Can 'Digital Computers' Be Surprised by Us? Rescuing Lovelace’s Originality Insight From Turing’s Hasty Dismissal

Abstract

This paper problematises the assumptions presiding over current endeavours to optimise the learning capacity of a system through 'surprise constraints'. The latter tend to overlook the need to distinguish between different types of surprises. This kind of oversight is not new. When Turing dismisses the 'Lovelace objection' by pointing out that `machines take [him] by surprise with great frequency’, he only considers trivial, downstream interpretation surprises. The latter are trivial because calculation failures are expected. They also result entirely from the recipient’s downstream interpretation. In contrast, some surprises may be said to be ‘co-produced’, in that they are also the result of some upstream endeavour to ‘originate’ something (a ‘pretension’ which current machines lack just as much as the Analytical Engine considered by Lovelace).

Had Turing considered different kinds of surprises, he might have concluded that, to be a fruitful demarcation criterion, his 'surprise question’ needs reversing: can computers be surprised by us? A positive answer to the latter presupposes a degree of interpretative sophistication that our ‘digital machinery' still lacks. Were it to overcome this interpretive mediocrity, such machinery may eventually be able to 'originate' something, and hence surprise (and be surprised by) us in a non-trivial, coproduced way.

Keywords: Machine Learning, Turing, Autonomous Systems, Decision-Support Systems, Ethics, Surprise, Habit, Value-Alignment Problem, Lady Lovelace