[SystemSafety] Google Willow ..... Extension to

[Prof. Dr. Peter Bernard Ladkin]

On 2024-12-12 13:24 , Prof. Dr. Peter Bernard Ladkin wrote:
> On 2024-12-12 12:28 , Roderick Chapman wrote:
>> On 12/12/2024 10:20, Prof. Dr. Peter Bernard Ladkin wrote:
>>> So what am I missing? 
>>
>> Try: https://scottaaronson.blog
>>
> Thanks. 

This sent me down another rabbit hole. Whether it counts as "System Safety" depends on what system 
you are looking at (an investment exchange may not be one of them) and what you count as "safety" 
(having a strong possibility that someone loses all your resources  does count for many people as 
"safety", though not as "engineering safety" unless you count "financial engineering" on digital 
"money" to be "engineering").

Scott Aaronson has a blog post on SBF just after the FTX scandal hit that hits the nail on the head 
concerning certain sorts of utility calculations that came up when SBF was discussing effective 
altruism. https://scottaaronson.blog/?p=6797

Aaronson also points to this substack post by Sarah Constantin which also hits some other nails on 
the head https://sarahconstantin.substack.com/p/why-infinite-coin-flipping-is-bad

There are plenty of people on this list willing to discuss, in detail, Bernoulli Processes and their 
worth (or not) for software reliability assessments. Constantin's post discusses their worth in 
utilitarian-ethics calculations, of the sort which Effective-Altriusm types apparently like to 
engage in. So even if one thinks the ostensive content is not "system safety", then the details, 
namely certain observations of the characteristics of Bernoulli Processes, certainly lie within the 
scope of this list.

Besides, I don't think it is wrong to occasionally discuss ethics. There are many occasions on which 
I have been in engineering-safety discussions, come up against ethical stances (in particular, 
stances I considered poor or inappropriate) and wished to be able to discuss them as such. Usually 
forlornly.

It seems to me in retrospect that there was a social-ethical dimension that SBF & Co seemed 
completely to ignore.

The discussion in Aaronson and Constantin concerns repeated coin-flips, i.e. Bernoulli Processes, 
but in fact certain conclusions generalise to repeated games (of the sort considered by Daniel 
Ellsberg in his belatedly-published PhD thesis Risk, Ambiguity and Decision, Routledge 2010). The 
main assumption here is "repeated". In these specific ethical instances, it is facile to the point 
of irresponsible to consider that the tosses all have the same probability, or that a specific game 
is repeated.

Suppose you are betting on repeated coin flips in a Bernoulli Process. Then the Kelly Criterion 
(1956, I believe) tells you how to bet on successive flips to maximise utility (= expected winnings) 
in a gain-or-quit betting situation (1 = you get more money back than you put in; 0 = the game stops).

What such calculations do not say is what you should do when you are engaging in such a game and 0 = 
you are wiped out. For if you play *such* a game, you are wiped out with probability 1 - \epsilon, 
(\epsilon coming in because you cannot in any practical sense go on for ever).

We have laws which say you can play this game as you like with your own money, but you can't play 
such games with money other people have given you, except under certain very constrained conditions. 
The courts have ruled SBF & Co were ignoring those constraints. That is essentially what Aaronson 
says, when the scandal broke (and of course before the trial started, let alone after the verdict 
was delivered). Those constraints essentially mean that you are legally proscribed from engaging in 
trading with investor capital in such a manner that you are probabilistically exceptionally likely 
to go broke. Aaronson and Constantin suggested, as the scandal broke, was what SBF appeared to have 
been doing. The courts agreed.

Is that a fair comment?

The phenomenon is well known in investing that someone makes a lot of money, is considered to be a 
"genius" or extremely talented or something, and then tanks. An example most people know: Dick Fuld 
and Lehmann Brothers. Of course, almost no one in actual investing is doing the "same thing" with 
billions of dollars as they were doing way back when with thousands of dollars, which is why the 
repeated-game model doesn't fit real-world behaviour all that well. About the only person who 
"carried on doing the same thing" for decades and continued making exceptional returns was Jim 
Simons and the Medallion Fund of his company Renaissance Technologies. Here, the "same thing" was 
extremely sophisticated mathematical analysis (Simons was am exceptionally talented mathematician 
before he gave it up to go into investing).

PBL

Prof. Dr. Peter Bernard Ladkin
Causalis Limited/Causalis IngenieurGmbH, Bielefeld, Germany
Tel: +49 (0)521 3 29 31 00