Sunday, May 31, 2015

Pattern language

Asa H can be thought of as inductively inferring an approximate tree grammar describing the regularities it sees in the world (including itself), similar to Oliver Wendel's model of neural activity as a pattern language (in Topics in case-based reasoning, Wess, Althoff, and Richter, Eds., Springer-Verlag, 1994, especially pages 433-434).

Feature extraction in Asa H

Asa H has employed various feature extraction methods/algorithms.  Once clusters have formed in a given level of the concept hierarchy the difference between two cluster centroid vectors can be taken as a candidate feature and potential category in the next lower level of the hierarchy. What distinguishes between different categories is important.

Friday, May 29, 2015

Independent modular learning at early times and low levels in the Asa H hierarchy

There is some evidence that things like color, motion, depth, form, and audition are processed independently of one another in the human nervous system.  This suggests that the lowest several layers of Asa H could be trained for each sense independently using simpler data sets. (see post on vision versus audition, 7 Jan. 2015) These are, in part, preprocessors.

Tuesday, May 26, 2015

Asa H, CBR, and the AI curriculum

Case-based reasoning (CBR) only works if the agent has seen a sufficiently similar situation in the past.  Asa H's extrapolators help with this issue but there is a need for Asa to have been trained on a rich enough curriculum prior to it's deployment.

Monday, May 25, 2015

A future for idealism?

Evolution gives us our values.  As with all living things humans should value that which promotes their survival and spread.  Our current science and engineering help us cure our ills, grow our food, travel from place to place, etc. etc.  So far the idealist philosophers offer us nothing that we can use to design bridges, fly to the moon, treat our wounds, etc. etc.

But quantum mechanics suggests that ultimate reality may not be matter and space and time.  Konrad Zuse suggested that the universe might be a computer, an idea taken up more recently by Seth Lloyd.  The quantum mechanical wave function has been thought of as a "wave of probability" (information) and Wheeler has proposed his "It from bit."  These might be steps toward fleshing out a theory of idealism.

But if so there is still a long way to go in order to compete with our current materialist science.

Sunday, May 24, 2015

WordNet categories and Asa H

Asa H has learned about 60% of WordNet's top-level noun categories: act/action/activity, animal/fauna, artifact, attribute/property, body/corpus, cognition/knowledge, communication, event/happening, feeling/emotion, food, group/collection, location/place, motive, natural object, natural phenomenon, person/human, plant/flora, possession, process, quantity/amount, relation, shape, state/condition, substance, time.

Wednesday, May 20, 2015

Capitalism, war and peace

Some political philosophers try to argue that capitalism tends to promote war while others try to argue that it tends to promote peace.  I will only say that collectivism is inherently cooperative in nature while capitalism is inherently competitive, and war is competition taken to its extreme limit.

Monday, May 18, 2015

The AI curriculum and Asa H training

Ideally I might like to have a curriculum which first supplies simple patterns that are learned in the lowest level of the Asa H hierarchy.  Only when these simplest patterns were learned would one then advance to more complex, more abstract patterns and allow signaling to pass on up to the next level of the hierarchy.  This process would then be repeated to further levels of detail and complexity on each of the successive layers.

I know how to do this for simple tasks occurring in simple environments but not for more general agents acting in more realistic worlds.

Monday, May 11, 2015

>5 senses

It is easy to give an AI more than the 5 human senses.  I have added a Vernier radiation monitor to Asa H's suite of sensors.  Asa quickly learns that the radiation level rises as it (a Lego mobile robot) approaches the orange disc gamma source.


Friday, May 8, 2015

Qualia

When the appropriate region in our visual cortex is activated by strong signals from red cones and green cones in the eye and when no substantial signal from blue cones is present, that is the experience of the color yellow.  Similarly, when Asa H has the appropriate lower level case activated by a signal of value 5.5 coming in from a Lego NXT color sensor, that too is an experiencing of the color yellow. Either of these may also have become associated with simultaneously hearing the word "yellow" spoken. If you receive slightly different strength signals from the cones in your eyes then you experience a slightly different shade of yellow.  If Asa receives a 5.4 or 5.6 from the color sensor then it experiences a slightly different shade of yellow.

Wednesday, May 6, 2015

Space drive

People are asking me about  Shawyer/White's "EM drive."  Years ago I did some work on the idea of "pushing against empty space," (R. Jones, American Journal of Physics,  vol. 37, pg 1187,  1969 ) but it was research more along the lines of  Jack Wisdom's work ( Science, 21 March 2003, pg 1865 ). Shawyer and Yang Juan  and White have all offered various different explanations for their results. Most of the work is unpublished and none of it has been independently verified. Extraordinary claims require extraordinary evidence. And, again, the majority of scientific papers may well be wrong (New Scientist, 30 Aug. 2005)


Shawyer and White's device:



It may well be that we need to explore the details of the measurement systems being employed.  Some years ago I had a Geiger counter that would respond when I turned on an electric motor in the same room.  The counter was battery powered and I assume its circuitry was acting like a radio and picking up RF interference generated by the motor.  When I did radiation background measurements with my mechanical roughing vacuum pumps turned off I measured a low level background.  When I was doing (plasma) experiments the pump motors were turned on and the Geiger counter registered a higher level of radiation.  At first I thought this was all coming from my plasma. 

One of the suggested EM drive spaceships is to have a mass of 90,000kg, be powered by a 2,000,000 Watt nuclear reactor and have a thrust of 800 Newtons (.4N/kW).  This would give an acceleration of almost .009 m/s/s and in a week or so of powered flight the ship would have a speed of almost 5400 m/s and kinetic energy of over 1.3 teraJoules.  But the nuclear reactor will have only supplied an energy of  E=P t , perhaps 1.2 teraJoules.  The violation of conservation of energy gets worse for longer powered flight times.

 With a constant thrust to mass ratio the acceleration is constant so the velocity increases linearly with time.  So the energy output, the ship's kinetic energy, must increase as time squared. But the input power from the nuclear reactor is constant so the input energy only increases linearly with time. So conservation of energy will always be violated if the powered flight time is long enough. This is the same issue I used to discredit the Dean Drive back in the 1960s. The EM drive would be an even better energy source than it is a propulsion system.

So the EM drive not only pushes on the vacuum it gets energy for free (from the vacuum??) as well.  2 miracles, each extremely unlikely to be true.  Like with cold fusion, when the number of miracles required exceeds 1, at that point I give up hope. It's just simple probability theory.

Tuesday, May 5, 2015

Benchmarks, beyond the Turing test

AI research involves software development and all software development involves regular and extensive testing.  Along with the AI curriculum I've been working on for years now (see my blog of 10 April 2015 and references therein) one would want some suitable test sets. 

I believe that intelligence, values/utility, consciousness,.... are all complex vector quantities.  Although a single test can certainly try to measure more than one quantity, still, it seems likely that more than one test might be needed in order to gauge an AI's overall performance. Jia You recently described how the Turing test might be replaced by a battery or suite of tests (Science, 9 Jan. 2015, pg 116).

In testing my own code I typically start with simple, and then more complex, logic functions (see chapter 1 of my book, Twelve Papers, for example).  For me, a follow on test is often times character recognition. But where should one go from there? I think a good test suite can only be developed in conjunction with the AI curriculum. Perhaps the school of  "test first" software development would have us create the test suite first and then the AI curriculum.  In designing an intelligence I would think that the opposite might be more reasonable, or, perhaps, working through both curriculum and tests in an iterative fashion.