3 days Workshops: Predictive AI in PTS, a complex Problem
Abstract
Prediction raises potential threats linked to the controllability of AI systems, which couldbuild information about the future of human subjects while exceeding the presenthuman knowledge and awareness. Following our recent meetings at Fort Detrick andthe MIT Lincoln Lab, and our conversations with leading professors in psychiatry andmilitary psychiatry in December 2019, we came to the conclusion that there is room forimprovement in the new multidisciplinary domain of prediction in PTS. Indeed, weforesee that reliable long-term prognosis can constitute an AI ethical problem, startingwith adding risk in the patient management practice. Additionally, the reliability of aprediction from an AI system is still challenged. Prediction explainability is due toguarantee accountability in medicine. Causality is one strong lead for predictingexplainability. Uncovering causal links between entities allows human experts tounderstand how the prediction is made, how to interpret the prediction, and what is theunderlying accountability. However, causality is still a debated subject in the communityand different approaches are being developed. For example, Clive W. J. Grangerdesigned a weak version of causality based on the time arrow direction. Their method isstill widely used on real world data today. On the other hand, Judea Pearl is anadvocate for using simple expert-based causal graphs to interpret still real-world butmore static data. Recent developments in deep learning tries to uncover new causalrelationships based on an agnostic approach despite increasing the difficulty to validateoutgoing models. Consequently, the impact of AI prediction in PTS could dangerouslybe underestimated: First, there is no consensus on the correct causality framework inthe scientific community; second, the forthcoming applications in human subjects arestill undefined and for some, unexpected. The problem is complex and ethics in humanresearch is engaged with potential threats in the near future. For these reasons, wedecided in conjunction with the Office of Naval Research Global to open the debate tohigh level experts in the 3 fields concerned, Militaries, Psychiatrists, AI specialists. Weexpect that pivotal changes and innovations will result from a collective andmultidisciplinary workshop to challenge the following main obstacles: Psychiatrists are for most part not aware of the kind of existing technologieswhich could support their practice. This gap of appreciation is preventing themedical community to imagine the impact of such tools. We would like to submitthis problem to prestigious representatives of the psychiatrist community in theUS and France. New technologies like chatbots and mobile applications designed to work standaloneand bypass the medical professionals are proliferating. In the clinical field,we are concerned by the risks raised by the suppression of human-to-humaninteraction. We would like to submit this problem to prestigious representatives ofthe psychiatrists and the AI communities in the US and France. Warfighters are a population facing a higher risk of PTSD or TBI. Tackling thisissue is key both at the personal and the institutional levels. Developingpreventive solutions can make a substantial difference if they are integrated inthe follow-up care everyone receives. We would like to discuss the specifics ofthe soldiers situation with representatives of the US Navy, the French and USArmies, along with the two communities, psychiatrists and AI specialists.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 06, 2021
- Source ID
- N629092112012
Entities
People
- Denis Fompeyrine
Organizations
- Office of Naval Research
- United States Navy