Wednesday, June 24, 2015
When Asa H has run plasma lab experiments and mobile robots it typically outputs things like voltages, forces, and torques. (see, for example, chapter 1 of my book Twelve Papers) Asa can, instead, provide an output that is the set point for a PID, or other, controller. (see, for example, PID Control, F. Haugen, Tapir press, 2004)
Tuesday, June 23, 2015
Most Asa H robotics experiments are done on simulators to save time and money. Sometimes we even turn off displays (renderings) to speed up the simulator. Although its easy to give a real mobile robot a wider VARIETY of sensor types than humans have (i.e., greater than the 5 human senses) it is difficult, with the exception of vision (cameras), to give the robot a large NUMBER of sensors. It is fairly easy, however, to give a simulated robot a larger number of virtual sensors. This is another reason to do as much as possible with simulators.
Sunday, June 21, 2015
While expanding its semantic network Asa H has reported to me that feature extraction can be taken to be function decomposition. That is, feature detectors aim to decompose input patterns into sub-patterns, each sub-pattern representing its own simpler function, at least as an approximation.
Thursday, June 18, 2015
With Eclipse (for example) if there are compilation errors you are asked if you still wish to continue with the run anyway. In AI you frequently can't anticipate all the environments your software will ultimately encounter. With my Asa H software I have sometimes received things like "out of range" messages and elected to proceed anyway. I find that I may still get reasonable responses from Asa when this occurs. Sometimes this amounts to extrapolation.
Tuesday, June 16, 2015
Multiple copies of programs like my Asa H 2.0 (see, for example, my blog of 10 Feb. 2011) each running in RobotBASIC on its own separate computer, can communicate with each other over the network using software described in chapter 8 and appendix C of Hardware Interfacing with RobotBASIC (Blankenship and Mishal, CreateSpace, 2011).
Wednesday, June 10, 2015
How might we distinguish between cognitive science and artificial intelligence? Cognitive science uses the methods of science (e.g., theory and experimentation) to try to understand the nature of thought and mind (consciousness, reasoning,....). Artificial intelligence is more engineering, it uses the methods/resources of science , mathematics and technology to create a useful product. There is, of course, considerable overlap between the two.
Friday, June 5, 2015
I find it very hard to convince my students that in order to really determine how large an effect is you must measure it multiple times. They want to think all the measurements should give the same result. I find it equally hard to convince my colleagues that we need to publish more replications. "Unrealistic scientific optimism," posted 4 June 2015 on gameswithwords.fieldofscience.com is an excellent argument for why scientific journals need to publish more replications and more null results.