Monday, February 29, 2016

Student study guides

There is a good blog post on study guides and student attitudes on the 25 Feb. 2016 angrybychoice.fieldofscience.com blog.  I resist giving study guides.  I want students to do more not less.  I want students to learn more not less. I'm sure this hurts my student evaluations.

Tuesday, February 23, 2016

A concept of triggered recurrence

As we experience the world, we observe,  record, and refine spatial-temporal patterns at various levels of abstraction ("concepts").  We then use these to describe/explain new observations/patterns.  I am interested in finding new concepts with which to describe reality. (see my blogs of 21 Jan. 2016 and 10 Jan. 2016 for example) Fermi-Pasta-Ulam-Tsingou recurrence is related to an interesting concept Asa H has developed.  The pattern of activity described as "a volunteer fire brigade" is triggered by and follows the occurrence of a fire.  Asa H has developed the concept of a spatial-temporal pattern that recurs but only when triggered by the proper circumstances (rather than at some regular time interval).
Some of my colleagues think that I have given up physics altogether.  But I am trying to find better ways to describe reality.  This is physics at its most basic level.  Maybe there is not mass, not particles, not waves, not quantum fields but rather.......................

Friday, February 19, 2016

word2vec

Google is making a big deal of being able to take vector Paris, subtract from it vector France, and then add on vector Italy and get as an answer vector Rome. (for example, AI Weekly, 11 Feb. 2016)
Chapter 5 of my book Twelve Papers (www.robert-w-jones.com, book) pg 56-57 does the same thing with vector kitten - vector cat + vector dog = vector puppy.  My example was (intentionally) a low dimensional toy example but worked in the same way.

Thursday, February 18, 2016

B.E.A.M. Robotics

In my small robotics lab I design and build robots that define and ground concepts for my artificial intelligence Asa H. (See my blog of 1 Oct. 2015) I also have a half dozen small BEAM robots. Is there any way they can be useful to Asa? Asa could operate on top of a set of BEAM reflexes or BEAM elements might be used in place of other pre or post processors. A light seeker could be used with my solar power panels.  A beacon seeker could direct a robot to a recharging station/"food".  Cliff avoidance would be a useful reflex/fear. A righting reflex might be useful for some robots. A thermophobic might keep itself from overheating. I could use a BEAM element for wall following.......

Humanoid robotics

I want my artificial intelligence Asa H to understand enough of the concepts (words) that humans use so that we can communicate with one another and understand one another. This wont be a perfect correspondence but then it isn't between individual humans either.  So far I have tried to give Asa any of these needed concepts in the simplest (and cheapest) way that I can.  My blogs of 1 Oct. 2015 and 5 Nov. 2015 explain how I've done this with several mobile robots and other computer interfacing methods.

Some philosophers believe that the details of our body configuration are also important if an artificial intelligence is to adequately share and understand the concepts humans use to describe our experience. (i.e., the idea that we might never understand what it is to be a bat, for example, just because its body configuration and sensors are too different from our own.) I have not tried to build a humanoid body for Asa H.  As expensive as robotics is humanoid robotics is typically even more expensive. Even Blankenship's minimalistic Arlo (Arlo: The robot you've always wanted, CreateSpace, 2015) would cost at least $2000.00 without sensors and without the computer(s) that would hold Asa itself.

At other times I want Asa to form concepts that humans don't yet have (for example my blog of 10 Jan. 2016). If these two goals are mutually exclusive I may simply have to follow two different training paths for two different versions of Asa H. (The old question again of what the learning curriculum should be for an artificial intelligence.)

Creativity

Concepts may be decomposed into a set of features.  Features from various different concepts can then be mixed and matched to define new concepts.  If the new concepts prove useful in describing the world they are retained in the pool of  known patterns.
It's quite common to see a new concept formed from an old one by the simple addition of a single new feature:
               horse concept  +  horn feature = unicorn concept
               horse concept  +  wing feature = Pegasus concept
               man concept  +  wing feature = angel concept
I see Asa H doing this as well.

Wednesday, February 17, 2016

Causal reasonings in Asa H

Asa H is designed to learn and hierarchically decompose the spatiotemporal patterns that it experiences (R. Jones, Trans. Kansas Acad. Sci., 109, 3/4, pg 159, 2006).

On one level in the Asa H hierarchy (or, more realistically, over a small range of levels) a pattern may be formed involving the concept of force, the concept of mass, and connected /leading to the concept of acceleration. 

On a higher level in the Asa H hierarchy a pattern may be formed involving the concept of thought, the concept of knowledge, and connected/leading to a concept of creativity. 

Patterns will also form between/across levels in the hierarchy.  The concept of knowledge, at a higher level, will be linked with the concept of memory (on a lower level).  Various concepts on lower levels, like the concept of a sentence or an utterance or message, will activate the concept of thought on a higher level.

Monday, February 15, 2016

eV3

I decided to buy a LEGO eV3 brick.  They are faster than the NXT brick, have more memory, will accept a microSD card, and run Debian Linux.  eV3 also has 4 output ports while NXT has only 3. But, of course, we have long been using multiplexers to expand both the number of input and output ports. EV3s  are expensive enough that I do not plan to retire my NXTs. (But I will need to redesign at least some of my various robots.)

Saturday, February 13, 2016

Point charge student lab experiment

Forty years ago if you wanted to measure equipotentials in the first year physics lab you filled a tray with saltwater, powered the electrodes with an induction coil, and located the equipotentials using the sound you heard in a hand held earphone.  I soon discovered that one could use a DC bench power supply in place of the AC induction coil and measure potentials directly with a digital voltmeter.  You could even measure electric field with a pair of probe wires attached to a voltmeter.  Rotating the pair until you got a maximum reading gave you the field direction.  I used this setup for a few years until the modern conducting paper experiment became available.  It was less messy and no dangerous saltwater around power supplies.  But a point charge experiment in two dimensions does not give a potential that varies as 1/r or an electric field that varies as one over r squared.

A 3 dimensional point charge experiment is possible using a large jar filled with saltwater.  A metal screen is placed along the wall of the jar to serve as cathode connected to a battery or power supply.  An insulated wire with just a tiny conducting wire tip exposed is hung in the center of the jar and connected to the DC supply as the anode (I.e., the point charge).  Another similar insulated wire with an exposed tip is then connected to a voltmeter and used as a probe that you move around inside the water.  I use stiff "bell wire."  A pair can be used to measure electric field.  Now one can get the 3 dimensional 1/r scaling of voltage as a function of distance from the point charge.

Friday, February 12, 2016

Flow-through systems

There are people that believe that some sufficiently large but closed formal system, beginning with a theory of everything, (some string theory perhaps) could, through computer simulations and emergence, come to develop reasonable theories of physics, then chemistry, then biology, then social science, etc. etc. essentially forever. On the other hand I have noted that my own creativity machines seemed to operate more as information flow-through systems, requiring a steady input of new facts in order to generate new original output. (see Trans. Kansas Acad. Sci. 102, pg 32, 1999)
One of the behaviorist's models of the brain is the finite state machine which, at each time step, receives a new input stimulus, changes its state, and outputs some new response, sometimes a creative one. (See, for instance, page 98 of  Behaviorism by John Staddon, Duckworth, 1993.)
My own scientific work seems to depend upon a steady input in the form of books, journal articles, experimental results, etc. in order to continuously revise my theoretical models (state) and produce original results like those reported here and in my other publications. (i.e., an open system)

Wednesday, February 10, 2016

Life after death for quantum computers?

With an artificial intelligence program run on a quantum computer the processing (thinking) takes place either in computers in Everett's many worlds or else in the many dimensional Hilbert space. (see my blog of 13 Oct. 2015 and Bull. Am. Phys. Soc., March, 2016) If the computer in our world were destroyed while the program was running wouldn't the processing continue in other worlds? (But not forever.)

Tuesday, February 9, 2016

Psychometric AI

Ben Goertzel and others have criticized much of current mainstream work as being "narrow AI." (Artificial General Intelligence, Springer, 2007)  Bringsjord and Schimanski have suggested we overcome this limitation using "one program for many tasks" as proposed in Newell's "20 Questions" paper (in Visual Information Processing, Academic Press, 1973).  Where as "everyone is carrying out work on his or her own specific little part of human cognition" Newell proposed we "stay with the diverse collection of small experimental tasks, as now, but to construct a single system to perform them all." "it must be a single system in order to provide the integration we seek."  This has been done to some degree by the connectionists.  What a wide number of problems have been attempted using the standard backprop algorithm for example. I have been doing similar things with my Asa H.

Monday, February 8, 2016

Who won the space race?

There's a new BBC documentary titled: "Cosmonauts: How Russia Won the Space Race." Thought I'd present my own view/argument on this issue.  Here are some important accomplishments:

                                   First rocket into space: German V-2

Russian firsts:                                                              U.S. firsts:

First I.C.B.M., R-7                                 First satellite recovered from orbit, Discoverer
First earth satellite                                             First orbital docking
First moon probes                                              First men to orbit moon
First planetary probes                                        First men to land on moon
First man in space, Vostok                                Robotic exploration of Mars
First soft landing on moon                                Robots probe outer planets
First animals around moon and back safely      Robots out of solar system
Robotic exploration of Venus                           Sample return from comet
First space station                                              Reusable spacecraft
First permanently manned space station

It looks like a tie to me! Both sides have much to be proud of. There is no need to belittle one another.

Why program in BASIC?

The weak link in computing today is the human.  It makes sense to make things as simple for the human as possible.  The QB64 compiler takes your BASIC code and converts it to C++.

Friday, February 5, 2016

A and P Consciousness

Ned Block talks about access and phenomenal consciousness.  He says "A state is A-conscious if it is poised for direct control of thought and action." (Consciousness, Function, and Representation, MIT Press, 2007, Page 168) In my Asa H 2.0 light (see my blog of  10 Feb. 2011) the nearest matching case vector becomes active in the module between lines 1020 and 1080 in the code and may output an action or make a prediction about the future in the module between lines 2300 and 2320. In some versions of Asa H search is reduced by retaining this best matching case over time (keeping it in consciousness) until it's degree of match drops below some threshold.

On page 171 Block says "A-conscious states are ... Thoughts, beliefs...e.g., the thought that grass is green." Asa H's case vectors are exactly things like "grass is green" and other beliefs that Asa H has learned from its experiences in the world.

Block also says on page 171 that "The paradigm P-conscious states are sensations..." And on page 170 that "...the feel of pain is a P-conscious type" or kind of state.  When the parts of a Lego NXT robot controlled by Asa H break apart torn sensor wires signal a pain which is recorded in a case vector (a state, a state of Asa's being). Other external and internal sensors provide additional components for each of Asa's case/state vectors.

Asa H does not have what I would call a "phenomenal consciousness module" however.  (Block, page 163) Unless such modules are the sensors and any preprocessors that are associated with them.