Brain and Mind 2: 253–259, 2001. © 2001 Kluwer Academic Publishers. Printed in the Netherlands.
253
John Horgan Responds JOHN HORGAN 241 Route 403, Garrison, N.Y. 10524, U.S.A. (E-mail:
[email protected]) (Accepted: 9 January 2001)
First let me say that I am delighted by these comments on The Undiscovered Mind. Although I intended my book for a general audience, I also hoped that it would provoke responses from professional mind-ponderers. I am thus grateful to Brain and Mind for persuading such distinguished scientists and philosophers to participate in this forum. I was also gratified that even the most critical respondents grant some of my basic points about mind-science, albeit reluctantly or inadvertently. I’ll address my critics’ remarks in the order in which I received them. Luciano da F. Costa contends that mind-science may progress to the extent that it embraces mathematics. Two points can be made here. First, as da F. Costa himself indicates, mathematical models might succeed in a strictly practical sense, giving us power to predict and control mental phenomena without providing the intellectual insight that we crave (and that non-mathematical theories such as psychoanalysis purport to deliver). That is arguably the case with quantum mechanics, which has enormous predictive power but is notoriously difficult to understand. But based on the record thus far, the mathematization of mindscience may yield neither deep intellectual insights nor practical benefits. After all, researchers have been spewing out mathematical models of the mind for decades. Cybernetics, chaos, complexity, catastrophe theory and information theory have all addressed the mind – with little success. What worries me is that the mathematization of a field may lead to the illusion of progress and thus to hubris rather than genuine advances. This syndrome can be seen in economics, which has become increasingly arrogant and divorced from reality (and driven by the concerns of Wall Street) as it has become more mathematical. What mathematics can do, perhaps, is help delineate the limits of social sciences such as economics and mind-sciences such as psychology. A fundamental lesson of chaos theory, after all, is that certain complex phenomena are intrinsically unpredictable because of sensitivity to initial conditions. Minds and all their byproducts, such as economies, are prime examples of phenomena subject to this so-called butterfly effect. That brings me to Joao Teixeira’s comments. He says it is not clear what I expect of mind-science. Actually, the subtitle of my book summarizes what I see as mind-science’s major goals: explanation of the mind, replication of its functions and treatment of mental illness. None of these goals has been achieved or is even
254
JOHN HORGAN
in sight. Teixeira seems upset that I single out mind-science for criticism, when other realms of science are just as deserving of approbation. He wonders: “Why are we supposed to believe that economics is more scientific than psychology?” Actually, as I said above, I believe economics is just as weak as mind-science, and for some of the same reasons. But I can’t accept Teixeira’s comparison of mindscience to other scientific enterprises, such as medicine in general. Where is mindscience’s smallpox vaccine, penicillin, sterile surgery, anesthesia? Teixeira rightly notes that physics has bequeathed us some frustrating paradoxes, such as particlewave complementarity and the uncertainty principle. But physics has also yielded microchips, lasers, thermonuclear weapons and insights into the entire cosmos, from quarks to quasars. No wonder mind-scientists have physics envy! If, however, Teixeira is implying that I hold mind-science to a higher standard than other fields, let me assure him that he is quite right. No science – not even nuclear physics – has a greater capacity for harm than the science that seeks to tell us what we are, what we can be and even what we should be. Stan Franklin is obviously confused by my critique of mind-science, so confused that in his attempts to refute me he often ends up agreeing with me. “Neuroscience is much, much harder than physics,” he informs us. Yes, I made that very point in my essay and book. I fear I will compound Franklin’s befuddlement by addressing his remarks, but here goes. He says that rejecting mind-science because psychotherapy is ineffective is like rejecting physics because we don’t have clean, cheap energy. But as I commented above, mind-science has failed on all major fronts, intellectual and practical, whereas physics has achieved extraordinary successes. Franklin declares that the missteps of human genetics are signs of that field’s vigor and health. He divulges his delight as a mathematician when he ends up with contradictory results. But when Franklin botches a proof presumably no one ends up maligned – like blacks tarred as innately inferior by intelligence researchers – or dead – like Jesse Gelsinger, who recently died in a reckless gene-therapy trial. In an effort to rebut my contention that mind-science has made little progress on a purely theoretical level, Franklin rattles off myriad models of cognition. He fails to realize that this surfeit of theories is one of the most striking symptoms of mindscience’s lack of genuine progress. Mind-science remains eternally mired in what Thomas Kuhn called the pre-scientific stage, when researchers are so baffled that they cannot agree on a common paradigm. By the way, Franklin should know that the neuroscientist Walter Freeman, with whose work I am familiar, does not claim to have solved the binding problem or any other problem with his chaotic model of cognition. Freeman says that his model merely indicates one possible mechanism whereby certain types of binding might occur. Freeman once told me that he would be surprised if his model turned out to be right. If only all mind-scientists were so modest! In defending AI, Franklin notes that it took “almost forty years of effort by some hundreds of computer scientists and computer engineers” to produce a chess-
JOHN HORGAN RESPONDS
255
playing computer that could beat the best human in the world. Moreover, computers still cannot play the more complex game Go as well as beginners. Exactly! And AI’s difficulty in mastering these narrowly defined games suggests that mimicking human intelligence in all its complexity and versatility will be extremely difficult and perhaps impossible. Franklin proclaims that the solution of a mathematical conjecture by a theorem-solving program produced at Argonne laboratory is AI’s “major accomplishment.” Franklin may be enthralled by Argonne’s program, which I have written about for Scientific American, but the major mathematicians I have spoken to are unimpressed. That is why no one has yet been able to claim the $100,000 Leibniz Prize – created by Edward Fredkin in the early 1980’s and administered by Carnegie Mellon – which goes to the first AI program that produces a “profound” mathematical result. Frank Jackson begins his defense of AI by reciting the ancient dogma that dualism is false; we are machines. Although we are very complicated and made of flesh and blood rather than silicon, all of our functions should be transferable to non-biological machines, such as computers or robots. I accept all this, but AI’ers have been mouthing this dogma for 50 years now and they have little to show for it. Jackson goes on to discuss more subtle theoretical issues raised by AI. Do artificial minds need an evolutionary history? Must they be shaped by some selectional process? How can they deal with the infinite subtleties of context that arise when confronting most real-world problems? Again, Jackson’s goal seems to be to show that there are no absolute, theoretical, in-principle obstacles to AI. AI is possible. But again, the relevant question is not, Can AI succeed in principle? but, Why has AI failed so far in practice? Why does it succeed in certain narrowly defined realms, such as chess, and fail in others, such as natural-language comprehension? Jackson himself alludes to some potential obstacles to AI, and his discussion suggests that certain approaches may work better than others at overcoming these obstacles. Neural networks, for example, which acquire intelligence by interacting with the world, may fare better than rule-based programs, whose intelligence is front-loaded into them. In my book, I suggest some reasons for AI’s failures. Sensitivity to context is one rather obvious technical challenge. AI’ers have also been slow to learn the importance of environmental interaction and emotion in shaping human intelligence. But I do not claim that any of these problems are show-stoppers. In my opinion, arguing that AI is impossible – by appealing to Godel’s theorem, for example, as Roger Penrose and others have done – is as futile as arguing that AI is possible. While I see context as a huge hurdle for AI, I certainly do not subscribe to what Jackson calls “radical contextualism,” which implies that not only machine intelligence but even the human kind is impossible. I stick to the empirical facts, pointing out what AI’ers have promised and what they have actually delivered. Based on the abysmal record of AI thus far in mimicking general intelligence, I predict that AI will not soon or perhaps ever realize its ultimate goal of equaling
256
JOHN HORGAN
or surpassing general human intelligence. No theoretical argument can prove me wrong, only the accomplishment itself. Like Stan Franklin, Valerie Hardcastle corroborates my argument more than she refutes it. In fact, she is even more pessimistic than I am about the state of mindscience. She thinks that mind-science is so fantastically difficult and complicated that you must read every relevant journal article and talk to every scientist before you are even qualified to voice an opinion on it! No one can possibly comment on the forest of mind-science before he or she has scrutinized every last tree. Hardcastle implies that she fulfills these daunting qualifications, and what conclusion has she reached about mind-science? The same conclusion that I reached! The mind is a mystery! Hardcastle lectures me on how my fuzzy, romantic vision of science has nothing to do with science in the trenches. Actually, as someone who has been reporting not only on mind-science but on virtually all realms of science, technology and medicine for almost 20 years, I am at least as aware as Hardcastle of how messy things can get as science stumbles toward truth. Hardcastle is so engrossed in selfaggrandizing posturing that she never comes to grips with the question that her own work raises: Why do some fields, such as nuclear physics, manage to arrive at truth in spite of all the obstacles to progress while other fields, especially mindrelated ones, keep spinning their wheels? One problem may be what Joe Levine has called the “explanatory gap” between materialistic theories involving neurons and neurotransmitters and so on and the subjective phenomena to which they refer. Hardcastle accuses me of misunderstanding Levine’s term. Actually, I have read Levine’s original essay and talked to him about it; I know precisely how he intended the term. I am also familiar with the subtly different ways the term has been used by Christof Koch, David Chalmers and others. My job as a science journalist is to take a rather technical philosophical concept and translate it into language that lay people can grasp, an enterprise that Hardcastle obviously disdains. My anecdote about Patricia Goldman-Rakic in The Undiscovered Mind is intended to show that the explanatory gap is not just an esoteric philosophical concept; it underlies the almost visceral sense of dissatisfaction that I and many other observers have when confronted with neuroscientific models of cognition. John Symons calls me “a lover of mystery” who is “happy” that mind-science “isn’t going to live up to the more outrageous hype any time soon.” I doubt that anyone who has read my book would say that I am sanguine about mind-science’s current ability – or inability – to deal with mental illness. I am horrified that psychiatrists are prescribing SSRI’s and other powerful drugs for kids as young as three years old when there is no evidence that these drugs help – or do not harm – children. I would be delighted to see neuropsychiatry advance beyond the level of bloodletting and leeches. Symons suggests that the obstacles to progress in mind-science are not as serious as I have indicated. Regarding variability, he points out that human brains are uniform in many respects, particularly on the molecular and cellular levels.
JOHN HORGAN RESPONDS
257
True, but of course what makes human minds so resistant to reductionist explanations is that each mind is unique, even though all are made of the same components. Symons points out that variability arises in all fields of science. True again, but Symons adds that “the situation [in mind-science] is not qualitatively different from that of any other science.” Is Symons kidding? The variability of human brains and minds poses profound scientific, philosophical, medical and ethical questions. Does Symons seriously believe that the variability of stars presents astronomers with remotely comparable challenges? In discussing the binding problem, Symons accuses me of suggesting that binding must take place in a particular spot in the brain; I am thus presenting an updated version of Descartes’s old pineal-gland theory. Actually, I assume along with Christof Koch and others that binding is a process that is probably distributed throughout the brain. Symons notes that “there is no shortage of plausible solutions” to the binding problem. At the risk of belaboring this point, let me reiterate that abundance of theories is not a sign of progress. Quite the contrary. In discussing the explanatory gap, Symons reveals himself to be a Dennettista, someone who believes that the explanatory gap does not really exist because consciousness does not really exist. Philosophy has produced many dumb ideas over the millennia, but this is surely one of the dumbest. Symons refers to “faith in the reality of qualia . . .” Faith? All we know with certainty are our qualia! Everything else we take on faith. Symons ends by accusing me of being antiscience, because I supposedly believe that science has “demystified and therefore devalued human life and the natural world.” Actually, as I have stated publicly many times, I became a science writer because I view science as the most noble and meaningful of all human creations. What I am against – and what I have devoted my career to exposing and fighting – is bad science, lies or half-truths masquerading as scientific truth. This pseudo-science – whether it is social Darwinism, eugenics, totalitarian Marxism, or biopsychiatry – not only limits and demeans us; it can quite literally kill us. I must say that after the overt condescension displayed by several previous commentators, Khalfa’s critique was a tonic. He makes some legitimate points, for example, when he says that I do not show the same skepticism toward Chomsky’s evolutionary argument for mysterianism that I show toward other evolutionary theories. But as Khalfa himself recognizes, my critique of mind-science is largely inductive and empirical rather than theoretical. I look at how far the claims made for mind-science fall short of their actual accomplishments and speculate on how much we can expect in the future. Although I predict that certain aspects of the mind may prove to be scientifically intractable, I acknowledge that I might turn out to be wrong. My goal is not to damn mind-science but to show just how difficult and dangerous it can be. In this way I hope to foster healthy modesty among the purveyors of mind-science and healthy skepticism among lay consumers. Khalfa regrets that I do not dwell a bit more on theoretical issues, if only to show the extent to which two different approaches, such as cognitive science and
258
JOHN HORGAN
psychoanalysis, contradict or complement each other. But the history of mindscience shows that theory often obfuscates more than it clarifies. I decided early on in writing my book that a blunt, hard-nosed method of assessment was needed to cut through all the theoretical hand-waving and show what has really been accomplished thus far. Theory aside, does this approach work? Khalfa wonders just how much the persistence of old paradigms such as psychoanalysis indicates lack of progress in mind-science. But can he cite another realm of science that exhibits this syndrome to such a striking degree? I barely mentioned behaviorism in The Undiscovered Mind, because I assumed that it really is dead. But in The New Behaviorism (Psychology Press, 2001), John Staddon of Duke University argues persuasively that behaviorism has been unfairly slighted – especially given the poor record of alternative paradigms over the past half-century – and deserves a second chance. The equivalent in cosmology would be a book arguing for reconsideration of the old steady state theory, which was routed by the big bang theory of the universe more than 30 years ago. (Fred Hoyle and a few other die-hards still tout steady state models but no reputable cosmologists do.) Khalfa worries that my method of criticism reduces all theories to the same level and thus neglects the possibility that some may have contributed more to our understanding than others. He notes that cognitive science, if nothing else, has revealed the sheer complexity of even the simplest mental operations. Actually, in The Undiscovered Mind I make this very point about neuroscience. Neuroscience keeps revealing more complexity within the brain, but the paradoxical result is that a unified theory of the brain seems more distant than ever. Lest I seem overly pessimistic, let me mention one potential advance that could lead to precipitous, revolutionary progress on many fronts of mind-science: the deciphering of the brain’s neural code, the “language” whereby neurons absorb, store and communicate information. This discovery would surpass even the genetic code in intellectual and practical significance. I hope to live long enough to witness the discovery of the neural code, although I must admit I am disturbed at its potential for harm as well as good. Finally, I am gratified that Khalfa took the time to read my critique of John Searle’s Chinese room argument, which was hidden in the endnotes of The Undiscovered Mind. Since Khalfa has apparently misunderstood my critique, let me try to clarify it. Searle thinks he has shown that computers do not really understand anything; they just dumbly carry out operational rules. They are like a man who does not understand Chinese but who has a book of rules for translating a string of Chinese characters representing a question into another string representing an appropriate response to the question. In my view, Searle’s experiment leads not to the conclusion he thinks it does but to two different, more subtle conclusions: first, temporal factors are crucial to our definition of understanding; second, AI has a long way to go before it will pass the Turing test for language comprehension. Here is why: A rule-book that would allow a non-Chinese-speaking person to respond appropriately to questions in Chinese would have to be massive, because
JOHN HORGAN RESPONDS
259
of the aforementioned context problem. In the real world, the man in the room would quickly get bogged down trying to figure out how to translate an incoming symbol-string into an appropriate response. Only someone who actually understands Chinese could respond to questions promptly and intelligently enough to mimic a Chinese-speaker! To put it another way, that man understands Chinese by definition. Unlike Searle, I am willing to grant that a HAL-like computer – a computer that can carry on intelligent, timely, complex conversations just like a real human – really understands language. But of course no current computer can mimic any ordinary human’s language comprehension, because of the context problem, etc. I hope this clarifies matters for Khalfa. Of course, given the contentious history of the debate over the Chinese room experiment, clarification to everyone’s satisfaction is unlikely. This mind-related puzzle, like so many others, resists a final solution. But if we solved all of our puzzles, what would we do then?