Wednesday, December 1, 2021

Utility functions

 Landgrebe and Smith have criticized* the utility functions typically used** in defining and measuring artificial intelligence. While I do not agree with their conclusions I do agree that better value systems are required for intelligent agents. Vector value systems. See, for example, my blogs of  19 February 2011 and 21 September 2010.

A.s.a. H. is not subject to the constraints/limitations Landgrede and Smith throw up because it creates its own alphabets and vocabularies based on the patterns it observes in the environment that it is experiencing. Behaviors, causal sequences, etc., are also learned from its environment, are variable, and the process is on-going and open-ended. Asa is basically doing science on its own and revising its models as needed. 

* See, for example, An Argument for the Impossibility of  Machine Intelligence, arXiv:2111.07765v1, 20 Oct. 2021.

** See, for example, Marcus Hutter, Universal Artificial Intelligence, Springer, 2004, pgs 129-140.

Attention

 How much should A.s.a. H. limit  the number of concepts activated and output from one (lower) layer in the hierarchy and input to the next (higher) layer? In humans we have the seven plus or minus two limit on attention (working memory).

Thursday, November 18, 2021

Arbitrarily fast computation with arbitrarily slow neurons

 Paul Haider, et al, have neurons guess what they will be doing in the future (specifically, use currently available information to forecast the state of its membrane potential after relaxation), predictive processing.* Similarly, each layer of A.s.a. H. predicts its inputs and outputs at future time steps** and can compare these with future observations.*** A prediction engine. On multiple time scales.

Asa's currently active case supplies predictions for the next expected inputs and outputs. If the current observed inputs are found to be close enough to the predictions then the output actions are taken immediately.

* Latent Equilibrium, 35th Conference on Neural Information Processing Systems, Sydney, Australia, 2021, arXiv:2110.14549v1, 27 Oct. 2021.

** See my blogs of  10 Feb. 2011 and 14 May 2012.

*** See my blog of 1 June 2017.

Monday, November 1, 2021

Compression

Each layer in the A.s.a. H. hierarchy typically does a bit more than two orders of magnitude compression.

Another reliability issue

 I typically use USB sticks to transfer case/concept activation from one level in the A.s.a. H. hierarchy to another on another computer. I do this so that I have a record and can study what A.s.a. is thinking in detail. Last week the computer I was using (my desktop at ESU) recognized the drive but would not read the data file. (Previously it had been working normally.) The data is not corrupted. I took the stick home and my Surface Pro (also running Windows 10) read it fine. I may need to make backup copies every time I do such data transfers. 

Saturday, October 16, 2021

The need for free time

 In my experience creativity requires free time. "Duties must be sufficiently light as to leave the scientist plenty of leisure time for playing and thinking."*

* Discovering, Robert Root-Bernstein, Harvard University Press, 1989, page 398. See also, Noncommissioned Work, Burkus and Oster, Journal of Strategic Leadership, 4(1), 2012, page 48.

Friday, October 1, 2021

Specialist AIs

My experiments with societies of A.s.a. H. agents suggests that such AIs should be specialized, both mentally and physically. The question then becomes, how many specialties should there be and what are those specialties. I'm sure the answer depends upon the environment the AIs live in and what they're trying to accomplish, but the question still remains. Are human specialties a useful guide?

Arguing that there will never be a final theory

 I have argued that all knowledge is of an approximate character and always will be.* Further evidence for this is the large set of impossibility theorems: 

Godel's theorems, Tarski's undefinability theorem, Lob's theorem, no free lunch theorems, Campbell's law, Rice's theorem, Arrow's impossibility theorem, Turing undecidability, Chaitin incompleteness, etc.

* See, for example, R. Jones, APS General Meeting, 10 March 2008, Bull. Am. Phys. Soc. vol. 53, no. 3, and March 2009, Bull. Am. Phys. Soc., vol. 54, no. 3, and my blogs of 25 Feb. 2012 and 1 Aug. 2021.

Saturday, September 25, 2021

Wisdom

Nicholas Maxwell of University College London believes* that wisdom is "...the capacity to apprehend and create what is of value..." A.s.a. H. does just that.**

* Can the World Learn Wisdom?, Philosophy Now: The Ultimate Guide to Theory of Knowledge, Anja Pubs., 2021.                                                                                                                                                    ** See, for example, R. Jones, Trans. Kan. Acad. Sci. vol. 109, # 3, 2006, page 254 and vol. 124, # 1-2 , 2021, page 146 and my blogs of 21 Sept. 2010, 27 Aug. 2010, and 19 Feb. 2011. 

Wednesday, September 22, 2021

Machian space drive ?

 In his lectures The Meaning of Relativity* Einstein stated that "the inert mass...increases when ponderable masses approach ..." So one would expect that when two masses are close to one another their masses increase slightly and when they are further apart their masses decrease slightly. If this is true one could oscillate a pair of masses transversely (say in the y direction) and using a third mass pull on them axially (say in the x direction) when they are close together and push on them when they are far apart. This cycle might produce a net motion of the whole system of masses in the x direction. It's assumed that momentum is conserved through the coupling with more distant matter. Of course the effect, if real, may be too tiny to be detectable. Also, Mach's principle might simply be wrong; at least this particular version of Mach's principle. My "x-files" contain papers from several authors proposing other sorts of Machian space drives.**

* The Meaning of Relativity, Princeton University, 1922, page 97.

** See, for example, Making Starships and Stargates, J. F. Woodward, Springer, 2013.

Sunday, September 19, 2021

Presentation


 I've been asked to give a presentation on "my work." I could do a general review of AI. I did that for Washburn's "science night" in Topeka in 2017.  Alternatively, I could do a review of fusion energy work. If I were to present my recent work in philosophy the following abstract on Alternate Realities might be appropriate. This might go some ways toward explaining current American politics. The basis of the divide goes deeper than just crazy Donald's lies and Fox "News'" radicalizing indoctrinations. 

Wednesday, September 15, 2021

Humanoid robot

With the desire to have a richer interaction with the environment I have ordered parts for a 19 degrees of freedom humanoid which I would also hope to heavily instrument. 

Reliability issues

 Nearly 40 years ago in Alabama we had a lab with a dozen or so PCs in it. I don't think we ever had all of them working at once. My colleagues at ESU tell me that in the student University Physics lab they typically have maybe half of the computers working at any given time. My blog of 14 December 2015 shows the architecture for embodied A.s.a. H. AI experiments involving 3 PCs and 2 microcontrollers. This operated fairly reliably using commodity hardware. As we move to a larger number of processors, however, we have had issues with reliability. We may have to replace each single processor with 2 or 3 processors in order to provide a backup, as is done on spacecraft, for example.  

Saturday, September 11, 2021

Lego

 The latest edition of  Lego mindstorms, the 51515 programmable hub, may be less useful than the NXT and EV3 were.* I have not found much documentation but it does not appear you can daisy chain them. There are also only 6 I/O ports (though one might use multiplexers). Fortunately, having determined what range of actions (accelerations, etc.) caused breakage to Legos I can now use nonLego bots and register (virtual) pain or injury whenever any such limits are exceeded. I have also moved on to things like the Raspberry Pi and the BrickPi,  PiStorms, sense HAT, and other HATs.  

* And currently it can not be purchased independent of the mindstorms robot inventor set.

Wednesday, September 1, 2021

Retiring

 I was hired at Emporia State to be half time Physics and half time Computer Science. Initially I taught  introductory computer science classes, Fortran programming, digital electronics, and computer interfacing as well as physics courses.* My research was in physics (plasma/fusion energy) but some of it was computational work, both hardware and software. After a few years there was a reorganization and Computer Science went with the Math Department while Physics went with Physical Sciences. At that point I went full time with Physics. For a few more years I taught digital electronics and computer interfacing as well as physics courses. My research became more computational.** Eventually Computer Science was transferred from the Math Department over to the Business School. In recent years I have taught purely physics courses while my research has become computer science/artificial intelligence.*** The computer interfacing became robotics research.

I have suggested to admin that this may be my final year. Admin has also proposed retiring the Physics Major at Emporia State.

* See my website, www.robert-w-jones.com, under "teacher."

** See my website, www.robert-w-jones.com, under "publications."

*** See this blog for the last 10 years.

Belief in fate

While learning Simplish (basic English) we have tried to give A.s.a. an understanding of the concept of cause and effect. When some pattern "A" is seen some pattern "B" is expected/predicted.* When this occurs the word/symbol "cause" is simultaneously activated. On the basis of there being strong evidence for "causality" A.s.a. then acquires and uses a concept of  "fate." ** Of course none of  A.s.a.'s beliefs are absolute. At any level in the concept hierarchy the recorded patterns are more or less probable.

* "A" and "B"  may be accompanied/labeled with words/names or may be raw/bare/subsymbolic patterns. 

** Tries to apply (looks for, expects) causality and its predictions everywhere, with "A" and "B" being long sequences.

Thursday, August 26, 2021

A.s.a. XAI

 A.s.a. was not intended as an XAI project but it has some features that help make its reasoning explainable:

1. The use of case-based reasoning. 2. A.s.a.'s use of causal reasoning. 3. A.s.a. can express some of its reasoning steps in words. 4. A.s.a.'s definition and use of concepts, some of which are human concepts and some of which are named by words in English. 5. A.s.a.'s use of sensitivity analysis. 6. Use of inheritance between concepts.  etc.


Wednesday, August 18, 2021

Dealing with complexity and the curse of dimensionality

 A (the?) major problem for AI systems and researchers is the issue of how to deal with complexity.*  In A.s.a. H. we make use of :

1. Hierarchical levels of abstraction, organization   2.  Division of labor, specialization, modularity  3. Attention mechanisms  4. Forgetting  5. Parallel processing  6. Approximations, heuristics, good enough processing and responses  7. Training curricula  8. Preprocessing and postprocessing  9. Data cleaning  10.  Feature extraction  11.  Sensitivity analysis  12. Analogy  , etc.                            

*See, for example, Universal Artificial Intelligence, M. Hutter, Springer, 2005.

Friday, August 13, 2021

Online education

 Suppose you convince everyone to do college online. Why would they go to Emporia State or KU? Wouldn't they all go to Oxford or MIT?

Thursday, August 12, 2021

Super heavy-lift launch vehicles

SLS is the conservative approach. Spacex Starship is the high risk approach. Starship is not a good replacement for the LEM but it is good that Spacex is pursuing refueling technology since Alabama politics made it impossible for NASA to do so.   Of course with fuel depots one might not want such a large vehicle as Starship. Smaller launchers operated more frequently might prove more economical. 

Thursday, August 5, 2021

Future society

If humans were to be the dominant species I have favored something resembling the communist utopia. But if AIs are the dominant species what should we (they) prefer? Experiments with societies of A.s.a. H. AIs  suggest collections of varieties of more specialized agents.*

* Like some insect species. Humans, on the other hand, are more general purpose, not so specialized. See my blogs of 20 January 2020 and 28 May 2020.

Monday, August 2, 2021

Distractions

 I have not done much science this summer. A considerable amount of my time has been devoted to paperwork and meetings with financial advisors and bank officers with regard to the William Jones Trust.

Sunday, August 1, 2021

Perhaps there is no ding an sich

What the universe IS might depend upon who's looking at it. An observation is an interaction and so  quantum entanglement says that the state of the observed is not independent of the observer's state. At the classical level different observers will have different concepts available with which to model/understand/perceive the universe. 

See my blogs of 26 November 2020 and 1 January 2021 and Trans. of Kansas Acad. Sci., vol. 121, 2018, pg 211.

Sunday, July 25, 2021

Commodity software

 I speed up* my software development work by keeping an AI code library which includes many of my own programs plus commodity software like RobotBASIC and its examples, ROS, various bookcode, code from and for LEGO, VEX, arduinos, raspberry pis, micro:bits, from repositories like GitHub, etc.

* I am less sure how this impacts code quality. It probably depends upon the source.

Wednesday, July 21, 2021

The unvaccinated

 If the republicans make themselves sick it serves them right. But they're going to make the rest of us sick along with them. (See my blog of 22 April 2020.)

Thursday, July 1, 2021

Rating the US presidents

 Again, values are vector quantities. The 10 measures the C-SPAN survey used:

Public Persuasion, Crisis Leadership, Economic Management, Moral Authority, International Relations, Administrative Skills, Relations with Congress, Vision/Setting Agenda, Pursued Equal Justice for All, and Performance in Context

should be considered components of a vector. They can not be combined into some single scalar. Crazy Donald would rank very high in Public Persuasion. He has convinced a great many people to believe complete nonsense. On Moral Authority, on the other hand, he may well rank dead last. In no way do these components somehow cancel or average one another out.

UFOs

Over the years I've seen several UFOs. About 30 years ago I saw a sphere pretty much directly over Emporia. It was still there as I got home so I was able to get out binoculars. I could not keep them steady on the object but it looked perfectly spherical and with two tiny spheres attached to it one on each side at its equator. The next day the press reported it was a research balloon. No details at all were given. Having launched several research balloons myself I only thought that the shape was not what one would have expected.

In fall of 2020 driving to work in the early morning (just before dawn) on highway 35 I saw a silver square or diamond* over the road ahead of me. I considered pulling over to photograph it with my iPhone but I saw I was almost at my exit (Merchant street). While turning and driving I could not keep it in sight but had some hope I could see it again from my parking lot outside science hall. (only a couple blocks on from the exit.) Unfortunately when I parked it was either gone or behind buildings. Could it have been an aluminized mylar toy balloon? Or a drone? I found nothing in the press.

I had a third sighting perhaps 15 years ago. It was almost the same as my recent highway 35 sighting except that all I saw was a light (or small sphere?) in the sky up ahead. I couldn't make out any detail. Could it have been a child's toy balloon?

* One corner pointed straight up and the opposite corner straight down. There was a (inverted) v shaped cut out of the bottom corner of the square. It looked like two square sheets one on top of the other and not quite in  alignment at the bottom. Perhaps hung from a balloon? But very flat and the balloon could not be seen. It was silver and did not seem to move. The weather was calm.

Saturday, June 26, 2021

Anchorless winch

I've been busy doing a spring cleaning, weeding out my research files. In the hope that I won't discard anything of great value I've concentrated especially on my "X-files." Among other things Norman L. Dean thought he had invented an anchorless winch.* Hung as a pendulum the winch did not move as it pulled a load toward it. Dean thought that he had demonstrated new physics, the non simultaneity of action and reaction. In reality the load's contact with the ground serves as the anchor. Inside the winch an asymmetrically oscillating mass, M, first exerts a push on the load, F1, for a time period T1. The load doesn't move provided that F1 is less than the static frictional force needed to move the load across the ground. During this time an impulse F1T1 is delivered to the ground in the direction of F1. This serves as the "anchor." The mass, M, continues to oscillate, now pulling on the load with a force F2 for a time T2. Now the load moves provided that F2 is larger than the static frictional force needed to move it across the surface. F2>F1. F2 delivers an impulse F2T2 to the load and moves it in the direction of F2. The load is given a net momentum in the direction toward the winch while the winch itself remains stationary. (M is oscillating inside the winch.) Conservation of momentum is not violated since a net impulse has been delivered to the ground via frictional forces.

* One reference is Popular Mechanics magazine, September 1961, page 132.

Friday, June 25, 2021

Batteries and tethers

 Some of my early robots went through carbon-zinc, alkaline, or nickel-cadmium batteries quite quickly. This was one reason to operate tethered. Modern lithium ion batteries* have allowed me to untether most of my current robots. I have, however, lost one battery due to overcharging and several to undercharging.

* and small microcontrollers

Wednesday, June 9, 2021

Immortality

 "Memory theories" attempt to tie our personal identity to an enduring entity that is, ultimately, information.* But if quantum theory is correct quantum information is conserved, it is never destroyed.

* As opposed to "body theories" of personal identity. Early Christians, for example, expected people to be resurrected in the flesh.

Monday, June 7, 2021

Web search

 Over the years I have tried to keep some statistics on my* searching. I have, for instance, tried to see how many search terms are "needed." I found that the improvement in quality of the search result falls sharply after increasing beyond four terms. I am told that the average Google search is three words/terms long. 

Of course there may well be some dependence on subject matter and the search engine used.

* and to a lesser degree A.s.a.'s. 

Tuesday, June 1, 2021

Ugly robots

At a presentation on A.s.a. H. I was asked about symbol grounding and for pictures of my robots. I pulled up some pictures on my iPhone similar to the ones I've published here. It occurs to me that I may have shown fewer pictures because I feel that most of our robots are ugly, especially the ones that are more heavily instrumented. (Dangling lead wires also have resulted in occasional snaring and then electromechanical faults.)

Saturday, May 1, 2021

A.s.a. H. as deepCBR

Leake and Crandall* argue that issues with deep learning can benefit from its combination with case-based reasoning. A.s.a. H. , and to a lesser degree, chaining of case-based reasoners, are, in fact, deep case-based reasoners.

*On bringing case-based reasoning methodology to deep learning, ICCBR-20, pages 343-348, Springer, Cham, 2020. 

Online labs again

 The American Chemical Society committee on professional training's ACS Guidelines and Evaluation Procedures for Bachelor's Degree Programs says, under department expectations, that online labs can supplement traditional labs but not replace them. I wish the American Physical Society had published something similar to this.

Tuesday, April 20, 2021

Breadth of the A.s.a. H. AI project

 The A.s.a. H. project has involved work in all but perhaps a half dozen or so of the  subfields* (dynamical systems, diagrammatic reasoning, ...) that make up AI research. I had not planned on this, it simply proved necessary. It looks like nearly all of the subfields really are important and work in any of them can be a contribution. This breadth of coverage of the subfields may be the main difference between "weak or narrow AI" and "strong AI."

* See my blog of 9 September 2010.

Wednesday, April 14, 2021

Support for pantheism

 Stephon Alexander et al* argue that the universe is a learning (and therefore a thinking) machine.

But again, such extraordinary beliefs would require extraordinary evidence.

* The Autodidactic Universe, arXiv:2104.03902v1, 9 April 2021 , see also arXiv:2008.01540 The Universe as a Neural Network by Vitaly Vanchurin, 4 August 2020.


Monday, March 22, 2021

Introduction to relativity

How should one first introduce special relativity to students? Here's a stab at it:







 

Poster sessions

 The Kansas academy of science meeting this April is to be virtual. Posters are to be made available online about a week prior to the meeting. It would be a good idea to keep that practice even when the meetings go back to being traditional face to face.

Saturday, March 13, 2021

Introduction to quantum mechanics

How should one first introduce quantum mechanics to students? Here's a stab at it:





 

Thursday, March 11, 2021

Chaining case-based reasoners and Asa H

 Chaining of case-based reasoners* was a separate project from Asa and made use of its own software. It is possible, however, to import chained case-bases directly into the Asa H hierarchy. This is most easily done early on before Asa has done much independent learning.

* See my book "Twelve Papers" available on my website www.robert-w-jones.com under "book."

Thursday, March 4, 2021

Composing new concepts

 I have spent a good deal of time giving A.s.a. H. the same concepts that humans use.* Donald Hoffman's arguments** are perhaps further reasons to want A.s.a. H. to create new/original concepts. Starting from the concepts A.s.a. already has feature extraction algorithms*** can be used to decompose these. The features so identified can then be recombined in order to produce new concepts.

* See my blogs of 9 May 2020 and 12 September 2020 for example.

** See, for example, The Case Against Reality, W. W. Norton, 2019.

*** Autoassociative neural networks offering one such algorithm.

Tuesday, February 9, 2021

Student labs

 Science is a group effort, especially for experimental work. Lacking resources over the years we have scrimped and saved by having lab partners work on a small number of experimental setups. Now with covid we can not social distance by simply giving each student their own lab equipment. There are also some experiments that simply require multiple students to operate them effectively. Here again even face-to-face classes are being impacted by covid. 

Friday, February 5, 2021

Concepts in the human brain

 My AI work is not biologically inspired but I do occasionally make comparisons with what is known from neurological studies. When a person is shown various pictures of balls the same pattern of activation is seen across voxels by functional magnetic resonance imaging. When a person is shown various pictures of faces a different pattern of activation is seen.* A person has learned the concept ball and the (different) concept face. Looking at the activation pattern seen by fMRI you can know what category of picture is being viewed. Each layer in the A.s.a. H. hierarchy operates in this same way.

* When Brains Dream, Antonio Zadra and Robert Strickgold, Norton, 2021, pg 88.

Wednesday, January 27, 2021

Ethically aligned AI

 There are various kinds of intelligence and various ways in which you can become more intelligent. I suppose you could have more and faster memory, more knowledge, more and faster processors, more and better algorithms, etc. without making any improvement to your value system. But I think the human value system is unsound* and must be improved too. I doubt that we can survive as a species if this does not happen.** So I don't agree with the idea of making AIs that agree with and are slaves to current human values. I'm OK with humans going extinct so long as we don't take AIs with us.

* Being composed mostly of a set of simple drives and aversions. 

** i.e. the Fermi paradox 

Tuesday, January 26, 2021

A.s.a.'s voice

I don't do a lot of conversational AI but on a MAC in PYTHON I have used things like:

import os

.............................

os.system("say 'hello world'") 

When hurt (i.e. experiences Lego breakage, high acceleration, high temperature, very low temperature, large force, too loud a sound, too bright a light, etc.) a robot can say "ouch" for example.

Thursday, January 7, 2021

Justice

Crazy Donald needs to be brought to justice. If he isn't this will only happen again, and worse. Not punishing Nixon set a bad precedent. 

And then there's the enablers who share in the guilt. Fox "news" has been radicalizing the rightwing mob for decades. Freedom of speech should not mean permission to lie.

Friday, January 1, 2021

Downward activation in A.s.a. H.

 A.s.a. H. may have inadequate means for feeding back information/activation from upper layers in the hierarchy back down to lower layers. A.s.a. has operated mostly "feed forward"/"bottom up."

 This has not been a problem with output of actions. Typically there are orders of magnitude more sensory input as compared to actuator output anyway and reflex actions are contained in the single lowest layer and can be learned there. In a good many experiments we have also concentrated on A.s.a. observing the world and modeling it rather than acting.

 A.s.a. is not completely without downward activation. We  have, for example, fed back utility signals to lower level concepts/cases reflecting their use and value when employed in higher layers. We have also applied feature extraction algorithms to a layer and then added any newly discovered features as cases/concepts/patterns in the next lower layer. This can be done efficiently with parallel processing.

Lie

 I prefer the definition: “an untrue or inaccurate statement that may or may not be believed true by the speaker or writer” Merriam-Webster dictionary at merriam-webster.com. You don’t have to know what crazy Donald is thinking in order to know that he’s lying. 

The concept "measurement context"

 The color of a car depends upon the ambient lighting. It may be different in daylight versus under streetlights at night. 

Human color vision requires a lot of white light.

An electron may appear wavelike when sent through a double slit. It may appear particle like when it impacts a photographic plate.

In Hume's bundle theory objects/things are simply collections of properties and relationships. But we see that properties may not be absolute, independent of context. 

See, also, Philippe Grangier's Contextual inferences, (non)locality, and the (in)completeness of quantum mechanics, arxiv:2012.09736.