Laurent, This & other papers in this thread are interesting in that they show a movement towards a more positive , *creative* kind of agent.
“We believe such agents (or their computational variants) should turn out to be useful to humanity in a dierent way than RL agents, since they should constantly be creative and solve interesting problems that we may not yet know.” I esp welcome “creative” because that is what AGI is about. Any other people in this or related fields thinking similarly? There does seem to be a definite trend here. What needs to be recognized though is that a creative machine represents an entirely different and opposed culture to that of rationality. For example, the whole of science would love to know what “optimal learning/discovery” constitutes in the real scientific world. Actually, there is no such thing as “optimality” in creative endeavours. What’s the optimal painting of a Madonna, or the optimal scientific discovery of the brain’s workings, or the optimal AGI machine? “Optimality” belongs to the rational world where you compute/calculate right answers – a monistic as opposed to a pluralistic culture. From: Laurent Sent: Monday, April 22, 2013 7:16 AM To: AGI Subject: Re: [agi] Causal Entropic Forces If you are interested in entropy maximization over future histories and AGI to define an optimal scientist, you will probably be interested in this paper: http://link.springer.com/chapter/10.1007%2F978-3-642-24412-4_28 and the pdf can be found here: http://www.agroparistech.fr/mmip/maths/laurent_orseau/papers/orseau-ALT-2011-knowledge-seeking.pdf Laurent On Sun, Apr 21, 2013 at 3:19 PM, martin biehl <[email protected]> wrote: Another information theoretic principle with capabilities possibly similar to the ones in the paper is empowerment. It can for example also make poles swing up: http://www.worldscientific.com/doi/abs/10.1142/S0219525912500798 (sorry for not providing an open link, if you are willing to wait a day or so, I can get it for you, just mail me). Empowerment also depends on the possible futures of a state, but on top of that on control over these futures ( here is an open access paper on it http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0004018) Thanks Robert for posting this, it is a very interesting paper. Also this came just when I was thinking about quitting my lurking on this list. Disclaimer: I work with the authors of the papers above. If you are interested here is some related work by other people (you guys all know Schmidthuber and Ben mentioned him so he is not in the list) I don't understand Friston, but it always sounds great. Karl Friston (2012) A Free Energy Principle for Biological Systems Entropy, 14, 2100-2121; doi:10.3390/e14112100 Georg Martius, Ralf Der, Nihat Ay (2013) Information driven self-organization of complex robotic behaviors http://arxiv.org/abs/1301.7473 Susanne Still, David A. Sivak, Anthony J. Bell, and Gavin E. Crooks (2012) Thermodynamics of Prediction Phys. Rev. Lett. 109, 120604. Naftali Tishby, Daniel Polani (2011) Information Theory of Decisions and Actions In: Perception-Action Cycle, Springer Series in Cognitive and Neural Systems, pp 601-636. 2013/4/21 Tim Tyler <[email protected]> On 21/04/2013 00:53, Giovanni Santostasi wrote: But here it is claimed that intelligent behavior is associated with entropy maximization over histories not minimization. What is going on? Roderick Dewar claims to derive more general versions of Prigogine’s principle of minimum entropy production from MaxEnt here: Maximum entropy production and the fluctuation theorem - R C Dewar - http://www.swarmagents.cn/thesis/doc/jake_210.pdf He claims this reproduces Prigogine’s principle without the usual near-equilibrium assumption. Maximum entropy production is much bigger than Prigogine's ideas. -- __________ |im |yler http://timtyler.org/ [email protected] Remove lock to reply. AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
