The history of alchemy provides an instructive analogy for AI: like alchemists who, buoyed by partia… — Hubert L. Dreyfus

The history of alchemy provides an instructive analogy for AI: like alchemists who, buoyed by partial successes such as distilling quicksilver, persisted for centuries in a fundamentally impossible quest to transmute metals, AI researchers can always justify continuing work despite repeated failures; yet if we attend to predictions versus results, empirical evidence shows current efforts are misdirected at the information‑processing level, and real progress requires stepping back to investigate the deeper structure of the problem rather than pouring more resources into the same paradigm.

By Hubert L. Dreyfus, from What Computers Can't Do

Key Arguments

Dreyfus recalls that 'Alchemists were so successful in distilling quicksilver from what seemed to be dirt that, after several hundred years of fruitless efforts to convert lead into gold, they still refused to believe that on the chemical level one cannot transmute metals.'
He draws a parallel: 'They did, however, produceas by-productsovens, retorts, crucibles, and so forth, just as computer workers, while failing to produce artificial intelligence, have developed assembly programs, debugging programs, program-editing programs, and so on, and the M.I.T. robot project has built a very elegant mechanical arm.'
On this basis he warns, 'To avoid the fate of the alchemists, it is time we asked where we stand. Now, before we invest more time and money on the information-processing level, we should ask whether the protocols of human subjects and the programs so far produced suggest that computer language is appropriate for analyzing human behavior.'
He formulates the key questions: 'Is an exhaustive analysis of human reason into rule-governed operations on discrete, determinate, context-free elements possible? Is an approximation to this goal of artificial reason even probable? The answer to both these questions appears to be, No.'
Anticipating resistance, he concedes that 'Artificial Intelligence workers who feel that some concrete results are better than none, and that we should not abandon work on artificial intelligence until the day we are in a position to construct such artificial men, cannot be refuted. The long reign of alchemy has shown that any research which has had an early success can always be justified and continued by those who prefer adventure to patience.20*'
He notes the limitations of purely formal impossibility proofs (e.g., Gödel), remarking that 'such proofs are irrelevant to AI,' and suggests that the proper check is empirical: 'If, however, one accepts empirical evidence as to whether the effort has been misdirected, he has only to look at the predictions and the results.'
He adds that 'Even if there had been no predictions, only hopes, as in language translation, the results are sufficiently disappointing to be self-incriminating,' underscoring that performance has fallen far short even of modest aspirations.
He concludes with a counterfactual about alchemy: 'If the alchemist had stopped poring over his retorts and pentagrams and had spent his time looking for the deeper structure of the problem, as primitive man took his eyes off the moon, came out of the trees, and discovered fire and the wheel, things would have been set moving in a more promising direction,' implying that AI should likewise redirect its efforts toward deeper theoretical understanding rather than more of the same techniques.

Source Quotes

And this way of being-in-a-situation turns out to be unprogrammable in principle using presently conceivable techniques. Alchemists were so successful in distilling quicksilver from what seemed to be dirt that, after several hundred years of fruitless efforts to convert lead into gold, they still refused to believe that on the chemical level one cannot transmute metals. They did, however, produceas by-productsovens, retorts, crucibles, and so forth, just as computer workers, while failing to produce artificial intelligence, have developed assembly programs, debugging programs, program-editing programs, and so on, and the M.I.T. robot project has built a very elegant mechanical arm.

Alchemists were so successful in distilling quicksilver from what seemed to be dirt that, after several hundred years of fruitless efforts to convert lead into gold, they still refused to believe that on the chemical level one cannot transmute metals. They did, however, produceas by-productsovens, retorts, crucibles, and so forth, just as computer workers, while failing to produce artificial intelligence, have developed assembly programs, debugging programs, program-editing programs, and so on, and the M.I.T. robot project has built a very elegant mechanical arm. To avoid the fate of the alchemists, it is time we asked where we stand.

To avoid the fate of the alchemists, it is time we asked where we stand. Now, before we invest more time and money on the information-processing level, we should ask whether the protocols of human subjects and the programs so far produced suggest that computer language is appropriate for analyzing human behavior: Is an exhaustive analysis of human reason into rule-governed operations on discrete, determinate, context-free elements possible? Is an approximation to this goal of artificial reason even probable?

To avoid the fate of the alchemists, it is time we asked where we stand. Now, before we invest more time and money on the information-processing level, we should ask whether the protocols of human subjects and the programs so far produced suggest that computer language is appropriate for analyzing human behavior: Is an exhaustive analysis of human reason into rule-governed operations on discrete, determinate, context-free elements possible? Is an approximation to this goal of artificial reason even probable? The answer to both these questions appears to be, No. Does this mean that all the work and money put into artificial intelligence have been wasted? Not at all, if instead of trying to minimize our difficulties, we try to understand what they show.

Artificial Intelligence workers who feel that some concrete results are better than none, and that we should not abandon work on artificial intelligence until the day we are in a position to construct such artificial men, cannot be refuted. The long reign of alchemy has shown that any research which has had an early success can always be justified and continued by those who prefer adventure to patience.20* If researchers insist on a priori proof of the impossibility of success, one can at best use formal methods such as Gödel's to prove the limitations of formal systems, but such proofs are irrelevant to AI.21* Researchers could in any case respond that at least the goal can be approached. If, however, one accepts empirical evidence as to whether the effort has been misdirected, he has only to look at the predictions and the results.

The long reign of alchemy has shown that any research which has had an early success can always be justified and continued by those who prefer adventure to patience.20* If researchers insist on a priori proof of the impossibility of success, one can at best use formal methods such as Gödel's to prove the limitations of formal systems, but such proofs are irrelevant to AI.21* Researchers could in any case respond that at least the goal can be approached. If, however, one accepts empirical evidence as to whether the effort has been misdirected, he has only to look at the predictions and the results. Even if there had been no predictions, only hopes, as in language translation, the results are sufficiently disappointing to be self-incriminating.

If, however, one accepts empirical evidence as to whether the effort has been misdirected, he has only to look at the predictions and the results. Even if there had been no predictions, only hopes, as in language translation, the results are sufficiently disappointing to be self-incriminating. If the alchemist had stopped poring over his retorts and pentagrams and had spent his time looking for the deeper structure of the problem, as primitive man took his eyes off the moon, came out of the trees, and discovered fire and the wheel, things would have been set moving in a more promising direction.

Even if there had been no predictions, only hopes, as in language translation, the results are sufficiently disappointing to be self-incriminating. If the alchemist had stopped poring over his retorts and pentagrams and had spent his time looking for the deeper structure of the problem, as primitive man took his eyes off the moon, came out of the trees, and discovered fire and the wheel, things would have been set moving in a more promising direction. After all, three hundred years after the alchemists we did get gold from lead (and we have landed on the moon), but only after we

Key Concepts

Alchemists were so successful in distilling quicksilver from what seemed to be dirt that, after several hundred years of fruitless efforts to convert lead into gold, they still refused to believe that on the chemical level one cannot transmute metals.
just as computer workers, while failing to produce artificial intelligence, have developed assembly programs, debugging programs, program-editing programs, and so on, and the M.I.T. robot project has built a very elegant mechanical arm.
before we invest more time and money on the information-processing level, we should ask whether the protocols of human subjects and the programs so far produced suggest that computer language is appropriate for analyzing human behavior:
Is an exhaustive analysis of human reason into rule-governed operations on discrete, determinate, context-free elements possible? Is an approximation to this goal of artificial reason even probable? The answer to both these questions appears to be, No.
The long reign of alchemy has shown that any research which has had an early success can always be justified and continued by those who prefer adventure to patience.20*
If, however, one accepts empirical evidence as to whether the effort has been misdirected, he has only to look at the predictions and the results.
Even if there had been no predictions, only hopes, as in language translation, the results are sufficiently disappointing to be self-incriminating.
If the alchemist had stopped poring over his retorts and pentagrams and had spent his time looking for the deeper structure of the problem, as primitive man took his eyes off the moon, came out of the trees, and discovered fire and the wheel, things would have been set moving in a more promising direction.

Context

Near the end of 'The Limits of Artificial Intelligence', Dreyfus develops an extended analogy between the history of alchemy and the trajectory of AI, using it to argue that early partial successes can mask deep theoretical impossibility at a given level of description and that empirical disappointments in AI should prompt a shift toward investigating the deeper structure of human reason and world‑relation.