New Book on Transparency

[RobW]

Hope you don’t mind, I just wanted to mention my new book on Transparency for Robots & Autonomous Systems, published by the Institution of Engineering & Technology (IET). It’s a combination of AI systems engineering, psychology experiments and discussion related to the ethics of robots and AI systems. More details here on my web site https://www.robwortham.com/new-book-transparency-for-robots-and-autonomous-systems/

[Andrew Martin]

Very interesting, what are the ways to read a preprint, or similar for those of us currently unaffiliated with an academic institution?

Either way, I wonder if biasing decisions based on ones previous experiences is necessarily at odds with transparency. It seems that, after having had some experiences there either is or is not a clear causal link between an individual experience and the effect of that experience on your final behaviour.

If there must be a clear causal link between experiences and decisions, then we are limited in which mechanisms may link experiences to decisions (namely those mechnaisms which are explainable). If there is not a clear causal link then we can use any computable mechanism to link experiences to decisions, but can only investigate the bias with some form of black-box analysis.

The process of writing the previous two paragraphs has cleared some things in my mind. I had initially felt there to be some kind of “no free lunch” relationship at play here. I felt that there must be a statement along the lines of “As you increase transparency in a system you necessarily must reduce the level of X the system”. I initially thought that X might be “bias” where bias means something vague like generalising from a set of experiences. Now I’m not so sure.

[Author]

Posts