Dennett on Competence without Comprehension
William A. Dembski June 28, 2012 2:10 PM
Daniel Dennett has published a piece in the Atlantic focused on Alan Turing's contribution to computer science, a contribution that Dennett treats as proof that a reductionist, materialist understanding of life and cognition is well in hand. He closes his piece with the extraordinary claim,
If the history of resistance to Darwinian thinking is a good measure, we can expect that long into the future, long after every triumph of human thought has been matched or surpassed by "mere machines," there will still be thinkers who insist that the human mind works in mysterious ways that no science can comprehend.
This inflated claim forms the logical conclusion of Dennett's essay, so let's begin by examining how he got there.
Dennett's watchword throughout his essay is that competence can be achieved without comprehension. Dennett here uses competence in the functionalist sense of a system achieving a level of performance that in human contexts would be ascribed to comprehension (i.e., intelligence). Accordingly, as an atheistic materialist, he sees life as emerging by an evolutionary process that did not comprehend what it was doing and likewise he regards cognition as emerging from computational modules that know nothing about their interaction or purpose.
Dennett's target throughout his essay is intelligent design, which he disparages as "the trickle-down theory of intelligence." According to him, this theory mistakenly holds that things exhibiting competence (such as life and cognition) do so because they are the product of comprehension (aka intention, purpose, intelligence). Dennett needs to reverse this order, making comprehension a product of mindless material forces and thus placing comprehension logically downstream from competence. (For him, the world does not begin with mind; mind is something that emerges out of it via a mindless process.)
At such a general level of description, there's nothing new or surprising here. Atheistic materialism admits a very limited set of answers to how life and cognition could have emerged and may properly be explained. Where this position becomes interesting, if at all, is in the precise character of its proposals. The devil, as always, is in the details. Dennett claims to fill in these details with the results of modern science. But, in fact, the science that he cites cannot do the heavy lifting that he demands of it.
Dennett thinks that Alan Turing provides support for his brand of atheism. Turing himself was an atheist, though Dennett makes nothing of this, turning instead to Turing's work. In 1936 Turing proposed a universal mechanism for performing any and all computations, since dubbed a Turing machine. In the last seventy-plus years, many other formal systems have been proposed for performing any and all computations (cellular automata, neural nets, unlimited register machines, etc.), and they've all been shown to perform the same -- no less and no more -- computations as Turing's originally proposed machine.
Now at first blush one would think that because Turing had "invented" a "machine," this might give even Dennett pause and lead him to take a second look at his claim that competence precedes comprehension. For any competence exhibited by a Turing machine would, on its face, presuppose comprehension by, in this case, a mathematician named Turing, who understood the nature of computability and invented a machine (albeit an abstract one) that could perform any and all computations. And one would be wrong -- atheists like Dennett have a knack for turning even the most damning evidence against their position into decisive confirmation of it. A failed Wall Street banker following his example would look not merely for a government bailout but would insist on taking over both the Fed and Treasury.
It's important here to distinguish the implications that Dennett draws from Turing's work from the actual substance of that work. According to Dennett, "Alan Turing created a new world of science and technology, setting the stage for solving one of the most baffling puzzles remaining to science, the mind-body problem." Okay, that's the implication, and the solution that Dennett envisions is one that reduces mind to computation. He continues to be an advocate of what has come to be known as "strong artificial intelligence," and it's the case that Turing held to this position as well.
What about the actual substance of Turing's work? Dennett quotes Turing's famous 1936 paper on computability: "It is possible to invent a single machine which can be used to compute any computable sequence." Dennett elaborates:
Turing didn't just intuit that this remarkable feat was possible; he showed exactly how to make such a machine. With that demonstration the computer age was born. It is important to remember that there were entities called computers before Turing came up with his idea, but they were people, clerical workers with enough mathematical skill, patience, and pride in their work to generate reliable results of hours and hours of computation, day in and day out.
Dennett's account of Turing's work misrepresents the reality at a number of levels. Dennett gets the history wrong: Turing was not the first to invent a general system of computation. Turing's advisor at Princeton, Alonzo Church, had in 1932 published a paper in which he introduced the lambda calculus, which proved to be logically equivalent to a Turing machine. And a hundred years earlier, Charles Babbage had invented an analytical engine (first described in 1837, though never implemented because of technological and cost constraints) that was also logically equivalent to the Turing machine.
Nor is it the case that prior to Turing, computers were people. Granted, Babbage's difference and analytical engines were never built. But abacuses have been around for a long time. Blaise Pascal invented a mechanical calculator in 1642. Commercially successful mechanical calculators were available by the 1850s (cf. Thomas de Colmar's arithmometer). In addition, people have for long envisioned computational devices capable of performing the tasks performed by people. Automated chess playing programs were envisioned during the Enlightenment. Mathematician and philosopher Gottfried Leibniz (co-inventor with Isaac Newton of the calculus) in the 1600s outlined an Ars Combinatoria, which took inspiration from the 13th-century philosopher and logician Raymond Lully, who had proposed an Ars Generalis Ultima, a general method (or machine, as Dennett might call it) for resolving all important questions. All of these computational forerunners of Turing, by the way, were Christians and didn't see computation as somehow posing a challenge to human exceptionalism or inviting a reduction of the human mind to computation.
Dennett's problems with portraying Turing's work (to say nothing of the implications he draws from that work) don't just end with history. He also misrepresents the nature of computation. It helps, in his misrepresentation, that Dennett never explains what computation actually is, so let me lay it out here. At the heart of computation is the idea of an algorithm. An algorithm is a finite, step-by-step procedure that at each step tells you precisely what to do and what step to take next. Take, for instance, the algorithm you learned in grade school for doing long division. Given a number that needs to be divided by another, this algorithm gives a finite sequence of steps for determining the quotient and the remainder. If you are given a problem, it is computable provided you can find an algorithm that resolves it. As it turns out, there are problems in mathematics that can be proved to be beyond resolution by any algorithm (e.g., the halting problem).
How, then, does Dennett misrepresent Turing's actual work? Turing proposed a machine that can be used to perform any computation. According to Dennett, Turing
showed exactly how to make such a machine ... The Pre-Turing world was one in which computers were people, who had to understand mathematics in order to do their jobs. Turing realized that this was just not necessary: you could take the tasks they performed and squeeze out the last tiny smidgens of understanding, leaving nothing but brute, mechanical actions.
The reason I say Dennett has misrepresented the substance of Turing's work is that he gives the impression that the Turing machine renders people unnecessary to the task of computation. But nothing could be further from the truth. The Turing machine, as characterized in 1936 by Alan Turing, was a general computational framework in which any particular computation could be realized. Yet to resolve any particular computational problem requires programming a Turing machine to solve it.
It is convenient for Dennett that just as he didn't define computation, he also didn't define a Turing machine, so let me define it briefly. Something is a Turing machine if it has a "tape" that extends infinitely in both directions, with the tape subdivided into identical adjacent squares, each of which can have written on it one of a finite alphabet of symbols (usually just zero and one). In addition, a Turing machine has a "tape head," that can move to the left or right on the tape and erase and rewrite the symbol that's on a current square. Finally, what guides the tape head is a finite set of "states" that, given one state, looks at the current symbol, keeps or changes it, moves the tape head right or left, and then, on the basis of the symbol that was there, makes active another state. In modern terms, the states constitute the program and the symbols on the tape constitute data.
From this it's obvious that a Turing machine can do nothing unless it is properly programmed to do so. Nor does it help to invoke a universal Turing machine. Basically, what a universal Turing machine does is cash in on the equivalence between programs and data, and allow a very simple fixed set of states to be used to perform any computation whatsoever provided that the initial data on the tape is suitably programmed to resolve the computation. Universal Turing machines can be quite simple. In the 1960s, MIT's Marvin Minsky came up with a 7-state, 4-symbol Turing Machine. Steve Wolfram currently touts a 2-state, 3-symbol Turing machine as the simplest.
How could universal Turing machines be so simple and still be universal in the sense of being able to solve any computational problem? It's because universal machines are not programmed to solve any problem in particular. You just have to load the program on the tape. A universal Turing machine reads through the tape, interpreting the data as a program, and then executing the steps of the program. Essentially, all a universal Turing machine does is: read instruction, execute it, read instruction, execute it, read instruction, execute it, etc. The challenge is to have the right set of instructions to execute. This is an inconvenient point that Dennett, in the interest of pushing his agenda, understandably wants to ignore.
I need here to say something about the concept of a material mechanism, which figures so prominently in Dennett's essay. Darwinian natural selection acting on random variations is, for Dennett, a material mechanism. Likewise, a Turing machine is for him a material mechanism. But there is an ambiguity in his use of the term mechanism that bears scrutiny, and which Michael Polanyi, writing specifically in reference to biology, clarified back in the 1960s:
Up to this day one speaks of the mechanistic conception of life both to designate an explanation of life in terms of physics and chemistry, and an explanation of living functions as machineries -- though the latter excludes the former. The term "mechanistic" is in fact so well established for referring to these two mutually exclusive conceptions, that I am at a loss to find two different words that will distinguish between them.
For Polanyi mechanisms, conceived as causal processes operating in nature, could not account for the origin of mechanisms, conceived as "machines or machinelike features of organisms." (See his article "Life Transcending Physics and Chemistry" in the August 1967 issue of Chemical and Engineering News.)
Dennett conveniently conflates these two uses of mechanism. Once a Turing machine is properly programmed, it will produce the solution to any computational problem. But humans -- read "intelligent designers" -- invariably do the programming. Turing, far from having obviated the "trickle-down theory of intelligence," actually underscores its preeminent role in the field of computation.
A running theme in Dennett's article is how Turing's achievement for computation parallels Darwin's for evolutionary biology. Turing supposedly gives us a mechanism that shows how mind may be reduced to computational processes. Likewise Darwin supposedly gives us a mechanism that shows how life may be reduced to evolutionary processes. Interestingly, just as his reduction of mind to computation fails (Dennett's reductionism fails to factor in the programmer), so does his reduction of life to purposeless evolutionary processes. The best evidence these days suggests that evolution, insofar as it operates at all, is teleological.
Here I need not refer solely to the ID literature. Molecular biologist James Shapiro sees evolution as proceeding not by natural selection but by natural genetic engineering, in which organisms guide their own evolution (see his book Evolution: A View from the 21st Century as well as his exchange with me here at ENV). Moreover, recent work at the Evolutionary Informatics Lab (see the publications page at www.evoinfo.org, especially the article "Life's Conservation Law") shows that Darwinian processes, if they are going to accomplish anything of consequence for the evolution of life, need to be "programmed" with an appropriate fitness landscape that is far from self-explanatory and, in fact, demonstrates that Darwinian processes are themselves deeply teleological (despite constant advertisements to the contrary by thinkers such as Dennett and Dawkins).
So what are we to make of Dennett's conclusion to his essay?
If the history of resistance to Darwinian thinking is a good measure, we can expect that long into the future, long after every triumph of human thought has been matched or surpassed by "mere machines," there will still be thinkers who insist that the human mind works in mysterious ways that no science can comprehend.
Dennett makes up in bluster what he lacks in evidence. Resistance to Darwinian thinking is on the rise because our best science is demonstrating its inadequacies. And the claim that the human mind is about to be equaled or surpassed by machines remains the stuff of science fiction (chess and jeopardy playing programs notwithstanding).
But I would close this response to Dennett with a question. Throughout his essay, he tries to minimize comprehension at the expense of competence. And yet the very last word of his essay is the word "comprehend." In his concluding sentence, he faults those, like me, who would question that science can comprehend the way the mind works (with science here conceived as a reductionist science that reduces everything ultimately to the interaction of material forces). But according to Dennett, the mind doesn't work by comprehending things, at least not really. It works simply by running various computational modules. Comprehension or understanding by humans is, as he puts it, just sorta comprehension or sorta understanding.
So my question for Dennett is this: In precisely what sense does he comprehend what he wrote in his essay and how does that comprehension justify his sneering contempt for those who disagree with him on his biological and computational reductionism?
William A. Dembski June 28, 2012 2:10 PM
Daniel Dennett has published a piece in the Atlantic focused on Alan Turing's contribution to computer science, a contribution that Dennett treats as proof that a reductionist, materialist understanding of life and cognition is well in hand. He closes his piece with the extraordinary claim,
If the history of resistance to Darwinian thinking is a good measure, we can expect that long into the future, long after every triumph of human thought has been matched or surpassed by "mere machines," there will still be thinkers who insist that the human mind works in mysterious ways that no science can comprehend.
This inflated claim forms the logical conclusion of Dennett's essay, so let's begin by examining how he got there.
Dennett's watchword throughout his essay is that competence can be achieved without comprehension. Dennett here uses competence in the functionalist sense of a system achieving a level of performance that in human contexts would be ascribed to comprehension (i.e., intelligence). Accordingly, as an atheistic materialist, he sees life as emerging by an evolutionary process that did not comprehend what it was doing and likewise he regards cognition as emerging from computational modules that know nothing about their interaction or purpose.
Dennett's target throughout his essay is intelligent design, which he disparages as "the trickle-down theory of intelligence." According to him, this theory mistakenly holds that things exhibiting competence (such as life and cognition) do so because they are the product of comprehension (aka intention, purpose, intelligence). Dennett needs to reverse this order, making comprehension a product of mindless material forces and thus placing comprehension logically downstream from competence. (For him, the world does not begin with mind; mind is something that emerges out of it via a mindless process.)
At such a general level of description, there's nothing new or surprising here. Atheistic materialism admits a very limited set of answers to how life and cognition could have emerged and may properly be explained. Where this position becomes interesting, if at all, is in the precise character of its proposals. The devil, as always, is in the details. Dennett claims to fill in these details with the results of modern science. But, in fact, the science that he cites cannot do the heavy lifting that he demands of it.
Dennett thinks that Alan Turing provides support for his brand of atheism. Turing himself was an atheist, though Dennett makes nothing of this, turning instead to Turing's work. In 1936 Turing proposed a universal mechanism for performing any and all computations, since dubbed a Turing machine. In the last seventy-plus years, many other formal systems have been proposed for performing any and all computations (cellular automata, neural nets, unlimited register machines, etc.), and they've all been shown to perform the same -- no less and no more -- computations as Turing's originally proposed machine.
Now at first blush one would think that because Turing had "invented" a "machine," this might give even Dennett pause and lead him to take a second look at his claim that competence precedes comprehension. For any competence exhibited by a Turing machine would, on its face, presuppose comprehension by, in this case, a mathematician named Turing, who understood the nature of computability and invented a machine (albeit an abstract one) that could perform any and all computations. And one would be wrong -- atheists like Dennett have a knack for turning even the most damning evidence against their position into decisive confirmation of it. A failed Wall Street banker following his example would look not merely for a government bailout but would insist on taking over both the Fed and Treasury.
It's important here to distinguish the implications that Dennett draws from Turing's work from the actual substance of that work. According to Dennett, "Alan Turing created a new world of science and technology, setting the stage for solving one of the most baffling puzzles remaining to science, the mind-body problem." Okay, that's the implication, and the solution that Dennett envisions is one that reduces mind to computation. He continues to be an advocate of what has come to be known as "strong artificial intelligence," and it's the case that Turing held to this position as well.
What about the actual substance of Turing's work? Dennett quotes Turing's famous 1936 paper on computability: "It is possible to invent a single machine which can be used to compute any computable sequence." Dennett elaborates:
Turing didn't just intuit that this remarkable feat was possible; he showed exactly how to make such a machine. With that demonstration the computer age was born. It is important to remember that there were entities called computers before Turing came up with his idea, but they were people, clerical workers with enough mathematical skill, patience, and pride in their work to generate reliable results of hours and hours of computation, day in and day out.
Dennett's account of Turing's work misrepresents the reality at a number of levels. Dennett gets the history wrong: Turing was not the first to invent a general system of computation. Turing's advisor at Princeton, Alonzo Church, had in 1932 published a paper in which he introduced the lambda calculus, which proved to be logically equivalent to a Turing machine. And a hundred years earlier, Charles Babbage had invented an analytical engine (first described in 1837, though never implemented because of technological and cost constraints) that was also logically equivalent to the Turing machine.
Nor is it the case that prior to Turing, computers were people. Granted, Babbage's difference and analytical engines were never built. But abacuses have been around for a long time. Blaise Pascal invented a mechanical calculator in 1642. Commercially successful mechanical calculators were available by the 1850s (cf. Thomas de Colmar's arithmometer). In addition, people have for long envisioned computational devices capable of performing the tasks performed by people. Automated chess playing programs were envisioned during the Enlightenment. Mathematician and philosopher Gottfried Leibniz (co-inventor with Isaac Newton of the calculus) in the 1600s outlined an Ars Combinatoria, which took inspiration from the 13th-century philosopher and logician Raymond Lully, who had proposed an Ars Generalis Ultima, a general method (or machine, as Dennett might call it) for resolving all important questions. All of these computational forerunners of Turing, by the way, were Christians and didn't see computation as somehow posing a challenge to human exceptionalism or inviting a reduction of the human mind to computation.
Dennett's problems with portraying Turing's work (to say nothing of the implications he draws from that work) don't just end with history. He also misrepresents the nature of computation. It helps, in his misrepresentation, that Dennett never explains what computation actually is, so let me lay it out here. At the heart of computation is the idea of an algorithm. An algorithm is a finite, step-by-step procedure that at each step tells you precisely what to do and what step to take next. Take, for instance, the algorithm you learned in grade school for doing long division. Given a number that needs to be divided by another, this algorithm gives a finite sequence of steps for determining the quotient and the remainder. If you are given a problem, it is computable provided you can find an algorithm that resolves it. As it turns out, there are problems in mathematics that can be proved to be beyond resolution by any algorithm (e.g., the halting problem).
How, then, does Dennett misrepresent Turing's actual work? Turing proposed a machine that can be used to perform any computation. According to Dennett, Turing
showed exactly how to make such a machine ... The Pre-Turing world was one in which computers were people, who had to understand mathematics in order to do their jobs. Turing realized that this was just not necessary: you could take the tasks they performed and squeeze out the last tiny smidgens of understanding, leaving nothing but brute, mechanical actions.
The reason I say Dennett has misrepresented the substance of Turing's work is that he gives the impression that the Turing machine renders people unnecessary to the task of computation. But nothing could be further from the truth. The Turing machine, as characterized in 1936 by Alan Turing, was a general computational framework in which any particular computation could be realized. Yet to resolve any particular computational problem requires programming a Turing machine to solve it.
It is convenient for Dennett that just as he didn't define computation, he also didn't define a Turing machine, so let me define it briefly. Something is a Turing machine if it has a "tape" that extends infinitely in both directions, with the tape subdivided into identical adjacent squares, each of which can have written on it one of a finite alphabet of symbols (usually just zero and one). In addition, a Turing machine has a "tape head," that can move to the left or right on the tape and erase and rewrite the symbol that's on a current square. Finally, what guides the tape head is a finite set of "states" that, given one state, looks at the current symbol, keeps or changes it, moves the tape head right or left, and then, on the basis of the symbol that was there, makes active another state. In modern terms, the states constitute the program and the symbols on the tape constitute data.
From this it's obvious that a Turing machine can do nothing unless it is properly programmed to do so. Nor does it help to invoke a universal Turing machine. Basically, what a universal Turing machine does is cash in on the equivalence between programs and data, and allow a very simple fixed set of states to be used to perform any computation whatsoever provided that the initial data on the tape is suitably programmed to resolve the computation. Universal Turing machines can be quite simple. In the 1960s, MIT's Marvin Minsky came up with a 7-state, 4-symbol Turing Machine. Steve Wolfram currently touts a 2-state, 3-symbol Turing machine as the simplest.
How could universal Turing machines be so simple and still be universal in the sense of being able to solve any computational problem? It's because universal machines are not programmed to solve any problem in particular. You just have to load the program on the tape. A universal Turing machine reads through the tape, interpreting the data as a program, and then executing the steps of the program. Essentially, all a universal Turing machine does is: read instruction, execute it, read instruction, execute it, read instruction, execute it, etc. The challenge is to have the right set of instructions to execute. This is an inconvenient point that Dennett, in the interest of pushing his agenda, understandably wants to ignore.
I need here to say something about the concept of a material mechanism, which figures so prominently in Dennett's essay. Darwinian natural selection acting on random variations is, for Dennett, a material mechanism. Likewise, a Turing machine is for him a material mechanism. But there is an ambiguity in his use of the term mechanism that bears scrutiny, and which Michael Polanyi, writing specifically in reference to biology, clarified back in the 1960s:
Up to this day one speaks of the mechanistic conception of life both to designate an explanation of life in terms of physics and chemistry, and an explanation of living functions as machineries -- though the latter excludes the former. The term "mechanistic" is in fact so well established for referring to these two mutually exclusive conceptions, that I am at a loss to find two different words that will distinguish between them.
For Polanyi mechanisms, conceived as causal processes operating in nature, could not account for the origin of mechanisms, conceived as "machines or machinelike features of organisms." (See his article "Life Transcending Physics and Chemistry" in the August 1967 issue of Chemical and Engineering News.)
Dennett conveniently conflates these two uses of mechanism. Once a Turing machine is properly programmed, it will produce the solution to any computational problem. But humans -- read "intelligent designers" -- invariably do the programming. Turing, far from having obviated the "trickle-down theory of intelligence," actually underscores its preeminent role in the field of computation.
A running theme in Dennett's article is how Turing's achievement for computation parallels Darwin's for evolutionary biology. Turing supposedly gives us a mechanism that shows how mind may be reduced to computational processes. Likewise Darwin supposedly gives us a mechanism that shows how life may be reduced to evolutionary processes. Interestingly, just as his reduction of mind to computation fails (Dennett's reductionism fails to factor in the programmer), so does his reduction of life to purposeless evolutionary processes. The best evidence these days suggests that evolution, insofar as it operates at all, is teleological.
Here I need not refer solely to the ID literature. Molecular biologist James Shapiro sees evolution as proceeding not by natural selection but by natural genetic engineering, in which organisms guide their own evolution (see his book Evolution: A View from the 21st Century as well as his exchange with me here at ENV). Moreover, recent work at the Evolutionary Informatics Lab (see the publications page at www.evoinfo.org, especially the article "Life's Conservation Law") shows that Darwinian processes, if they are going to accomplish anything of consequence for the evolution of life, need to be "programmed" with an appropriate fitness landscape that is far from self-explanatory and, in fact, demonstrates that Darwinian processes are themselves deeply teleological (despite constant advertisements to the contrary by thinkers such as Dennett and Dawkins).
So what are we to make of Dennett's conclusion to his essay?
If the history of resistance to Darwinian thinking is a good measure, we can expect that long into the future, long after every triumph of human thought has been matched or surpassed by "mere machines," there will still be thinkers who insist that the human mind works in mysterious ways that no science can comprehend.
Dennett makes up in bluster what he lacks in evidence. Resistance to Darwinian thinking is on the rise because our best science is demonstrating its inadequacies. And the claim that the human mind is about to be equaled or surpassed by machines remains the stuff of science fiction (chess and jeopardy playing programs notwithstanding).
But I would close this response to Dennett with a question. Throughout his essay, he tries to minimize comprehension at the expense of competence. And yet the very last word of his essay is the word "comprehend." In his concluding sentence, he faults those, like me, who would question that science can comprehend the way the mind works (with science here conceived as a reductionist science that reduces everything ultimately to the interaction of material forces). But according to Dennett, the mind doesn't work by comprehending things, at least not really. It works simply by running various computational modules. Comprehension or understanding by humans is, as he puts it, just sorta comprehension or sorta understanding.
So my question for Dennett is this: In precisely what sense does he comprehend what he wrote in his essay and how does that comprehension justify his sneering contempt for those who disagree with him on his biological and computational reductionism?
No comments:
Post a Comment