Feedback is where an action in some way influences the original system or actor that initiated it. This happens in physical and social dynamical systems, for this see feedback phenomena. It also happens in more discrete interactions between robots, or software egents and their environments and between people and automated systems.
- feedback from actions on the world – When a robot or software agent (or human) performs an action on the world, it is important to be able to sense the change that has happened in order to verify that the action has the expected effect.. See also feedback
- human–human conversation – In human communication we offer a combination of explicit confirmation (e.g. "yes, that's right") amd implicit feedback (e.g. "would you like a single or double shot" after a request for an espresso, which demonsrates undersanding of the request). Some of these techniques can be emulated in human–machine communications.
- feedback from systems to users – When a user perfoms an acton on a system, it is normally essential that they receive some form of feedback so that they can tell that (a) the system correctly understood their request and (b) has executed it as expected.
- feedback from human to machine – It can also be heloful for AI systems to know if they have provided pertinent or useful information, requring some form of human confirmation. Often this is designed to minimise conscious effort for the user, for example in epistemic interaction, but may include more explicit 'thumbs up' or stars for media or news choices in recomender systems.
Used in Chap. 14: pages 205, 215; Chap. 15: pages 230, 231, 234