purplecat: An open book with a quill pen and a lamp. (General:Academia)
[personal profile] purplecat
Both in a new conference called ERAS (Engineering Reliable Autonomous Systems). The first looks at how a system - e.g., a drone on a patrol - might decide to skip some patrol points if it realises it can't reach all of them. The second attempts to catalogue the kinds of things people want from a computer explanation. Explainability is big in AI at the moment, but explanation is quite a slippery term and its not clear that the support for explainability that's being developed actually meets what people want.

(no subject)

Date: 2025-10-02 08:49 am (UTC)
a_cubed: caricature (Default)
From: [personal profile] a_cubed
I've seen some stuff about "explainability" that really doesn't map to anything we can turn into a logical chain of reasoning that a human would understand. It's a bit like the "fairness" issue in ML from the mid-2010s. ML researchers asked ethicists for a definition of fairness they could check for/train to and got back about 17 different definitions, mostly pretty much mutually exclusive, and certainly not averageable or anything like that.

(no subject)

Date: 2025-10-03 01:39 am (UTC)
a_cubed: caricature (Default)
From: [personal profile] a_cubed
There's always an XKCD.

Profile

purplecat: Hand Drawn picture of a Toy Cat (Default)
purplecat

May 2026

S M T W T F S
     12
3456789
1011121314 1516
17181920212223
24252627282930
31      

Tags

Page Summary

Style Credit

Expand Cut Tags

No cut tags