“Artificial Intelligence and its Discontents": A Comment and Rejoinder
K. Vela Velupillai writes:
‘It was typical for [Alan Turing] … to seek to outdo Bell Telephone Laboratories with his single brain, and to build a better system with his own hands.’ (Hodges, 2008, p. 4)
Venu Madhav Govindu (henceforth VMG for simplicity) has written two articles titled “Artificial Intelligence and its Discontents” (13 August and 20 August, 2021). The two articles (I shall refer to them as I and II) are both interesting, informative, and mildly provocative. I broadly agree with VMG’s assessment of AI’s power and potential – especially on the lives of downtrodden people in the less developed countries (LDCs). My disagreement is with his harnessing of economics, philosophy, mathematics, politics – and even sociology, whereas he is ‘a computer vision researcher and Associate Professor at the Indian Institute of Science, Bengaluru.’
Apart from the contentious statement that the ‘popular Turing test’ tells us ‘nothing about the attributes of intelligence or the means of achieving them’ (Note 2, in I)—in this VMG echoes John Searle’s scepticism—VMG does not mention Alan Turing at all, the “father” of Mechanical Intelligence or MI (although there are many ancestors for MI and AI).
Firstly, Artificial Intelligence (AI) is a word coined by John McCarthy, for the proposal of the Dartmouth conference of 1957, in contradistinction to cybernetics (cf. Mitchell, 2019, chapter 1), whereas Turing did not use “artificial” in his definition of MI. Secondly, intelligence of machines is used in the context of ‘perfect information games’ (Sadler & Regan, 2019, p. 63), or Arithmetical Games, also by AlphaZero and AlphaGo (by DeepMind). Thirdly, Turing, too, developed intelligence for machines in the context of (arithmetical) games (particularly in Turing, 1992, pp. 107.127, chapter on Intelligent Machinery, 1948 and Turing, 1950). From all these points of view, VMG is deficient in his “History” and the section on “Intelligence” (I, pp. 2-3; and I, pp. 3-4).
VMG’s statement on Keynes (II, p.6) is part of his dogmatic stance (how does he know, even if he is a trained economist, that Keynes was ‘brilliant’ and an ‘old fashioned imperialist’; there are many reputable scholars who dispute both these assertions); so is his attempt at defining ‘utility’ in terms of ‘objectives’ (I, p. 4). Incidentally, Keynes’ prediction ‘for the progressive countries’ in 100 years, ranks with Turing’s (1950, p. 442): both were for the developed countries, which were then located in Europe and (North) America. Now, it is different, with the inclusion in this group, of Singapore, (South) Korea, Taiwan, and so on.
VMG talks about ‘the sudden rise of the idea of autonomous driving of vehicles’ (II, p.2; italics added). It was not ‘sudden’. In fact, it rose almost together with ‘the Fordian ideology of the personal car’ (II, p. 2), which was a result of the Carnot cycle, the rise of thermodynamics and the phenomenological basis of the second law of thermodynamics (although it is alleged that personal cars antedated even these).
Although I do like, even approve, of VMG’s many strictures against power, fairness, and efficiency in discussing inequity and inequality in all societies, I don’t think it should be AI algorithms to confront them. In fact, it is not the business of any kind of algorithms to be the strawman in the fight against technology and automation. Algorithms are (mathematical) methods, and tools, for implementing decision systems. VMG must be aware that none of standard (or orthodox) economics is algorithmic (not even Herbert Scarf’s so-called ‘algorithm’ which depends on, inter alia, the Bolzano-Weierstrass theorem and Sperner’s lemma for which no implementable method can be devised).
In my opinion, VMG depends too much on Kate Crawford, who is not a computer scientist at all. VMG’s criticism of social power, especially in India, depends on Crawford’s views, and they are tendentious. For example, his (historical) description of Jeremy Bentham and the design of the panopticon, and the role of Samuel Bentham in the construction, is directly attributable to Crawford (this is similar to the view VMG has, of Keynes). It is, at least, disputable given Jeremy Bentham’s enlightened view of prisoners and how they should be treated. To invoke this in the Indian context is not fair – a word and a concept that VMG likes to use, but here I am with him!
To endorse machine intelligence in arithmetical games, is one thing; to approve of the evils of automation on the lives of people – whether they be underprivileged, for whatever reasons of status or colour – is quite another. To be in favour of machines is not to endorse automation at any cost, as Shannon said in an Interview with Liversidge (Shannon, (1993), p. xxx):‘ [W]e are basically machines, but a very complex type.’
Let me end this with an observation by Turing at the end of his paper in the journal MIND (Turing, 1950, p. 460): ‘We can only see a short distance ahead, but we can see plenty there that needs to be done.’
References:
Crawford, Kate (2019), Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence, Yale University Press, New Haven, USA.
Hodges, Andrew (2008), Alan Turing, Logical and Physical, pp. 3-15, in New Computational Paradigms – Changing Conceptions of What is Computable edited by S. Barry Cooper, Benedikt Löwe and Andrea Sorbi, Springer Science+Business Media, LLC, New York, USA.
Michell, Melanie (2019), Artificial Intelligence – A Guide for Thinking Humans, Farrar, Strauss & Giroux, New York, USA.
Sadler, Matthew & Natasha Regan (2019), Game Changer, New In Chess, Alkmaar, The Netherlands.
Shannon, Claud Elwood (1993), Collected Papers, edited by N. J. A. Sloane & Aaron D. Wyner, John Wiley & Sons, Inc., NJ, USA.
Turing, A. M. (1950), “Computing Machinery and Intelligence”, MIND, Vol. LIX, No. 236, October, pp. 433-460.
Turing, A. M. (1992), Mechanical Intelligence, Collected Works of A. M. Turing, edited by D. C. Ince, North-Holland, Amsterdam, The Netherlands.
Venu Madhav Govindu writes:
Dr. Velapillai finds my two-part essay 'Artificial Intelligence and its Discontents' interesting and informative. However, given his ad hominem strictures, it would seem that it was more than “mildly provocative”.
The history of AI is extensive and its coverage in my essay is necessarily limited to ideas that are germane to the discussion. Someone au fait with modern AI could have rebuked me for other elisions, especially inductive learning, statistical learning theory, empirical risk minimization and the like. Pointing to the problems with the Turing test does not amount to a rejection of Turing's body of work on machine intelligence. Moreover, much of the focus of contemporary AI is not on arithmetic games. I am sure that there are deficiencies in the essay, but not in the sense that Dr. Velapillai would suggest.
Dr. Velapillai queries: “how does he know, even if he is a trained economist, that Keynes was ‘brilliant’ and an ‘old fashioned imperialist’” and asserts that it is part of my “dogmatic stance”. It would have been useful if he had given his reasoning instead of referring to the unstated opinions of unnamed scholars of repute. In any event, most trained economists are entirely innocent of the role of Keynes in determining colonial India's monetary policy.
Briefly, during the inter-War years, Keynes consistently and influentially endorsed the artificially high sterling-rupee ratio that harmed India's economic interests. Designed to prevent India from absorbing bullion, inter alia, this policy paid rich dividends for Britain when it abandoned the Gold Standard in 1931 while retaining the rupee's linkage to sterling. The result was the deepening of India's agrarian crisis and the flight of 'distress gold' to London. During the Bretton Woods negotiations, Keynes went to extraordinary lengths to prevent the Sterling Credits owed by Britain to India and Egypt from becoming a matter of multi-lateral negotiations (Dhar, Bimalendu. 1956. The Sterling Balances of India. Nababharat Publishers. p. 92).This greatly compromised India's position in the subsequent settlement (See, for instance, Amiya Kumar Bagchi's essay 'Keynes, Kaldor and Economic Development' and the references therein. Published in Jomo, K. S. (ed.). 2005. The Pioneers of Development Economics: Great Economists on Development. Tulika Books).
More broadly, in his canonical three-volume biography, Robert Skidelsky states that Keynes “assumed the Empire as a fact of life and never showed the slightest interest in discarding it” (Skidelsky, Robert. 1983. John Maynard Keynes: Hope Betrayed, 1883 – 1920. Macmillan Press. p. 91.). Similarly, the editors of a retrospective volume on the Keynes' essay in question point out: “The least we can say is that he had a manifest ethnocentric view and that he did not pay much attention to the destiny of the rest of the world” (Lorenzo Pecchi and Gustavo Piga (ed.),. 2008. Revisiting Keynes: Economic Possibilities for our Grandchildren, The MIT Press. p. 4.).
I agree that endorsement of algorithms is not the same as approval of the evils of automation. But when intellectual ideas – be they from economists or AI researchers – are deployed in the real-world, it is very much the business of sociologists and others to examine their assumptions as well as the consequences. A doctrinaire refusal to contend with their legitimate criticism is certainly part of the problem we are faced with. Further, I am puzzled by the argument that “it is not the business of any kind of algorithms to be the strawman in the fight against technology and automation.” In the case of AI, automation manifests in the physical world in the form of algorithms.
I am unable to see the connection drawn between autonomous driving and the second law of thermodynamics. Finally, it is stated that my argument “in the Indian context is not fair”. But 'fairness' is used in my essay in its computer science sense and not that of its common usage.
--Venu Madhav Govindu, Indian Institute of Science, Bengaluru