Högre seminariet i filosofi bjuder in till seminarium med Eric Campano och Kalle Grill. Seminariet ges på engelska och har den engelska titeln "The tension between physician benevolence and patient autonomy: artificially intelligent patient preference predictors".
Abstract (på engelska)
A patient preference predictor, or PPP, is a new clinical medical technology which, over the last decade or so, has been developed into different stages of prototypes. PPPs are also sometimes referred to as “autonomy algorithms”. The purpose of a PPP is to guess what a medical patient would want for his or her treatment, particularly in cases when the patient cannot communicate this in the moment — for example, if he or she is unconscious. A PPP uses artificially intelligent processes to gather and analyze large amounts of data about the patient, such as demographic information, clinical records, and even personal writing such as social media postings or email archives. The PPP then guesses, to a certain probability, what the patient would prefer if he or she were conscious. In an end-of-life dilemma, would the patient prefer to be kept on support, or be allowed to die passively? Research has suggested that PPPs can perform considerably better than loved ones or doctors at guessing a patient’s preferences.
The stakes involved are more nuanced than has so far been appreciated in the literature. In addition to clinical interests and third-party effects, there are several autonomy-related values that can all be predicted or estimated digitally. There is patient control over treatment, in the sense of explicit choice, informed by preferences and values. There is preference over health outcomes. In some cases, there are independent preferences over treatment methods. And then there is control and preference over the way decisions are made, which is particularly pertinent when the patient is currently incapacitated. In predicting or estimating the content of each of these, we will always face some degree of uncertainty, and so the moral landscape includes trade-offs between uncertain prospects. Given all this complexity, it is not clear that our best machines, now or the near future, will provide useful information in a clinical setting.
