Training curriculum for physical AGI

The learning curriculum* is a much under appreciated part of AI research and development. My current sequence of training for an AGI is something like:

1. First expose the agent to smells, wind, tastes, touch, wetness, temperature variations, light, sound, dust, IR, colors, and train the camera/ANN to learn shapes, numbers, letters, faces, people, plants, and animals.

2. Next teach grasping, motion, line detection, fire detection, acceleration, near and far, hunger, and hardness.

3. Then, approach/retreat, search, pain, carry, collision, hill, path, raise and lower, and object.

4. Then tool use and the self concept.

Training will differ for various specialist agents of course.

* See my blogs of 18 July 2014, 18 May 2015, and 29 January 2015

Why collective leadership

There is ample evidence that groups make better decisions than individuals do. See: F. Galton, Nature, 75, pp 450, 1907.    J. Surowiecki, The Wisdom of Crowds, Random House, NY, 2004.    R. Larrick and J. Soll, Management Science 52, pp 111, 2006.   L. Hong and S. Page, Proc. of the Nat. Acad. of Sci., 101, pp 16385, 2004 and Economic Theory 79, pp 275, 2025.   C. Davis-Stober et al, Decision, 1, pp 79, 2014.    D. Prelec et al, Nature 541, pp 532, 2017.   I. Douven et al, Artificial Intelligence, 275, pp 235, 2019 and Collective Intelligence 5, 2026

Exploring the contents of consciousness

A common model of consciousness* is the blackboard model. At each time step every layer** of the A.s.a. hierarchy outputs its N most active cases/categories to the blackboard using something like: 

1010 OPEN "f:asahb.dat" FOR APPEND AS #5

1011 FOR case = 1 TO N

1012 PRINT #5, level; case; activity(case)

1013 NEXT case

In turn, each level/layer of A.s.a. H. reads the file/blackboard for input (the M strongest activations).

In this model the asahb data file contains the contents of consciousness recorded as a function of time.

* See my blog of 29 June 2011

** See my blogs of 10 February 2011 and 14 May 2012 as possible simple examples.

Social Pluralism

There is not just one "correct" model of the world*. Rather, pluralism argues that it is better to have a number of different overlapping models*. This is true for models of society as well as scientific theories. I therefore favor a society of intelligent agents** and collective leadership***.

The scientific research community is an especially successful example of such a collective intelligence.**** To function properly it needs to be "open source".*****

* See my blogs of 17 August 2012 and 26 September 2010 and Bull. Am. Phys. Soc.,vol. 55, #2, February  2010. Also pages 381-382 of Kelly's The Logic of Reliable Inquiry, Oxford Univ. Press, 1996.

** Humans and/or A.I.s.  See also my blog of 16 September 2010.

*** Perhaps involving things like my blog of 10 September 2023. 

**** Despite various issues such as those noted in my blogs of 26 August 2011, 3 February 2012, and 2 May 2026.

***** See my blog of 20 February 2026.