Wednesday, June 28, 2017

Karl Popper's science "demarcation problem"

Even with scientific pluralism when an artificial intelligence reads from the internet what should it believe? What should it reject, or at least devalue? Up until now this has been a showstopper in trying to learn from the internet. Similarly, can we show our students how to recognize pseudoscientific nonsense?

In a famous paper Laudan actually argued that the demarcation problem is intractable (in Physics, Philosophy, and Psychoanalysis, Cohen and Laudan, D. Reidel, 1983, pg 111). Now Pigliucci uses a 2 (or more) dimensional vector criteria* to distinguish science, protoscience, and pseudoscience (Philosophy of Pseudoscience, Pigliucci and Boudry, U. Of Chicago, 2013, pg 23). One of Pigliucci's vector components is empirical content/support while the second vector component is theoretical understanding, internal coherence, logic. "Sciences" like physics and biology have large vector magnitude. Both of the vector components are large. "Protosciences" like economics and psychology have a smaller vector magnitude. One or both of the vector components are smaller. "Pseudosciences" like intelligent design and astrology have a small vector magnitude. Both vector components are small. (Another use of a vector value system.)

* The vector representation, or "cluster concept", is a realization of  Wittgenstein's "family resemblance" view of concepts (Philosophical Investigations, 1953). It is the way in which my artificial intelligence A.s.a. H. represents and manipulates concepts.

Sunday, June 25, 2017

Hacking the Hexbug Vex robots

The Hexbug Vex ant, spider, scarab, and strandbeest robots are inexpensive and are compatible with the Vex IQ components. They come with a simple "brain brick" which has only 2 inputs and 2 outputs and can be programmed in a limited way using a set of 6 switchs. One can replace this totally with the Vex IQ brain brick (ARM cortex M4 processor with a total of 12 input/output ports) and Vex IQ servos and sensors but it is also possible to retain the (1 or more) simpler "brain" bricks in a subsumption architecture (using a few relays). Each of the simpler brain bricks can perform a reflex (such as obstacle avoidance).

Friday, June 23, 2017

Asa's pain system again (see my blog of 31 March 2016)

I've had a Lego robot fall off a (3 foot high) table top. Bricks went everywhere but there was no other noticeable damage. My pain system would accurately assess such an event.

It takes about a 3 pound force to disconnect one Lego stud and about a 2 pound force to disconnect one Lego brick pin.  On the other hand it takes about a 5.5 pound force to disconnect one Vex IQ connector pin.

A Vex robot may possibly be susceptible to damage which is not detectable by Asa's pain system alone.

Wednesday, June 14, 2017

Reaping the whirlwind?

How many republican congressmen will need to be shot before we get meaningful gun control?

Saturday, June 10, 2017


According to Everett's theory human intelligences are "complex internal processes" in Hilbert space*. (The Emergent Multiverse, D. Wallace, OUP, 2014, pg 100 ) Could god simply be an even bigger (or even an infinite) "complex internal process" in Hilbert space? And possibly entangled with our world? To believe in such an extraordinary idea would, of course, require extraordinary evidence. Its just a thought.

* " can regard the state functions themselves as the fundamental entities." "...all of physics is presumed to follow from this function alone." The Many-World's Interpretation of Quantum Mechanics, Hugh Everett III, pg 9

Thursday, June 1, 2017

A.s.a. H. as a prediction engine

Andy Clark believes that intelligence is all about the ability to make accurate predictions. (in Brockman, This Idea Must Die, Harper, 2015, pg 310) My artificial intelligence Asa H predicts what will be seen in the next time steps. (see, for example, my blog of 10 Feb. 2011) We can assess the accuracy of each such prediction by taking a dot product of the prediction vector and the next actual observed input vector. A running average of this prediction accuracy can then be kept and made an additional vector component of that case's overall vector utility. (see my blog of 19 Feb. 2011)