2010

Charlie Went Through the Dryer





Son #2 put his Charlie Card (the electronic card that lets you ride on Boston public transit) through the dryer.

Twice.

Believe it or not, it still works.

Wednesday, May 19, 2010

Turing vs. the Creationists

Today we bring you your daily dose of breathtaking inanity from the creationist blog Uncommon Descent. A poster named "niwrad" (that's "Darwin" spelled backwards - the creationists are oh-so-clever) finds yet another reason to reject Darwinism.

"niwrad" draws an analogy with computer programming, and then "explain[s] as clearly as possible exactly what a program is". He goes on to say, "In order to process information – i.e. create software – it is necessary to create data and programs. Data is passive information: it cannot change or decide anything by itself" while "a program, in its simplest concept, is a blueprint specifying the reiteration of basic decision structures, about what to do and when to do it."

He/she then goes to say, "The argument which I am putting forward here cuts through these definitional controversies, because from my informatics-based perspective there are really only two possibilities, which can be summarized as follows: either (a) genes are data (which corresponds to the above old definition of a gene); or (b) genes are functions (which corresponds to the new definition)", gives incoherent reasons for rejecting both views, and concludes, "To sum up: Darwinism, from an informatics point of view, has absolutely zero credibility."

Poor "niwrad". He/she/it never learned much computer science, because we have known at least since 1936 that the artificial distinction between "program" and "data" is illusory. The existence of a universal machine shows that we can treat programs and data in the same fashion. That, indeed, was one of the fundamental insights of Alan Turing in his famous paper, "On computable numbers...". As Kleinberg and Papadimitriou remark, "Indeed, our style of writing programs and then executing them is so ingrained in our view of computation that it takes a moment to appreciate the consequences that flow from a universal machine. It means that programs and data are really the same thing: a program is just a sequence of symbols that looks like any other piece of input; but when fed to a universal machine, this input wakes up and begins to compute. Think of mobile code, java applets, e-mail viruses: your computer downloads them as data, and then runs them as programs."

Furthermore, we know from the field of DNA computing that very simple abstract operations, corresponding to the physics and chemistry of DNA, can simulate universal computation.

If creationists want to avoid being defeated by the ghost of Alan Turing, they need to spend more time reading about what is known in computer science and biology, and less time proclaiming (as "niwrad" did) that "God, also in this case, expects far less from us than what He Himself did and does".

Tuesday, May 18, 2010

Atheism - The Energy-Efficient Worldview

Spotted on a bus in Boston yesterday:



I guess that makes atheism the energy-efficient worldview. Go green!

Wednesday, May 12, 2010

An Academic Challenge

Today I saw the following poster at MIT:



This is the brainchild of the folks at phdchallenge.org. They want someone to get the phrase "I smoke crack rocks" into a scientific paper before December 1 of this year.

But someone at the challenge must have been smoking something, because in the FAQ, they say, "If your potential contest entry contains at least 90% of the words from challenge phrase, you may still submit it. For example, for the 2010 PhD Challenge, submissions containing the phrases I SMOKE CRACK and SMOKE CRACK ROCKS are valid and would be considered in the final judging." They seem to think that 75% is greater than or equal to 90%.

Tuesday, May 11, 2010

A Question for Christians

Here's a question for Christians - I'm genuinely interested in hearing your thoughts and I won't take issue with what you say.

Read this short blog post by Christian philosopher Doug Groothuis:

The more we submit to the higher, the more control we have over ourselves, and the more we find our place with that which is equal to us and and with that which is lower. Otherwise, we botch the hierarchy and bring (even more) chaos into this wounded world.

What do you make of it? Do you find it deep, or shallow? Insightful or fatuous? Meaningful or meaningless?

I'll give you my opinion in the comments, eventually, so as not to poison the well too much.

Thursday, May 06, 2010

Do Asthmatics Really Sneeze at Plastic Flowers?

Do asthmatics really sneeze at plastic flowers, presuming them to be real?

So a variety of papers, books, and websites -- many of them arguing that the mind is "immaterial" -- would have you believe. Just to list a few:

James Le Fanu, in his 2009 book Why Us? How Science Rediscovered the Mystery of Ourselves, writes (p. 219): "it is well recognised, for example, that asthmatics sneeze in the presence of plastic flowers, presuming them to be real", but he gives no citation.

Kenneth R. Pelletier, in a chapter "Between mind and body: stress, emotions and health" in a 1993 Consumer Reports book edited by D. Goleman and J. Gurin entitled Mind Body Medicine: How to Use Your Mind for Better Health, claims in the very first line of his article that "Asthmatics sneeze at plastic flowers" (p. 19). He gives no citation, but repeats this claim in his 1995 book Sound Mind, Sound Body: A New Model for Lifelong Health, writing (p. 79): "Control is the vital link between mind and body. It is the pivotal point between psychological attitudes and our physical responses. Asthmatics sneeze at plastic flowers."

Anees A. Sheikh, in his 2002 book Healing images: the role of imagination in health quotes Pelletier, as does Max Velmans, in a paper entitled How could conscious experiences affect brains? that appeared in the Journal of Consciousness Studies 9 (11) (2002), 3-29.

Arthur V. N. Brooks, in a paper presented at the Philosophical Club of Cleveland in 1993, asks "Why do asthmatics sneeze at plastic flowers?" No citation is given.

Christopher Gilbert, a physician currently at Kaiser Permanente in San Francisco, writes in a 1998 article in the Journal of Bodywork and Movement Therapies that "Asthma attacks, for example, can be set off by plastic flowers if the individual has had asthma responses to flowers in the past, and believes the flowers are real."

Naomi Judd, in her 1994 autobiography Love Can Build a Bridge, writes (p. 449) that "The mind is the deepest influence on the body. Asthmatics have been known to sneeze at plastic flowers." She seems to be citing Pelletier (1993).

A 1923 address by W. Langdon Brown that appeared in Proc. Roy. Soc. Med. 16 (1923), 1-16. He says, "That asthma often occurs in neuropathic families, and that asthmatics are unduly suggestible, are well-known facts. The paroxysm excited by an artificial flower figures in every text-book."

And what appears to be the original citation is an 1886 paper entitled "The production of the so-called "rose cold" by means of an artificial rose", by John Noland Mackenzie, in the American Journal of the Medical Sciences.

With all these citations, it would certainly seem to be a fact, right?

Not so fast.

First, single anecdotal examples (like that given in the 1886 paper) are not very convincing. There are lots of reasons other than psychological ones why someone allergic to flowers might sneeze in the presence of a plastic flower - such as, for example, having an allergy to plastic or residues of chemicals used to make plastic. That's why a controlled study would be needed to be truly convincing.

Second, Neville J. King, in the Journal of Behavioral Medicine 3 (2) (1980), 169-189, wrote, "The classical conditioning of human asthma attacks also has yet to be reliably demonstrated. Despite the many anecdotal cases of human asthma attacks elicited by the sight of dust and artificial flowers (Dunbar, 1954; McGovern and Knight, 1967) which are consistent with a classical conditioning model of asthma, attempts at the classical conditioning of human asthma in the laboratory have not been successful. Of approximately 100 asthmatic patients, Dekker et al. (1957) could succeed in classically conditioning asthma attacks to apparatus used in allergy investigations in only 2 of these patients. Also, Knapp (1963) reported negative findings after a careful and elaborate series of experiments on conditioned asthma in human beings. As Purcell and Weiss (1970) concluded in their review, "it appears accurate to state that, with either animals or human beings, the successful conditioning of asthma remains to be demonstrated, even in the opinion of those investigations whose original positive reports on conditioning are cited frequently" (p. 607).

Asthma is a common disease with a strong genetic component. If you have two asthmatic parents, your chances of developing asthma are about 1 in 2. Asthmatics react to various allergens and to infections of the upper respiratory system. In the 1930's, under the influence of Freud's batty ideas, many people claimed asthma was due to "separation anxiety", a claim that has since been discredited. Later, it has been claimed by some doctors that asthma was just a conditioned response, but as we have seen, this has not been demonstrated rigorously.

All in all, I am skeptical of the claim that asthmatics sneeze at plastic flowers. Even if it were true, I don't see how it supports the claim that the mind is "immaterial". I know very little about neuroscience, but it doesn't seem that farfetched to me that an neural network could, after repeated stimulation, associate some autonomic response such as sneezing, with some visual stimulation. No supernatural explanation seems to be required.

Our House on TV

The CTV show Dan for Mayor is filmed in Kitchener, Ontario, where I normally live. (This year I'm on sabbatical at MIT.)

I don't find the show particularly funny, but I have a soft spot for it, because last fall they filmed on our street. In Season 1, Episode 2, you can see the interior of our house. If you're in Canada, you can watch it on the CTV web site, but elsewhere you can buy it from the Itunes store or brave Megavideo. Our house appears about the 9 minute mark. Here's a screenshot:



Oddly enough, where we're temporarily living, in Newton, MA, another TV show was filmed just around the corner: a pilot for a CBS show called "Quinn-tuplets".

Oh, yeah, I forgot to say - that's Carl Zehr, mayor of Kitchener, in the pic above. Too bad we weren't there.