No abstract

Although rapid advances in machine learning have made it increasingly applicable to expert decision-making, the delivery of accurate algorithmic predictions alone is insufficient for effective human-AI collaboration. In this work, we investigate the key types of information medical experts desire when they are first introduced to a diagnostic AI assistant. In a qualitative lab study, we interviewed 21 pathologists before, during, and after being presented deep neural network (DNN) predictions for prostate cancer diagnosis, to learn the types of information that they desired about the AI assistant. Our findings reveal that, far beyond understanding the local, case-specific reasoning behind any model decision, clinicians desired upfront information about basic, global properties of the model, such as its known strengths and limitations, its subjective point-of-view, and its overall design objective-what it's designed to be optimized for. Participants compared these information needs to the collaborative mental models they develop of their medical colleagues when seeking a second opinion: the medical perspectives and standards that those colleagues embody, and the compatibility of those perspectives with their own diagnostic patterns. These findings broaden and enrich discussions surrounding AI transparency for collaborative decision-making, providing a richer understanding of what experts find important in their introduction to AI assistants before integrating them into routine practice. CCS Concepts: • Human-centered computing → Human computer interaction (HCI).

Machin e lear nin g (ML) is incr easingly being use d in image retrieval systems for medical decision making. On e app lication of ML is to retrieve visually similar medical images from pas t patients (e.g. tissue from biops ies) to reference whe n making a medical decision with a new pat ient. Howeve r, no algorithm can perfectly captu re an expert ' s ideal notion of similarity for every case: an image th at is algorithmi cally determin ed to be similar may not be medically relevant to a doctor' s specific diagnostic needs. In this pape r, we identified the needs of patho logists whe n searchin g for similar images retrieved usin g a deep lear nin g algorithm , and develope d tools that empower use rs to cope with the search algorithm on-the -fly, communi cating what types of similarity are most import ant at different moment s in time. In two evaluations with path ologists, we found th at th ese refinement tools increased the diagnos tic utility of images found and increased user trus t in the algorithm. Th e tools we re preferred over a traditi onal interface, without a loss in diagnostic accuracy. We also observe d that users adopted new str ategies whe n using refinement tools, re-purpos ing th em to test and understand the underlying algorithm and to disambiguate ML errors from their own errors. Taken togethe r, these findings inform futur e hum an-ML collabo rative systems for expe rt decision-m aking. CCS CONCEPTS• Human-centered computing --> Human computer interaction (HCI); KEYWORDS Human -AI int eraction ; machin e learnin g; clinical healthPermission to mak e digital or har d copies of part or all of this work for personal or classroom use is grant ed without fee provi ded that copies are not made or distributed for profit or commercial advanta ge and that copies bear this notice an d the full citation on the first page. Figure 1: Medical images contain a wide range of clinical features , such as cellular (1) and glandular morphology (2), interaction between components (3), processing artifacts (4), and many more. It can be difficult for a similar -image search algorithm to perfectly capture an expert's notion of similarity ,

The interpretation of deep learning models is a challenge due to their size, complexity, and often opaque internal state. In addition, many systems, such as image classifiers, operate on low-level features rather than high-level concepts. To address these challenges, we introduce Concept Activation Vectors (CAVs), which provide an interpretation of a neural net's internal state in terms of human-friendly concepts. The key idea is to view the high-dimensional internal state of a neural net as an aid, not an obstacle. We show how to use CAVs as part of a technique, Testing with CAVs (TCAV), that uses directional derivatives to quantify the degree to which a user-defined concept is important to a classification result-for example, how sensitive a prediction of zebra is to the presence of stripes. Using the domain of image classification as a testing ground, we describe how CAVs may be used to explore hypotheses and generate insights for a standard image classification network as well as a medical application.

No abstract