EVOLVING SEMANTIC SYSTEMS
(Organised by The ESSENCE Project)
Edinburgh University 24-28 Aug 2015
Summer School Poster (PDF)
An introduction to the Turing-Inspired Meta-Morphogenesis Project
School of Computer Science, University of Birmingham ----------------------------------------------------------
Expanded notes for tutorial now online:
[NASA artist's impression of a protoplanetary disk, from WikiMedia]
EXTENDED ABSTRACT (Liable to change)
Unsolved problems about natural intelligence
Despite all the successes of AI there remain deep gaps between what AI systems can do and the competences of humans and other animals, for example nest building birds, such as weaver birds, and squirrels that defeat "squirrel-proof" bird-feeders. At present, I don't think any current robot comes close to having the visual, manipulative, and other competences required to build nests made of knotted leaves like the weaver birds shown in this 10 minute BBC video: https://www.youtube.com/watch?v=6svAIgEnFvw
Moreover, current AI language learning systems are nothing like the young deaf children who created a new sign language, as reported in https://www.youtube.com/watch?v=pjtioIFuNf8 . Humans don't merely learn languages: they create languages, collaboratively. But for that, there could be no human languages since initially there were none to learn.
What else do infants and toddlers do that we cannot yet explain? A toddler who can barely speak (aged just over 17 months) seems to be able to do topological experiments with a pencil and a hole in a sheet of card driven entirely by her own internal motivation (I just happened to have a poor quality video recorder running):
It's interesting that after withdrawing the pencil from the hole, she manages to control the motion of the pencil over the edge of the card without looking at the pencil, though she seems to look intently at the hole while directing the pencil point to it, accurately enough to go through first time. Moreover she controls the motion of the pencil point while holding the pencil at the opposite end. The actions seem to be goal-directed throughout, rather than random movements that merely happen to produce interesting results. What selects the goals? Something produced by the genome that "notices" opportunities? There does not seem to be any ulterior motive. Yet presumably she learns from such actions. But what is she learning? Something about the topological structure of 3-D space?
What sort of ontology (for physical objects, spatial locations, spatial relationships, spatial processes, ....) does she need in order to formulate the goals in advance of executing them? What sort of process ontology and action control ontology does she need in order to direct the processes to the achievements of the goals? What sort of language or form of representation could she be using to encode the information about the current state of affairs, the intended goal state and the constantly changing actions that produce that goal state? How do her perceptual mechanisms make use of that ontology? How are the perceptual mechanisms related to motive generation, to plan formation, to fine grained control of movements during exploration or plan execution.
Another question concerns the motivational state with that goal. Where did the goal come from? How/Why do organisms acquire goals? A common assumption is that all motivation must be reward-based (and many AI learning systems depend on this). In contrast, the playful activities of many young animals including humans seem to be motivated without any expectation of reward. There are rewards, but they may come much later when knowledge acquired has been reorganised into some new deep, powerful theory. But the child doing the exploration cannot possibly know that: evolution however may have produced motivational mechanisms to produce such effects. I call that "Architecture Based Motivation" (ABM) in contrast with "Reward Based Motivation" (RBM). The individual performs actions that previously "rewarded" the genome!
I suspect that none of the formalisms known to logicians, mathematicians, AI researchers/Roboticists, neuroscientists or developmental psychologists is capable of playing the required role in modelling or replicating the toddler's achievements. Can we formulate requirements for forms of representation and information-processing mechanisms that are adequate to the task? Would the forms of representations overlap with those used by other intelligent species, e.g. weaver birds, squirrels, and elephants?
There are many human competences that current AI systems are not even close to replicating (as far as I know), for example the processes that led to the mathematical (geometric, topological, arithmetic) discoveries known to Euclid over two thousand years ago (long before modern logical notations and theories had been thought of), and the processes by which a young human who is unable to understand any such mathematical content can develop into a mathematical student who not only understands but who can also discover proofs and theorems without being told about them. Profoundly important discoveries in geometry, topology and arithmetic leading up to Euclid's Elements must have started before there were any mathematics teachers. How? How did the first engineers manage without teachers?
The Meta-Morphogenesis project
One way to try to bridge those gaps in our understanding is through the Turing-inspired Meta-Morphogenesis project, which aims to identify and understand the many transitions in information processing produced by biological evolution since the very simplest organisms or pre-biotic molecules came into existence on a lifeless planet.
A key hypothesis is that a major theme throughout biological evolution is production of new derived construction-kits (DCKs) all ultimately derived from the fundamental construction kit (FCK) provided by physics and chemistry.
In addition to production of new physical materials, new physical designs, and new physical behaviours, derived construction kits also provide ever more complex and varied forms of information processing.
An outline theory will be presented: concrete, abstract and hybrid (concrete+abstract) construction kits produced by evolution and development can help to explain the variety of types of information processing in living things, and help to draw attention to forms of information processing (computation) that have not yet been studied or replicated but which may play important roles in animal intelligence. Some preliminary ideas about evolved construction kits are assembled here (extending the material presented at the tutorial):
The tutorial will give an introduction to the ideas in this project and some preliminary results. One strand that may be of special interest to ESSENCE is the development of an ontology for construction kits and their powers, limitations, and relationships. The above paper has a crude first draft collection of ideas presented informally. This could also contribute to research on ontological requirements for deep, explanatory, scientific theories, and the research questions that lead to them.
It is hoped that more researchers will develop an interest in this large and complex project and help to speed up progress.
The presentation will be highly interactive, with opportunities for participants to contribute ideas as well as questions.
Before the event, this web site will be expanded with more up to date results
and invitations to potential attendees to contribute ideas, or links to related
work, to be added here if appropriate. Anyone who requires more information is
welcome to write to
a.sloman @ cs.bham.ac.uk
Aaron Sloman interviewed by Adam Ford at AGI 2013, St Anne's College Oxford.
Toddler theorems -- building on work of Kant, Piaget, Karmiloff Smith, and many
Ring magic - evidence for untaught topologists
Evolution of vision and language: Some common/linked themes
Some background: an attempt to understand the powerful implications of AI for
philosophy (and vice versa):
The Computer Revolution in Philosophy: Philosophy Science and Models of Mind
(Partly updated version of 1978 book.)
A variant of this tutorial will be presented as a workshop at SGAI-2015.
School of Computer Science
The University of Birmingham