Tuesday, July 1, 2025

Obsolescence

By one count I have 68 small robots and the component parts* for at least 20 more. Some of these are obsolete and others soon will be**. I try to counter this to some degree using reconfigurable designs* but this can increase the costs involved*** and so is not a universal solution.

* For example LEGO, Meccano, Vex IQ, Vex EDR, Raspberry Pis, etc.

** For example LEGO RCX, NXT, Ev3, early Arduinos, etc. The LEGO Technic cables from my RCX Mindstorms sets are actually disintegrating due to age.

*** Hacking toys, for instance, can be cheaper. See my blogs of 1 March and 7 April 2025.

My SCSs

My specialist AIs* are my version of SCSs (Self-Contained Systems**), software AND hardware. With cointelligence some of the specialist agents can even be humans.

* See, as one example, my blog of 1 July 2023.

** See scs-architecture.org

Sunday, June 22, 2025

Spacex Starship explosions

It needs to be remembered that Starship is the high risk approach (my blog of 12 August 2021). There will be failures.  

Friday, June 20, 2025

An integrated intelligent traffic management system NOT simple f.s.d.

 The environment needs to be automated too, with roadway navigation beacons, digital traffic signs and signals, smart traffic lights, etc. so as to monitor and then control vehicle traffic.

Saturday, June 14, 2025

Infrequent but important

 During neural network training the impact of rarely occurring training examples* can suffer "dilution" as compared to more frequently occurring examples. They can get "averaged out". When such uncommon examples are important* one can/should artificially increase the number of times the ANN/LLM sees these during its training. I.e., frequency of training exposure should be weighted by example importance/value. I've employed this technique since my earliest work with ANNs.

* A key example would be dangerous scenarios during level 5 vehicle autonomy.

Sunday, June 1, 2025

One robot's ontology

At a conference someone well versed in g.o.f.a.i. asked me what A.s.a. H.'s ontology was. I answered that it was something like the diagram presented in my blog of 4 September 2024. (Including some nodes/patterns/concepts that may go unnamed.)

An economical space launch system

A fully reusable launcher might not be the most economical launcher. At least for certain size payloads. See my blog of 1 January 2025.  

Sunday, May 11, 2025

Prototyping

 I recommend to students that they develop both hardware and software incrementally. A.s.a. H. software, for example, was developed incrementally and in modules*: inputing and saving cases, normalizing, comparing to past cases,** finding match(s)**, making any predictions, updating matching old cases and adding new cases, making extrapolations** and interpolations, etc., etc. Slowly adding functionality. For hardware like mobile robots I would develop chassis, arms, grippers, sensors, navigation, etc. as individual prototypes, get them functioning, and then seek to combine and coordinate them.

* each tested individually (I favor modularity whenever possible but I do not favor object oriented programming.)

** individually trying out and comparing various algorithms

Thursday, May 1, 2025

When would we need humans in space?

 Military space stations like the American Manned Orbiting Laboratory (MOL) and the Soviet Almaz were rendered obsolete* by robotic spacecraft.** Studies are needed to determine what tasks really require humans in space.*** Any such list is presumably going to be reduced as AI and robotics technologies improve over time. 

* Have studies with Tiangong done anything to change this conclusion?

** Reconnaissance, communication, navigation, weather, intercepter satellites, etc. and, for example, today's Boeing X-37. Note the use of specialist robotic agents once again. Notice also that they are not of a  humanoid form.

*** If the Hubble robotic repair mission had been pursued purely as an experiment then it might have helped to establish some of these limits. And, if you ever wish to repair something (be it by humans or by robots) you should design it with that in mind from the very beginning. (Not to say that "single use" items should not exist. But they stay "single use.")

Sunday, April 13, 2025

U.S. space program 2025

Will there be money enough (1.) for low earth orbit so as to allow a "commercial" space station to financially breakeven, (2.) to fund a manned Mars landing and return, (3.) compete with a manned Zhongguo/Rossiya* moon base, the ILRS, and (4.) do some actual space science? I don't believe it. Once again, try to do too many things at once and you end up doing nothing well.

* "Chinese/Russian"

Monday, April 7, 2025

Hacking toys again

This toy robot* can carry a small sensor pack and an Arduino or Raspberry Pi SBC.

* product model No. 968 "engineering vehicle" Shantou Jiabaile products co. ltd.

This robot* is large enough to carry a sensor package like the one in my blog of 10 February 2025.

* The chassis is taken from an amphibious remote control car manufactured by Spobot and the gripper assembly is from the innobot coding robot kit from Thin Air Brands.
 

Hacking toys has a long and honorable history, the first programmable pick and place robot was built from  Meccano (Erector set) in the mid 1930s. See Meccano Magazine, March 1938, pg 172, and The Robot Gargantua, cyberneticzoo.com under early industrial robots. And the earliest Roomba prototype was made with Lego in 1989.

Tuesday, April 1, 2025

How do you get there from here?

How does one peacefully get from todays hyper capitalism to a just democratic socialist state? I have suggested that one step might be to make all businesses nonprofit. This might be practical since something like half of businesses today are not currently profitable.

A.I. agents

In the mass of hype* that surrounds the world of A.I. research today the definition of "agent" has been lost. An agent is reactive, autonomous, goal oriented, and continuous running. Agents are frequently also communicative, mobile, flexible, and learn and have "personalities." A.s.a. H. A.I. agents have existed for many years now.

* i.e., lies 

Tuesday, March 25, 2025

Physical A.I.

A rebuild of one of my first Lego A.s.a. H. A.I. robots. The original was made using Lego that belonged to Emporia State University and that was left behind when I retired.


 I've had best results with a society composed of 4 or 5 different specialist agents, each having no more than 6 degrees of freedom.

Thursday, March 20, 2025

Manned flight to Mars

I'd be happy to see a project to send a small crew to Mars and back but not at the expense of any science projects. And humans should only fly when they are really needed/useful.

Emporia State University

 I retired from ESU in spring of 2022. The next semester ESU broke the law in firing a number of tenured faculty. Now they are trying to get out of this by asking the state to change the law RETROACTIVELY. This is what capitalism is like. Capitalism cheats. (See my work on value, e.g. chapter 2 of my book Twelve Papers, at www.robert-w-jones.com)

Tuesday, March 18, 2025

full self driving

I don't enjoy driving. I would love to have a truly self driving car. But I am not interested in being a beta tester. And it would be ridiculous to pay to be a beta tester. 

Saturday, March 1, 2025

Hacking consumer robots

 Industrial robots are extremely expensive. As an alternative I hack robots like ArcBotics' Sparki:


Kleenex light

 In a dark room you can see a spot of light emitted when you pull a Kleenex from its box. The tissue is rubbing along and away from the plastic at the opening. Static electricity I assume.

Wednesday, February 26, 2025

Cloud repatriation

Almost all of my Java software was run on the cloud* but I have not used Java for a long time now. I've run BASIC in the cloud a tiny bit but I'm not using cloud computing at all for Python, C/C++, Prolog, or Lisp programs. 

* from a Windows PC.

Monday, February 10, 2025

Sensor packs


 The sensor pack for use on one of my Asa H AI robots.* (I think that having several packs that are smaller than this** is probably best.)

* See, for example, my blogs of 8 January 2018, 1 October 2015, and 4 September 2024.

** On multiple robots.

Friday, February 7, 2025

AI race?

I don't believe that there is only one way to build an intelligent system* and that it will be ideal for all applications/specialties. So I don't think that there is a race that will have a single "winner." In discussing "space races" I have noted issues that I have with "crash programs".**

Also, the first way that is discovered that is capable of doing something is frequently not the way that will prove best in the long run. An example would be DC electric power distribution being replaced by AC.

Is all this hype*** needed to fund AI research? It's diverting resources away from other equally important AI work.

* or one architecture or one set of hardware

** See my blog of 21 October 2024.  

*** lies

Saturday, February 1, 2025

Mars colonization

The purpose of life is to survive and spread so it makes sense to colonize Mars. But this can involve mechanical life, it need not involve human life. 

LLMs today

LLMs currently are a system for learning* large complex functions from example data**. Currently those functions do not represent very long temporal sequences.*** They are also extremely energy inefficient.****

* approximating

** The data should really be cleaned, curated, and presented in the right order.

*** i.e., there is limited short term memory. This somewhat limits the tasks these LLMs are good for. But, again, different architectures are good for different tasks/specialties.

**** this is a rather brute force approach

Wednesday, January 1, 2025

A.s.a. H. question generation

 If A.s.a. generates an output signal that is too weak it will be unable to drive its effector (motor, servo, etc.). But if the output signal is above some preset threshold it can, instead, trigger a question, e.g., "should the xxxxxxx  effector be triggered?" Bystanders may then respond or ignore the question as they see fit. 

An issue with LLM generated computer code

 It is my impression* that LLM generated code requires more debugging as compared to my hand written code. I think most of us hate debugging. Especially debugging other people's code.

* I admit that I don't have enough experience to quantify this impression.



On the simulation hypothesis

First and foremost I am my mind. Our best current model says that my thoughts are electrochemical signals in a biological brain*. The simulation hypothesis would have me be signals (electrical?) in a vast computer. This doesn't seem any less "real". We've already been adding lots of layers to our models of reality*.

* Or, at a deeper level, quantum fields and the like?

Rocket recovery and reuse

For every pound you must invest in the heat shield for the recoverable rocket's orbital* stage you will lose one pound of payload. Does it make sense then to recover simple tankage? Perhaps one should only recover the engines, pumps, and avionics. I.e., something like** ULA's "SMART" but applied to the upper stage***.

* typically second stage 

** may or may not use inflatable heat shield

*** or think drop tanks