Thursday Morning Talk “Work in Progress”: Dimitri Coelho Mollo (SCIoI), “Modelling Intelligence: the good, the bad, and the plural”
Abstract: I argue that artificial intelligence research has been both fuelled and hindered by the use of ‘model tasks’, that is, tasks the solution of which are taken to be sufficient for, or at least indicative of intelligence. Before AI proper, cybernetics explored model tasks involving basic real-time and world-involving action control aimed at the maintenance of homeostasis, an approach echoed more recently by the embodied AI movement. Logicist AI, in contrast, took as model tasks for intelligence the solution of abstract problems, such as theorem-proving and proficiency in combinatorially complex games, chess having pride of place. Connectionist AI – including the current deep learning wave – despite privileging model tasks tied to learning from ‘experience’, shares this focus on abstract, disembodied behaviours as key to intelligence, with particular effort being done in language processing, categorisation, and combinatorially complex games, such as Go. Reliance on model tasks has led to considerable progress in solving those specific tasks, but against expectation they did not lead to theoretical insights about the nature of intelligence in general, and how to build it. This outcome, I argue, is in part due to the failure of recognising the limited scope of model tasks, as well as the abstractions and idealisations of real-world intelligent behaviour that they embody. All mainstream frameworks in AI research, in brief, focus on circumscribed, idealised models of intelligent behaviour, those for which the respective approaches tend to generate cumulative progress and satisfactory solutions. Such models, however, abstract or idealise away important features of intelligence, and, if unchecked, close off potentially rewarding paths of research. Bringing to the fore the limitations tied to such model task choices, as well as the abstractions and idealisation involved in each, I argue, opens the way for a more integrative and plural approach to AI.