Tuesday, August 30, 2016

Graphical models in Asa H

I am reading David Danks' book Unifying the Mind (MIT Press, 2014) dealing with cognitive representation by graphical models. My AI  Asa H grows just such models, see, for example, the one in my blog of 4 March 2015.
Danks argues that at least some of human thinking involves operations performed on cognitive representations that are structured as graphical models. Again, Asa H does just that.

Monday, August 29, 2016

Evaluation functions

Referring back to my 16 September 2010 blog defining several different sorts of intelligent systems and to the definition of intelligence in my letter in Skeptic, vol 12, #3, pg 14, 2006. A system having a performance element only (no evaluation function) has functions that map sensory inputs to motor outputs and yet it still maximizes some quantity (see Barrow and Tipler, The Anthropic Cosmological Principle, OUP, 1986, pg 151 and Yourgrau and Mandelstam, Variational Principles in Dynamics and Quantum Theory, Dover, 1968, pg 176). But this may not be doing a good job of maximizing the things we want, utility, reproductive success, or lifespan.  Evolution acting on a set of such agents can tend to improve this over the time scale of generations.  If the environment is not changing no further learning might ever be required.
Adding an evaluation function and some sensors that can measure reproduction, pain, damage, etc. can tend to improve the "utility" on a time scale shorter than an agent's lifespan.  This is important, for instance, in dealing with environments that are rapidly changing.
We do not necessarily expect to ever be optimal we are simply seeking to improve over time.

Saturday, August 27, 2016


Human hands are soft, flexible, and can be cupped.  Current robotic hands are hard and rigid.  If a robotic hand with one thumb is used as a scoop to pick up small items (nuts and bolts or small Lego bricks) some of them spill off the side of the hand.  (Though tilting the hand helps some.) The addition of a second thumb (see my blog of 18 August 2016) prevents this and makes for better use as a scoop.

Thursday, August 25, 2016


As a child I played with the old metal meccano/erector sets.  As an adult I even occasionally used metal meccano parts as inexpensive fixturing inside of vacuum chambers. I decided to buy one of the new (plastic) meccano meccanoid robot kits.  Meccano's meccabrain has more outputs than do Lego NXT or EV3.  Speech recognition and the ability to use a smartphone camera as input for motion capture is also interesting.  There is nothing like the range of commercially available NXT compatible sensors, however.  I plan to try to hack the meccabrain inputs.

I have seen people who have hacked the meccabrain software.  We'll see if this platform can be interfaced with Asa H.

Wednesday, August 24, 2016

Division of labor

Although it is fair to call A.s.a. H. An artificial general intelligence we have trained A.s.a. Repeatedly using different (specialized) syllabi.  The size and hierarchical depth may also vary from agent to agent. A typical A.s.a. Is a specialist.  We assume division of labor just as in human society.  The agents can be organized and interact/cooperate in something like councilism.

IBM's Watson has been criticized when it has been necessary to specially tune the algorithm and/or training set for each application.  But, like humans, A.I.s probably will be specialists.  This helps to control complexity and works quite well in human society.

Thursday, August 18, 2016

A hand with two thumbs

I have published several papers on what are called discovery machines or creativity machines.  See, for example, Trans. Kansas Acad. Sci. vol. 102, pg 32, 1999 and my book Twelve Papers (www.robert-w-jones.com under "book") In the last few years my artificial intelligence A.s.a. H. suggested that a hand with 2 thumbs might be better than a hand with 1 thumb.  Often times one can not tell how a creativity machine comes to its conclusions.  The same is frequently true of human creative thought.  In this case, however, A.s.a. was using its extrapolation algorithm.  It had been told that a hand with a thumb is better than a hand with no thumb.  (One might challenge this, of course, even though many people have stressed the importance of the opposable thumb.) Extrapolation evidently led A.s.a. to suspect that two thumbs MIGHT be better still.  (A.s.a. knows to treat extrapolations as uncertain postulates.)

We can, of course, build robot hands having two thumbs.  In fact there is a continuous spectrum of possible geometries.  (You can google search "robot claw hand" and find quite a few.) A simple pincer made of two opposed "fingers" might be thought of as one using just their thumb and forefinger. Going beyond a hand with two thumbs might get one to the usual "claw" geometry, i.e. 2 pincers oriented at 90 degrees to one another.  The number of fingers can be varied in each geometry.  So the question is, which geometry is best?  I suspect there is not a simple answer to this.  Probably the question is, best for some particular task and environment. Perhaps we can get an engineering student to work on this.

In order to learn *

A.s.a. H.'s training curriculum attempts to expose Asa to the simplest concepts first and then gradually grow more complex concepts on top of these (hierarchically). Complex concepts are assembled (emerge) out of combinations of simpler concepts. This is rather different from what happens with human infants.  Infants can not be totally shielded from the influences of the (complex) outside world. To what degree will this make it harder for Asa and humans to understand each other? To what degree will Asa's (model of) reality differ from a human's?

* reference to Ritter, et al's book by that name, Oxford U. Press, 2007

Thursday, August 4, 2016

Vectors as pluralism

Vector value was forced on us for the kind of reasons outlined in chapter 2 of my book Twelve Papers (www.robert-w-jones.com, book) and is used in ways like those shown in my blogs of  21 Sept. 2010 and 19 Feb. 2011. Pluralism was developed independently, forced upon us by the kind of reasoning described in my blogs of  17 August 2012 and 21 July 2016, for example. But, clearly, substituting vector quantities for scalars is a kind of pluralism. The two lines of research are now linked together.

Virtual LEGO

I have very little documentation of my Asa H-LEGO NXT/EV3 robots. I have installed and am now running LDraw in an effort to try to improve this situation in the future and in order to aid with designing. I am assembling a robot design library that I can draw from in much the same way as my A.I. Code libraries.

Tuesday, August 2, 2016

Intelligence, broadly defined

I am reading Mancuso and Viola's book defending the idea that plants are intelligent. (Brilliant Green, Island Press, 2013) Intelligence comes in degrees, even among humans. Plants are able to detect gravity, temperature, humidity, light, chemical gradients, etc. and respond by moving their roots, leaves, stems, flowers, and producing chemicals of various kinds for various purposes dependent upon their current situation and needs.  Like animals they try to survive and reproduce. They satisfy the definition of intelligence I defended in Skeptic, vol. 12, #3, pg 14, 2006. They have a cognitive architecture that is highly distributed. (I defined 5 different levels/kinds of intelligence in my 16 September 2010 blog.) I agree with Mancuso and Viola that if we were to contact space aliens we might need to have had experience with a wide variety of different intelligences.

Monday, August 1, 2016

The mental health of a presidential candidate

I personally don't think any human being should have access to nuclear weapons.  I favor something like Gorbachev's global zero option.  But under the present circumstances if a candidate is suspected of having cluster B personality disorder shouldn't the mental health community be vetting the candidates and offering their assessment prior to the election?