New Contest II

Last time we considered my need for constructive contests. Recall that we have  now distinguished contest of an hypothetical nature and those contest with real world benefits beyond the purpose of the game. It was also a concern that some games are not fun.

Recently I mentioned to a coworker that brute-force solutions to games are so 20th century. Deep blue already proved it was possible. A large neural network is so 2010's, and alphaGo didn't just beat one world champion, it beat a lot of champions online as well. 

What about this? Let's have a competition that is played by human and computer together and that their enegy consumption is a divisor in the calculation of final score.

This is along the same line of thought from a post long ago before New Contest considerations: automated race car driving competitions where in addition to automotive restrictions that there will be CPU restrictions. Core limited to less than 2^16 cores, each running at no more than 4GHz, total power consumption is to be less than 1kw, etc.

So in a similar idea, one can imagine there be additional restriction on storage. The code the ANN model, the memory about current course and opponents, these all take memory space. So the contest could have an additional rule that there be a 256Gb limit to onboard memory, of any kind, hdd, ssd, cache, ram, GPU ram, etc. 

Suppose the contestants scored S in a scale where each score is real number originally associated with each player. and the highest score wins, the score used to determine the competition result can be S/C where represent the costs paid by competitors to win. For one example this could be the amount of code executed as measured by storage space or as measured by records kept on the CPU itself.

Therefore, in the post-AI years, we will have new contests: the winners must not only accomplish the objectives, they must also complete them efficiently.

TAS for Chomsky

Continuing the thoughts around expressing Chomsky’s ethical opinion regarding US military actions as expressed by the documentary that I watched during 2016-2017 break.

It became obvious that while propositional logic and set theory are comforting languages, they are not the most convenient when operationalizing an ethic. For example, there seem to be a need to distinguish these prescriptive targets: must do, must not do, may do, may not do.

\detokenize{must_do_actions}\subseteq \detokenize{may_do_actions}

\detokenize{must_not_do} = \detokenize{may_not_do}

It seems “may not do” is “not may do” due to common English use even though technically both expression should be subset relation. I.e. “You may not smoke” states “you must not smoke” instead of “you can choose not to smoke.” Although that imprecision is inconsequential to the current discussions.

So in fact the expression can be simplified by including do_not_do_ actions for each action in simple TAS. We must also impose a contemporaneous interpretation to a \in TAS as to mean a is an action viewed at some references moment, ostensibly now. 

Then, the predicate must\_do(a, p1, p2), reads “p1 must do a to p2” is expressed as ethical(a, p1, p2) \land \neg ethical(do\_not\_do\_a, p1, p2). may\_do is simply ethical.

may\_not\_do is \neg ethical
The need to restrict preposition to single moments in time is in retrospect necessary. All preposition can be sub-indexed with reference time: prescriptive proposition ethical_t means it is ethical at time t, while descriptive preposition do_t means something is done at time t. do_w(A,ally, axis) \implies ethical_u(A, axis, ally) \forall{w,u} \in WWII

That’s a mouthful. But at least we can avoid the pitfalls such as the “may not” fiasco we have in English.
(Disclaimer, I watched a thirdprty documentary about Chomsky. Some videoed statements were stitched together and I watched that. doesn’t mean I am writing to explain what he actually said or meant.)