May 2017 Newsletter

 |   |  Newsletters

Research updates

General updates

  • Our strategy update discusses changes to our AI forecasts and research priorities, new outreach goals, a MIRI/DeepMind collaboration, and other news.
  • MIRI is hiring software engineers! If you’re a programmer who’s passionate about MIRI’s mission and wants to directly support our research efforts, apply here to trial with us.
  • MIRI Assistant Research Fellow Ryan Carey has taken on an additional affiliation with the Centre for the Study of Existential Risk, and is also helping edit an issue of Informatica on superintelligence.

News and links

  • http://entropicai.blogspot.com Sergio HC

    Hi all, I haven’t found a better place to present myself, so here I go:

    I am a spanish mathematician -I studied decision theory on my grad, but too many years ago, so I am not really any expert- and a computer programmer in my own company -not related to AI at all- that have developed a general “decision making theory” and the corresponding AI algorithm that actually -and exactly- solves the problem stated in this paragraph:

    “Decision Theory: Say I give you the following: (1) a computer program describing a universe; (2) a computer program describing an agent; (3) a set of actions available to the agent; (4) a set of
    preferences specified over the history of states that the universe has been in. I task you with identifying the best action available to the agent, with respect to those preferences.”

    It is based on Causal Entropic Forces (by Alexander Wissner-Gross, 2013) and a new math method I developed for solving the many-path integral formula of those forces (quite similar with Feynman many-paths integral on amplitudes) based on fractals.

    I am actually writting it as a formal theory, even if there are some parts I can not really define or prof right (actually the equivalence on the causal cone Shannon entropy and a the fast aproximation of it I build and use, but it work as a charm).

    In short: given any system = agent+environment with n degrees of freedom that you know how to simulate (state at t+dt can be computed -deterministically or stochastically- form state at t and a dt), and given a utility function defined over the domain (the portion of the phase space the system can actually be), I can calculate the “intelligent” decision over the n degrees of freedom (forces applied on them) so that utility is maximized on a given time horizont.

    You can get an idea of the decision making algorithm on this blog post:

    http://entropicai.blogspot.com.es/2016/04/understanding-mining-example.html

    But I have also used this same idea for other problems, like global function optimisation (https://arxiv.org/abs/1705.08691).

    During this summer I hope I will finish the document defining what I call a “fractal intelligence”, covering the decision taking part of the intelligence + associative memory of past events + consciousness as a general way to manipulate and optimise the meta-parameters values (the “personality” of the AI).

    I anyone is interested, I could prepare a serie of “guest posts” on the subject.