3
$\begingroup$

So I read somewhere there are 2 different views of Artificial Intelligence.

  • One is the normal conventional approach where we use ML and AI logical inference programs to mimic a human brain.
  • Another view is that we cannot create an intelligent agent just by the above procedure. We humans or animals are only intelligent due to our social interactions.

If we think about it, the second view has a lot of credibility since a brain in a jar is not at all intelligent and will not evolve things that make an animal possess intelligent traits like language, logical reasoning, etc. Whereas an ant or bee colonies are apparently intelligent due to their collective information gathering even if their brain are basically hard-wired logical sensors which just perform an action on a given stimulus.

So my questions are:

  1. Are all this artificially intelligent agents/programs being created or will be created in near future, just a set of rules and probabilities?
  2. What is the current progress and views on collective/swarm intelligence?
  3. And what is the level of intelligence or brain capabilities required by an agent which if works collectively (like bees/ants broadly super-organisms) can exceed normal human intelligence?

EDIT: If we look at it nodes in a neural net are also kind of ants/bees. They just perform a hardwired calculation and output the result. Can we think of an ANN as swarm intelligence?

A collective answer explaining the reasoning for the views is highly appreciated. Thanks in advance!!

$\endgroup$
3
$\begingroup$

First, it is not possible to fully define or duplicate the actions of a conscious agent using only AI theory derived from Church-Turing. In other words, the human brain is not a type of Turing Machine. This can be proven.

Are all this artificially intelligent agents/programs being created or will be created in near future, just a set of rules and probabilities?

No. These are common methods in AI but they are not sufficient for cognitive behavior.

What is the current progress and views on collective/swarm intelligence?

We have not had a general theory of human reasoning since the Dartmouth conference in 1956. Because of this prolonged lack of theoretical explanation some have fallen back other explanations such as pan-psychism, spontaneous emergence, or things like swarm theory. Swarm systems can produce fairly elaborate behavior but they have the problem of lacking an independent, global pattern. This means that anything complex with a swarm must have a fractal-type pattern since the overall pattern can only be a result of an accumulation of local patterns. This does not describe human reasoning.

And what is the level of intelligence or brain capabilities required by an agent which if works collectively (like bees/ants broadly super-organisms) can exceed normal human intelligence?

As I've already stated, this is not possible with a swarm system.

If we look at it nodes in a neural net are also kind of ants/bees. They just perform a hardwired calculation and output the result. Can we think of an ANN as swarm intelligence?

Not exactly. A swarm system has local independence. In other words, every piece of a swarm is the same as every other piece. This is true even when individual units are specialized such as having queen, drone, worker, and warrior specialties in an ant colony. In contrast a neural network can have cross-linkages that provide global pattern recognition. But, this too has a problem when applied generally. For example, a Boltzmann machine has extensive cross-linkages but this causes it to stop learning when it is scaled up. So, we have the Restricted Boltzmann machine with no cross-links within the same hidden layer. In other words, neural networks are capable of behavior beyond that of a swarm system but still well below what a brain can do.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Comments are not for extended discussion; this conversation has been moved to chat. $\endgroup$ – nbro Mar 7 at 4:47
3
$\begingroup$

Swarm intelligence was first demonstrated by Craig Reynolds 1987 in the Boids / Flocking domain. There is a huge amount of agents which are interacting together. From a game-theoretic standpoint, swarm intelligence depends on the problem which has to be solved. For example, if the aim is to pick&place an object with 100 ants at the same time, than the question is, what the position of every ants has to be. So it is less a question of biology and neurons but more an optimization problem.

In general it is difficult to solve swarm like problems, because the first step is to generate a blackboard, in which the swarm stores his information. For example, ant #1 can only see a small part of the world, and ant #2 can see a another part. So both visual information must be matched together for creating the overall map.

Answering the second question “What is the current progress in swarm intelligence?” is simple. The Pixar animation studio is releasing every year a new movie. That is a good example, what the current technology looks like. Swarm Characters: The Principle of Distinctness in the Character Analysis

| improve this answer | |
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.