Smart sensors*, process control**, and robotic applications*** are common examples of edge AI, especially when running on single board computers like Raspberry Pis.****
* see R. Jones, Trans. Kan. Acad. Sci., vol. 104, pg 31, 2001
** see R. Jones, Trans. Kan. Acad. Sci., vol. 107, pg 32, 2004
*** see R. Jones, Trans. Kan. Acad. Sci., vol. 109, pg 159, 2006
**** see my blog of 27 Aug. 2017
One way of trying to deal with complexity* is to employ a society of specialist AI agents.** These different agents will likely develop differing worldviews.*** This can lead to interagent conflict.****
* See my blog of 18 August 2021.
** See my blog of 1 October 2021.
*** See my Trans. Kansas Acad. Sci., vol. 121, pg. 211, 2018.
**** See my blog of 15 June 2022.
In general different clustering algorithms find different clusters. In A.s.a. H. these clusters are A.s.a.'s concepts with which it models the world. There are many clustering algorithms that we have yet to try in A.s.a. This would be another way to attempt to force a reconceptualization of reality.
Wikipedia says that exploratory programming "...is an important part of the software engineering cycle when a domain is not very well understood or open-ended, or it's not clear what algorithms and data structures might be needed..." This is the situation with artificial intelligence development. An example is the work mentioned in my blog of 11 August 2020.
Many of our experiments produce noisy inputs. We sometimes use an auto-encoder neural network for data cleaning as a preprocessor. Whenever possible this is pretrained using clean data sets.
What constitutes life depends upon the environment it's found in. Hominids on the savannah, microbes in the soil, viruses in host cells, computer viruses attached to emails, A.I.s in computer systems**, ...
**See my blog of 19 October 2010.
The ESU Bulletin is reporting that the ESU administration is ending ("suspending") the physics program. See my blogs of 1 February 2022 and 1 September 2021.
I see that the Emporia State University administration has violated tenure, firing tenured professors. Perhaps the ESU faculty should go out on strike. In my view the only legitimate leadership is collective leadership.
It's been suggested that humans have a special purpose language module.* Modern humans typically resort to pocket calculators in order to do common math calculations.
A.s.a., and other AIs,** could be given a math module "add-on." The AI would learn associations like: one -> 00110001, plus -> 00101011, equal -> 00111101, 00110010 -> two, etc. The AI could then pass math calculations to a simple calculator program and receive back results.
* See, for example, Chomsky or Fodor. (Though the modern view of language modularity is weaker.)
** Neural networks are believed to be bad at learning common math operations.
A.s.a. H. decomposes sensory input into hierarchical sets of commonly recurring patterns.
Adding additional sensors to its AI robots should allow A.s.a. to make greater use of lower level patterns.
As an example: humans have unique scents that enable dogs to identify/differentiate each of them, one from another. A dog need not learn to identify various different human faces, for example. (Faces are a complex pattern situated further up in the concept abstraction hierarchy.) Adding a sense of smell allows differentiation to occur at a lower (simpler) level of abstraction.
Does NASA have launch fever? It seems to have skipped over some issues in its run-up to launching Artemis. Will this cause failure?
1.) to perceive the meaning of words.
As perhaps the simplest example A.s.a. hears/receives the word "stop" and halts its motion.*
2.) interpret in a particular way (perhaps resolving ambiguity).
As a simple example hearing the word "stop" when in the act of lifting or moving A.s.a. knows to halt that particular action.
3.) infer something from information received.
With wheels turning and no motion of the robot body observed A.s.a. concludes it is stuck.
4.) an observation is classified as an example of a general pattern.
A.s.a. displays more complex examples of each of these.
* see, for example, Experiments with A.s.a. H., in my book Twelve Papers, available on my website Robert-W-Jones.com under "Book."
There are common/recurring patterns on various spatial and temporal scales and to different degrees of abstraction. The A.s.a. H. hierarchy* decomposes behaviors/knowledge into various networks/sets of these.
* and other deep learning systems
Some categories defined at one level of abstraction may be causally related. Some categories defined on a higher level of abstraction may not be. The wave function may evolve causally but the relationship between capital and labor may not be totally causal. etc.
I am working to try to improve my robots' dexterity* by increasing the number of sensors and actuators (degrees of freedom) they employ.
* and to encourage a richer definition of concepts
As a part of natural language understanding (and symbol grounding) work I have given A.s.a. H. preprocessors that recognize numerals and simple shapes.* I do not yet know what curriculum (training sequence) is needed in order to give A.s.a. the concepts required for a "natural mathematical understanding." Just how such mathematical cognition develops in humans is not well understood.
* A.s.a. also has had builtin translation, rotation, reflection, scaling, logic functions, and similarity measures. Is there more that needs to be builtin (innate)?
Some alternate realities are:
Materialism: The entities described by physics are all that there is.
Idealism: Mind is all that there is.
Dualism: Materialism and spirit interacting with one another.
Theism: One (or more) supreme being and its creations.
Multiverse: Everett quantum mechanics, inflationary bubbles, fecund universe, etc.
It would be interesting to know what percentage of people have adopted each of these weltanschauung. The fragmentation of human society is quite natural, we experience some very different realities.
The concepts A.s.a. H. learns* are defined** by their place in the hierarchically structured memory/case-base and the neighboring concepts they are linked with. This will vary a bit from one intelligent agent to another and one training history to another. I think this is also true among humans. This is one reason I don't find the definition of words to be as important as some philosophers do.
* and words which name them (see my blog of 5 November 2015)
** and meaning is established
Most of the experiments that my A.s.a. H. robots have been tasked with* can be accomplished using a small set (society) of specialist agents**:
transporter, arm, various grippers, walker, flyer, swimmer, and brain
each with appropriate onboard sensors.
* See, for example, my blog of 5 November 2015.
** See, also, my blogs of 1 and 16 February 2020.
If an agent needs something he values it. Since a human's needs* differ from those of an AI humans and AIs will have different values. This may likely lead to some conflict between humans and AIs.
* humans need air, water, food, mates, etc. (The problem with money, scalar utility, is it is one dimensional while the world is not.)
A.s.a. H. decomposes its model of the world into concepts and sequences (transition models over the vector state space). In the version of A.s.a. H. 2.0 light published in my 10 February 2011 blog utility controlled extrapolations can occur around line 3000 of the code. Other learning/extrapolation algorithms have been discussed in various of my publications on A.s.a. H.* With these algorithms A.s.a. rethinks the past and past actions on multiple levels of abstraction over its hierarchical memory trying to improve the expected rewards/utility. Responses to alternative possible environmental conditions are also explored/evaluated. i.e., mental time travel, another piece of machine consciousness.
* See, for example, Trans. Kansas Academy of Science, vol. 109, no. 3-4, pg 159-167, 2006 and vol. 124, no. 1-2 , pg 146, 2021 as well as work with A.s.a. F.
I am looking at Steven Grossberg's summary* of his 50 years of cognitive science research, Conscious MIND Resonant BRAIN (Oxford Univ. Press, 2021). Grossberg argues that a (the?) most important feature of an intelligent system is its use of "laminar computing." Grossberg's LAMINART is organized in layers like A.s.a. H.**, i.e., "laminar computing."
* But you have to go to his original publications for the mathematical details.
** The algorithms Grossberg uses differ from those of my A.s.a. H. (Though I have sometimes used his ART as the clustering algorithm in Asa.)
I have students bring two fairly strong ceramic magnets (but not supermagnets) together in opposition. When repelling you experience something that feels like an invisible putty ball between the magnets. I point out that there is a "sixth sense" that magnetotactic bacteria have* but humans do not. I also point out that when a compass needle turns, or if the magnets in their hand flip, work is being done so magnetic fields have energy and by Einstein's E=mcc magnetic fields must also have mass. So fields are like matter but not composed of atoms.
* See for example Magnetic Navigation in Bacteria, Scientific American, Dec. 1981.
Spacio-temporal scaling*, translation, reflection, rotation, and shearing are important for pattern matching. I did some work on those algorithms during the early development of A.s.a. F. But computer vision, speech, and the like are such popular fields that more recently I have left this work to others.** At some point this must be better integrated with A.s.a. H.
* Average human speech, for example, is 120-150 words per minute while the fastest (barely understandable) is perhaps 667 WPM.
** I have employed temporal scaling and translation in A.s.a. H.
Having used robots' senses to ground/define some basic concepts one might then consider turning off** one or more (all?) of these input channels after some initial learning phase. I am currently experimenting with doing this with A.s.a. H.
* See, for example, my blogs of 27 May 2014 and 1 May 2019 and references therein.
** Permanently or perhaps only temporarily (as another attention mechanism).
The conditions that made possible the war of 1914-1918 appear to exist again today.
1. Mutual defense alliances: especially N.A.T.O.
2. Imperialism: today the American and Russian empires
3. Militarism: in America and Russia
4. Nationalism: American, Russian, Ukrainian
5. Immediate cause: The war in Ukraine drawing in more and more military aid from N.A.T.O. countries.
Will this lead to a catastrophic global war?
It has always been my intention (hope) that in retirement the only thing that would change would be that I would teach (much) less. I have been in partial (if unofficial) retirement for years in that its been a decade or so since ESU had any summer courses for me to teach. In moving my books and research files out of my ESU office I have to sort out (and discard) all of the old and unneeded materials that I can. (Although its a lot of work moving does have the benefit of forcing you to weed out.) I was short of lab and office space even before retiring. That will be an even bigger challenge now.
In the monograph Strong Artificial Intelligence: The approaches to supermind* Alexander Vedyakhin et al argue that the human brain employs a prototype theory of categorization, comparing observations to previously stored representations of categories. A.s.a. H. is such a program. At each time step attribute bundles (features) are learned and causal relationships are learned across multiple time steps.
* Intellectual Literature, 2021
In his paper "Reward Is Not Enough: can we liberate AI from the reinforcement learning paradigm?" Vacslav Glukhov argues once again* that a scalar utility like money is unsound. These arguments also constitute important arguments against the whole capitalist system.
* arXiv:2202.03192v1 3 Feb. 2022
One of the things A.s.a. H. does is learn behavior trees. It was designed to, among other things, learn hierarchically structured "sequences of sequences."*
* R. Jones, A hierarchical architecture for software agents, Part 1, Part 2, and Part 3, 14 November 2003, http://groups.google.com/group/sci.cognitive and Transactions of the Kansas Academy of Science, vol. 109, no. 3/4, pg. 159, 2006.
In my blog of 1 September 2021 I mentioned that the university administration was proposing that we retire the physics major. They claimed that physics was running at an annual loss of about $88,000.00. Several of us disputed that. Among other things they had omitted our Space Science course in their calculation. I'm told a recalculation now says we are delivering a small net profit of about $7000.00 each year. (And ESU is not a for profit school.)
When I was a student you couldn't go to school if you hadn't had your vaccinations. We actually got the polio vaccine at school. We should be requiring our college students to be vaccinated.
In their paper Scalar Reward is Not Enough* Peter Vamplew, et al, argue that scalar rewards are insufficient for an intelligent system.
* arXiv:2112.15422, 3 January 2022. See also my blog of 19 February 2011 and Capitalism is Wrong in my book Twelve Papers (www.robert-w-jones.com, Book).
In his paper "What are the Realities?" arXiv:2112.10282v1, 19 Dec. 2021, James Hartle outlines a view of multiple realities that is essentially the same as mine and A.s.a.'s.
I've been playing with S. C. Gruget's Mini Kore and Mini Mundo languages in place of Toki Pona and Simplish.
I have bought a Seeed Studio reTerminal. It can be substituted for the Raspberry Pi in any of my robots and they would then have the advantage of no longer operating "headless." The 5 inch touch screen is adequate for display purposes though rather small for input. One can temporarily plug in a mouse and keyboard if need be. The fact that there is no internal battery is no problem for my use in robots. In fact, it is helpful to be able to choose from a variety of battery sizes.
Elon Musk's email regarding possible Spacex bankruptcy reminds us of the fact that, large or small, most companies fail in 3 to 5 years, about 90% of companies fail within 10 years, and hardly any company survives 100 years. And what happens when Musk dies? Capitalism suffers from a shortsightedness and value system that makes it hard for a capitalist society to tackle projects like controlled nuclear fusion, artificial intelligence, space travel, and climate change.
I tried a computer search "real labs are messy." On my ESU machine I got no search results and a notice that I would have to change my Bing SafeSearch filter settings in order to allow for adult content. I tried the same search on one of my home computers and got reasonable results discussing things like lab safety and clutter. Crazy.
I keep code libraries in BASIC, Prolog, C/C++, Python, Lisp, Java, etc. In Python, for example, I have sample programs for clustering, neural networks, genetic programming, statistics, logic programming, expert systems, etc. I am putting together a library of robot designs for walkers, grippers, arms, humanoids, etc. Such libraries help to speed up the design of new experiments.
