This essay will discuss two types of algorithms: those capable of learning from their own ‘mistakes’ and those not fitted with such capacities. The first concern machine learning (ML), the second can be summed up as ‘dumb’ if this then that algorithms (IFTTTs). I will suggest that each can have added value as well as drawbacks, depending on how they are used, in what context and for what purpose. As the decision to engage either one of them may have far reaching consequences for those subjected to their outcome, I propose that both should be constraint in ways that open them up to scrutiny and render their computational judgements liable to being nullified as a result of legal proceedings.