The AI Delusion

This refreshing, amusing and frank book dispels many myths about the nature of AI when compared with human intelligence, with a stimulating range of examples. The introduction shows how Barack Obama successfully used data mining in his 2008 presidential campaign, while Hillary Clinton's approach failed in 2016, partly by not applying common sense and trusting the selective findings of an algorithm with notable blind spots. As the cover states upfront: 'while computers are very good at discovering patterns in data, they are useless at judging how best to apply them in the real world because they lack human wisdom and common sense.' They are obedient rather than intelligent and do not understand meaning. The author draws a revealing parallel between computer functioning and the feat of Nigel Richards who memorised the 386,000 words of French Scrabble and won the French language Scrabble World Championship despite not understanding a word of French! So computers are more useful than intelligent, despite well-publicised examples of beating world champion board game players. As the author points out, 'computers do not know what words mean because computers do not experience the world the way we do… they do not have the common sense or wisdom that humans accumulate by living life.' (p. 236)

A further crucial point also illustrated with many telling examples concerns the limits of data mining in view of the fact that when statistical models analyse large number of potential explanatory variables, the number of possible relationships not only becomes astonishingly large, but, invariably, some random combinations will be correlated with what we are trying to predict, thus tempting analysts into seeing real rather than artificial patterns. In one case, the author manufactured a blood pressure study based on random variables, which still showed up significant patterns – of course, these also apply to real-world studies, so one should be cautious about drawing conclusions or mistaking coincidence and correlation for causation. The same applies to forecasting stock-market performance. Algorithms are frequently used in the process of job applications but again, real human judgement has to have the last word. Overall, book is a cautionary tale delimiting the strengths and limitations of computer analysis.