Causality and Strong AI
by Sing Koo
October 13 2022 - Causality is indeed a major topic in Strong AI. Currently, there is effort underway to backdoor causality into ML and are labeled as Explainable AI. In general, it uses post-hoc or intrinsic methods. Doing so reminded me of knowledge engineering or rule based expert systems in the eighties. These processes are based on known taxonomy and cannot scale across different domains of discourse.
In order for AI to be truly working across any domain of discourse, a Strong AI System must be able to leverage on the narrative to derive logic and taxonomy dynamically and applies imagination to go beyond the boundary of a given narrative. I recalled the author of "The Book of Why" Julea Pearl once said:
"Yet even tool users do not necessarily posses a "theory" of their tool
that ells them why it works and what to do when it doesn't. For that, you need to have achieved a level of understanding that permits imagining"
To demonstrate the economic value of this idea, we applied ELAINE's narrative analysis to executive presentation of Earnings Conference Call Transcript and used the result to predict subjects of interest in the subsequent Q&A session. This kind of tool would help the executive team to better prepare their presentation. A well orchestrated Q&A session will likely help in preserving share values.