Robustness of AI Systems
Robustness of AI Systems - Mindful AI

Robustness of AI Systems

Steadily increasing technological advances in artificial intelligence encourage more and more businesses and governments to deploy AI in high-stakes environments, including autonomous driving, diagnostics in medicine, managing the power grid, and controlling autonomous weapons systems. However, for such applications, AI methods must be robust to the known unknowns (the uncertain aspects of the world that the computer can reason about explicitly) and the unknown unknowns (the elements of the world that system models do not capture). Thus, we need to pay more attention to the challenges of dealing with both known and, above all, the unknown unknowns.

These issues are essential because they address the fundamental question of how finite systems can survive in a complex and dangerous world and thrive for humankind and nature's benefit.

Murat

(Excerpt from my new book "MINDFUL AI - Reflection Artificial Intelligence")

Darius Burschka

Professor CIT (TUM), Member Scientific Board - Munich Institute of Robotics and Machine Intelligence

2 年

Aren't most AI systems "detectors" finding similarities to previous knowledge? If this is true then - they will have hard time to discover *unknown unknowns* and the *known unknowns* must be presented in training (which makes the expression "unknowns" more to *unusual* or *undesired*) - as detectors, they are not really well equipped to do predictions and discover new solutions. I guess that you will bring up VAEs, but sweeping through latent spaces is really not an efficient way of finding new solutions, because it is very tedious to filter them out (this is usually not mentioned in all the "success reports of AI") I accept to use them in autonomous driving to detect traffic structures, but using them to plan actions?

回复
Margrietha H. (Greet) Vink

Director | Research Development and Innovation | Smart Tech | Public Affairs | External Relations | Government - European Commission - United Nations | Gender & Diversity | Funding | Research | Education | Training

2 年
回复
Juan Martín Ricardo Ramirez Hurtado

Senior Quality Controller en H&M

2 年

Dear Murat, congratulations about you Recent release!!! as the first is very focused and usual.

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了