2023/09/01 by Matthew L. Jones, Matthew Jones · 9 citations
Medicine · Social Sciences · #American history #Ancient history #Art history #Classics #Columbia university #Computational and Text Analysis Methods #Computer science #Family history #History #Internal medicine #Library science #Media studies #Medicine #Sociology
paper · doi:10.1093/ahr/rhad361
published in The American Historical Review 128(3), 1360-1367 (Oxford University Press)
openalex publication_date 2023/09/01 · openalex created_date 2023/09/28 · openalex updated_date 2026/07/29
Late in 1982, Edinburgh professor Donald Michie explained the fundamental error that plagued earlier efforts to create artificial intelligence. “The inductive learning of concepts, rules, strategies, etc. from examples is what confers on the human problem-solver his power and versatility, and not (as had earlier been supposed) power of calculation.”1 A minority position in 1982, learning from examples came to dominate artificial intelligence early in the new millennium. In a key 2009 manifesto celebrating the “unreasonable effectiveness of data,” three Google researchers argued “sciences that involve human beings rather than elementary particles have proven more resistant to elegant mathematics.” We should “embrace complexity and make use of the best ally we have: the unreasonable effectiveness of data.”2 The computer scientist John McCarthy coined the term “artificial intelligence” originally in search of funding; in the mid 2010s, the term was dramatically redefined to describe large-scale algorithmic decision-making systems and predictive machine learning “trained” on massive data sets.3 Through most of the Cold War and beyond, AI researchers focused on “symbolic AI” largely ignored data collected from everyday and military activities.4 Such empiricism of the quotidian paled in prestige in comparison with logic and numerical computation and the more empirically oriented approaches such as neural networks and pattern recognition were widely lambasted.5 Learning from data seemed to be the wrong approach for producing intelligence or intelligent behaviors. Alongside this symbolic approach, in the USA, USSR, and beyond, a far less prestigious empiricist stratum developed comprising congeries of techniques for dealing with large-scale military, intelligence, and commercial data. Our contemporary world of AI, with its often-biased algorithmic decision system, owes far more to this empirical strand of inquiry than to the previously higher status and much studied symbolic artificial intelligence.6