Professor James,

Any effort in this way is ussefull. I think there are very important aspects to take into account and they are difficult to manage: 1) The man is a monkey that thinks itself superior instead of different (that vision of what chimpanzees do is very appropriate), therefore, the accumulated of actions always ends in a result similar to the one that corresponds to a limited rationality. 2) Any model is dependent on hypotheses that can change over time. 3) It is necessary to contemplate homeostasis, for not to repeat the experiences of that report on the Limits to Growth (for example).

Always a good computational model can give important ideas after the analysis. I wish you success in the proyect.

Prof. Doran:

The application of AI-ABM could be quite interesting, but framing the research might be a bit of a challenge. As I'm sure you know, the use of, or lack of use of, nuclear weapons is already a well developed line of research. Likewise, the development of game theory was accelerated by the development of nuclear weapons. The application of an ABM could allow for nation-states to be represented with their decisions making driven by a developed strategy. One of the challenges is going to be determining the number of agents that represent a nation-state though. A single agent per nation-state might be too simple considering the complex dynamics that influence national level policy. A collection of agents organized as a nation-state might be an interesting model though.

Nation-states are one of the key points since in the modern era nuclear weapons are typically seen as weapons intended to deter existential threats. Development of nuclear weapons by countries such as North Korea (NK) is evidence of this. While the costs of a conventional invasion of NK was previously extremely high, their development of nuclear weapons effectively means that their country is immune to threats of invasion. They now have a "guard" against existential threats to the nation. As such, while some of the motivations for the development of the nuclear weapon could be framed using greed, pride, ambition, family loyalty, aggression, etc. the development of the weapon doesn't imply deployment of the weapon.

A more interesting question might be to see if the AI-ABM could develop a profile of a potentially "irrational" actor that would deploy the weapon in spite of the threat of nuclear retaliation or an existential threat to themselves. While it seems intuitive that terrorists fit that profile, could a stable nation-state also fit that profile? Another question might be how much resistance needs to be overcome to develop the weapon in the first place, along with the profile of possible nation-states that do so. The technical challenges involved means that resources may have to be diverted from other areas of interest - a model that takes all of that into account could be quite interesting!

Dear Jim,

My own (current) views on these kinds of endeavours is that perhaps it's useful to take a step back to ask what kind of answer / model / result would be of use and to whom? Suppose you could wind back history and be there in 1945 knowing everything you know now. What would you do? Suppose you had a model that was very persuasive and said "this intervention" would prevent a nuclear war. What if that intervention was that rich countries should redistribute 50% of their wealth to poor countries. How would you make this intervention?

First reactions to some of the very interesting points made in replies to my question.

As a preliminary, please note that in my informal “Hiroshima” talk I had in mind my own current project (MoHAT – see my profile here on RG) which does have working computer code associated with it – not just ambitious ideas!

So..

J (Rafiee) feels that I am working on “an absolutely wrong problem” and suggests that “competition is a ‘positive’ force and ‘comfort’ a negative one”. I’m not sure whether the next step in this line of thought is to accept that catastrophes killing millions of people (eg the Tai Ping rebellion, the Holocaust) are inevitable features of the road to human destiny or whether he/she deplores them as I do but feels I am going about my search for a means to avoid them in a fully misconceived way. I suspect the latter. Maybe elaborate a little J?

Manuel makes several persuasive points. As regards the importance of homeostasis, I should stress that my project (MoHAT) uses the idea of DISCOVERING effective models rather than hand crafting them. Hence, to the extent that Manuel is right, the MoHAT system should discover models that incorporate homeostasis. OK!

Robert reminds me that research on the use/non-use of nuclear weapons is already extensive. Yes, indeed. I have the greatest respect for it. And it is very thought provoking and relevant as some of Robert points indicate. All that I am suggesting is that computational models of the AI/ABM/MoHAT variety may provide a newish line of attack on the need for more global cooperation. Compare with MoHAT computer modelling in the research and political context of climate change -- NB the latter's potentially catastrophic consequences. The stakes are high enough that almost any possible road forward, however small its chance of ultimate success, needs to be followed.

And, in addition, the actual AI/Computational problem of discovering/creating relevant models is both precise and, to me anyway, very interesting.

Dave asks:

What if that intervention was that rich countries should redistribute 50% of their wealth to poor countries. How would you make this intervention?

This is a key issue for MoHAT . Currently the idea is that the model discovery system is to be designed to answer the question: given that only one specific component (let’s call it “Xland”) of the socio-cultural system being modeled can be relied upon to take the advice of the modelling system what is the BEST (or LEAST INEFFECTIVE) action that particular component (i.e. Xland) can take to achieve the overall recommendation i.e. all should redistribute 50%.... Amongst many, three possible answers are: (i) Xland actually redistributes half of its own wealth, hence setting an example and a precedent; (ii) Xland threatens to spread worldwide an unstoppable toxin it happens to have lying around unless every country redistributes; (iii) Xland organises a national year of prayer.... Perhaps these examples are a touch frivolous, but hopefully they clarify…

An actual MoHAT math/comp model is inevitably a multi-agent system and we are essentially finding what is the best action one particular agent (or sub-system of agents) can take to bring about (maximize probability of) some overall property of the entire system. This is a well-defined question that can be precisely formulated and answered. [E.g. that the actions of an agent correspond to a subset of the signals it can possibly pass ...]

But won’t any useful model created by the MoHAT system (or similar) of the global socio-cultural system be impossibly complex? I believe that this is something still to be discovered – we cannot know in advance of trial how simple/simplified a global social model can be and yet still be effective towards the goal of promoting cooperation.

Thank you again for your responses that I shall digest further.

Further comments and criticisms welcome!

Best regards

Jim

@J. Rafiee

@Manuel Chacón Mateos

@Robert Zupko

@David Hales

Ul2sF�17Ɔx�