Post date: Jun 30, 2020 1:32:42 PM
Hi Scott
thanks.
Re humane algorithms, that is very close to another thing I am involved in, and which we formally launched last night:
Humanising Machine Intelligence
Risk and uncertainty are absolutely central to what we want to do, and we have several people on the team with a deep interest in reasoning under uncertainty. You may find my most recent work on this of interest: http://proceedings.mlr.press/v97/williamson19a
b
On 10 Aug 2019, at 6:05 am, Ferson, Scott <Scott.Ferson@liverpool.ac.uk> wrote:
Dear Bob,
Thanks a lot for the link to your scanned pdf. It is very helpful. I should have mentioned that it was Marco De Angelis (https://marcodeangelis.github.io/) who was trying to find your dissertation.
Your proposal is very interesting. We’ve been talking for a while about humane algorithms in Bill Kahan’s sense. It is perfectly clear after looking at your text that trustworthiness of data and possible algorithms has to be a part of any such conception.
Thanks also for the Heuer text. I don’t think I’ve seen that one, so I’ll definitely take a look. We’ve been worried about the issue of how you go from an expert’s utterance to a quantitative probabilistic statement. The “words of estimative probability” are popular around here, but I kind of think confusion over phrases like these have gotten many folks into trouble. What is remarkable to me is that no one seems to argue that there should be an uncertainty penalty that we should slap on expert elicitations as a matter of course.
I have been loosely following the probabilistic languages crowd for a long time. I tried to get them interested in what we were doing but they are not interested. But I noticed that the Wikipedia page for PPLs includes Analytica, so I’m thinking that, if it counts, then what we’re doing counts too!
Thanks a lot.
Best regards,
Scott
From: Robert Williamson
Sent: 05 August 2019 11:43 PM
To: Ferson, Scott
Subject: Re: your dissertation
Hi Scott
good to hear from you. I had indeed noticed you had moved to the UK a while ago.
Here is a scan of the thesis with all figures: http://axiom.anu.edu.au/~williams/ProbabilisticArithmeticScanned.pdf
(too big to email I think)
Your pre-compiler sounds interesting. I guess you are aware of all the work in machine learning on probabilistic languages? Not quite the same thing, but related.
I still have an interest in the general issue of uncertainty management. You will see mention of it in the attached proposal I wrote a couple of years ago. (Not funded alas).
I looked at your Natural Language of Uncertainty. It reminded me of the great graphic on page 155 of the attached. You might find this fun. The book is really cool.