We are all surrounded by other human beings whose thoughts we cannot see but to whom we extend the benefit of the doubt (grudgingly at times) that they, like ourselves, are thinking beings. We do this partly on the basis of metaphysics, ethics, and upbringing, but also from observing our fellow humans’ behavior. We are all also surrounded by machines whom we generally do not regard as thinking beings. Again, we do this partly on the basis of metaphysics, ethics, upbringing, etc., but also partly on the basis of observations of how machines behave (or, if you think “behave” is too anthropomorphic, how machines respond to stimuli and interact with their environments). There has long been discussion as to what machine behaviors might lead us to extend the benefit of the doubt to machine intelligence. Alan Turing’s 1950 “test” was one proposal for how we might use conversation (and, interestingly, deception) as a basis for evaluating machine intelligence. Ever since, there have been a number of claims to have passed the Turing test, at least under certain conditions. We are now at a point where most of the students and staff in this course have probably had a “conversation” with a machine, perhaps even without knowing it. Yet few of us would say that, for example, Siri is a thinking being. Thus, we are at an interesting point in history when both the achievement of AI and the limits of the Turing test are becoming visible even to quite ordinary people.

If the classic Turing test doesn’t quite achieve what Turing hoped, therefore, then maybe we can replace it, or – more likely – supplement it with new tests. Your task in this assessment is to create a new Turing test. That is, you should propose some task which, if a machine could do that task, would cause us to extend it the benefit of the doubt and regard it as a thinking being. This could be a largely intellectual task: for instance, if a computer could write a book proposing a new theory for the origins of the First World War such that historians would regard that book as a significant contribution to their field, well, then most historians would have to regard that machine as a thinking being. Alternatively, the task could be more oriented to know-how or to creative expression. If, for instance, a machine could devise a new recipe for a soufflé, and could cook that soufflé to a standard that impressed a panel of chefs, well, then, those chefs would at least have to tip their toques to their new machine colleague.

You should propose, explain, and defend your test in an essay of 1000 words. The essay should precisely lay out the conditions of the test and of the evaluation of the machine’s performance. You should explain how this test helps us better understand and gauge machine intelligence, while also reflecting on the shortcomings – and possible “hacks” – of your test.