Open Questions

I am creating a Discord channel for discussions that may lead to a theory of intelligence. You know that I have one, but it is important for me to be open-minded about alternative approaches. Therefore, the channel will welcome any suggestions, as long as they respect some requirements. Those are not mine, those requirements are imposed by two very real things - the world and time.

At the core of a good theory of intelligence should be answers to some key questions. The channel will revolve around them. I call those Open Questions. I will try to answer them myself, but it is more important for me to invite answers from the community. As well as discussions of the proposed answers. What is also important is not only how well each answer covers a respective question, but how well all the answers from the positions of one theory connect with each other.

Having a good answer to only one of the following questions is not only insufficient but also dangerous. It may mislead one into thinking in a particular way, trying to find supporting evidence for that. Finding such evidence is not difficult, but will that evidence be relevant? It is better to fail sooner under the pressure of all the questions and requirements. In a way, consider this project as a set of guidelines to develop a theory of intelligence you will be OK with. On top of that, I will try to ensure that all discussions are respectful and helpful.

Consider leaving comments to polish the proposed list.

Requirements and Expectations

The real world is amazing for its variety, complexity, and interactions. To an intelligent agent, it provides two challenges: to categorize the observed phenomena and to figure out the causality rules. All those need to be represented efficiently in terms of both storage and retrieval.

An even bigger challenge is imposed by real time. Not all the decisions and reactions need to be fast, but when there is a need for that, there must also be a way. Hence, an algorithm with logarithmic complexity at worst is expected. There should also be ways to handle more time-consuming tasks.

Open Questions

1. What are categories?

Those are not words. Decision-making relies on categories and may well proceed without any verbalization. Categories, concepts, classes, types - there are many terms. Whatever one you are going to use, try to explain what they are, and how they are used. Remember that they exist for objects, abstractions, actions, relations, facts, situations, etc. They are the starting point for any theory of intelligence.

2. Symbol grounding problem

Then there are words that stand for categories. How do we assign words to categories? What are synonyms and polysemous words?

3. How does categorization work?

You should explain this in terms of a single, core algorithm applied to a particular cognitive task. Humans categorize multiple phenomena extremely fast. This should be a logarithmic complexity algorithm.

4. What is the algorithm of intelligence?

Provide a generalized algorithm that may be applied to various cognitive tasks.

5. Causality

There is no need to explain the various laws here. But we need to explain how we figure out those laws, how we represent them internally, and how we apply them in various forms of thinking. How do categories affect causality? Does causality affect categories?

6. Memory

How do we keep track of categories and words, of causality rules, of objects and facts (general and specific)? It should provide a way to retrieve information fast. How does the memory of specific objects use categories?

7. Information

Shannon explained what he meant by that. You need to explain what you mean by that. Intelligence seems to rely on a different kind of information, semantic one. This may be about the distinction between objects and categories. Or it may be about functions and roles. Or anything meaningful.

8. Thinking

Decision-making, planning, adaptation, inference, intuition - there are many manifestations of thinking. How does your theory explain it, using all of the above? How does intelligence avoid analysis paralysis? Is it symbol manipulation or some other operation with categories?

9. Knowledge transfer

How do references work? What does grammar refer to? How does disambiguation proceed? How do we validate what we hear? What else does it take for knowledge transfer to succeed? For example, there is a saying that some information may only be acquired when a student is ready. On the other hand, an expert thinking about a hard problem may find a sufficient hint in a simple word or gesture.

***

Let me know if any of my readers know how to set up a Discord server and are willing to do that with me. It will speed up the process. But it will happen no matter what. I promise!