The following is an abstract I am preparing for a conference early next year.
I have decomposed thought into remembering, generalization, comparison, explanation, deduction, organization, induction, classification, concept formation, image manipulation, feature detection, analogy, compression, simulation, and value assessment. (Trans. Kansas Academy of Sci., vol. 108, pg. 169, 2005, vol. 109, pg 159, 2006, vol. 109, pg 254, 2006, vol. 110, pg 302, 2007, vol. 111, pg 174, 2008, vol. 112, pg 143, 2009, vol. 113, pg 127, 2010) Previous AI experiments like SOAR, ACT-R, and CYC have all lacked some of these capabilities. In order to implement these processes on a computer each of them is, in turn, decomposed into sorting, searching, vector averaging, vector differencing, vector dot product, sensitivity analysis, renormalization, interpolation, extrapolation, concatenation, time warping, and image manipulations like rotation, shifting, scaling, etc. These later are all well established algorithms for which there already exist efficient, error free, standard software components. There is no claim that this decomposition is unique. Hopefully it is adequate, or do we need explicit attention, affect, or language modules for instance? Also, are the linkage paths between modules correct and adequate?
In this fashion we hope to build minds which are more rational than humans and which possess better values. (Humans are not rational. See, for example, Predictably Irrational by D. Ariely, Harper, 2009) We should also note that we do not expect our AI Asa H to fill the same niche that humans fill. Asa is better than humans at certain mental tasks but certainly not all.
No comments:
Post a Comment