Non-spatial Abstract Entities

I'll start with the things designers need to think about, of which a subset will be required by Fido (unless it too becomes an intelligent designer!).

The abstract entities designers need to think about divide into several main classes, with various sub-classes, some of which Fido will think about:

• Abstract entities whose existence does not depend on intelligent animals or machines or anything with the capability to refer, for example

  times, time-intervals, temporal relations, and many timeless non-spatial entities that are relevant to thinking about spatial concrete things and processes but are not themselves spatially located, touchable, perceivable, etc, including numbers, shapes, colours, correlations, propositions, questions, proofs, explanations, analogies, causal relations, categories, taxonomies, patterns, similarities and differences between shapes or other things, .... and

• Abstract entities that are states, processes or events in virtual machines, for instance

  aspects of the mental state of an animal or intelligent machine (of the sort described here.) However it is arguable that the sematic contents of those mental states are not themselves mental. Thus someone may be puzzled by an unanswered question, which is a mental state of that person, but the question is not something mental and is not a part of that person's mind, since (in almost all cases) the same question could be considered by another individual, who may or may not be puzzled by it.

• Abstract entities that are not parts of mental states and processes within individuals but are products of such processes, including such things as

  virtual machines in computers, the internet, and various social and economic entities including rules, laws, family structures, money, obligations, rights, etc.

• Abstract entities that are patterns or commonalities between components, states, processes, in intelligent individuals, but whose existence does not depend on the existence of such things.

  Thus we can talk about the notion of having a concept as something that refers to a wide variety of patterns of capabilities within information processing systems, including animals and robots, without assuming that any such things need to exist for the notion of having a concept to exist, just as there need not be any dodecahedral physical objects for the notion of an object having a dodecahedral shape to exist. Of course if we are talking about concepts then that presupposes that we exist, but that's trivially obvious. The possibility of talking about them existed, like the possibility of intelligent animals and machines, existed before any instances existed.

What sorts of abstract (non-mental) entities will Fido need to know about?
The requirements for Fido could vary enormously depending on the sort of household in which the robot is deployed. For instance, one could imagine a robot helping to look after a disabled mathematician or chess player, learning to discuss abstract kinds of mathematics or abstract properties of chess strategies.

Likewise if Fido is a companion for a philosopher or psychologist, or is looking after a precocious child who starts asking questions about what's going on inside insects, dogs, humans or robots when they perceive and act on their environment, then Fido may need to be able to talk about things like concepts, propositions, beliefs, desires, sensing, learning, etc.

For purposes of this project we wish to restrict attention to generic capabilities required of a domestic robot, and in particular to the subset of capabilities that might be found in a typical human child aged up to about five years at most. That still leaves a wide range of possibilities, but would exclude the ability to discuss advanced mathematics or advanced chess strategies, or many of the philosophical and psychological problems that might fascinate Fido's owner or charges.

Of course it might turn out that providing the more generic and fundamental capabilities found in young children suffices, if done right, to provide most of the information-processing platform required for learning to think about the more abstract things.

But in early experiments there will inevitably be simplifications that mean it is not 'done right'.

Exactly which sorts of abstract entities the robot might be expected ot know about, or be capable of learning about, would need to be investigated by considering a number of scenarios in depth.

In particular we could expect Fido to know about locations, routes, questions, answers, rules, and many abstract entities that have physical instances, e.g. physical properties and relations, like colours, shapes.

What Fido can do with these various abstract entities and how they are represented within Fido are topics for further investigation. This will depend enormously on what we expect Fido to be able to do, and to talk about.