TU Delft & NFI Science Café – deepdive edition
Justification in forensic machine learning: what is it and why do we want it?
31 May 2023 15:30 till 17:00 - Location: TU Delft | The Hague at Bezuidenhoutseweg 63, The Hague | Add to my calendar
In criminal cases, forensic experts are requested to submit assessments on the evidential value of a piece of information. Whether or not in the triage of facts a judge accepts the reasoning of an expert, depends on the expert’s demonstrable expertise and impartiality.
In practice, experts explain their reasoning, such as pointing out the features of the evidence they considered, to increase trust in their judgment further and thus give grounds for the judge’s final decision.
The use of machine learning for forensic analysis (such as speaker comparison and facial recognition) suffers from at least two difficulties that cast a shadow on the experts’ assessment: 1) the algorithm's output is difficult to explain and 2) virtually impossible to understand. The core reason is known: algorithms are like a black-box. In the case of forensic science, this means that the algorithm is proprietary. In this context, how can the expert reconcile the value of using these technologies with issuing credible assessments?
In this science café we will discuss that justification could bridge this problem by providing reasons to believe the ML output. The idea is to shift claims on explanation and understanding to claims on reasons to believe. In this context, the talks and discussion concern what justification is for forensic ML and why we want it.
Organised by NFI, TU Delft Safety & Security Institute and TU Delft | The Hague
For whom: legal, forensic and AI researchers, experts, policy makers and those working in the field. Limited capacity.
Enrolment: send an email with your name, affiliation, and why you are interested in the session.