Why do we need multiple representations?

 

Robustness. Flexibility. Non-brittleness. Diversity.

 

If one representation fails you can switch to another.

For many-many (many causes/large effects) in causal diversity it is the only possibility.

Each representation has strengths and weaknesses.

Each representation has a bias, assumptions about how to look at the world.

Single representations haven't scaled up to human-level intelligence.

It allows you to examine ideas from different perspectives a key aspect of thinking.

Present day programs are single function, brittle constructions.

The right representation can render a problem trivial or conversely with the wrong representation a problem can be impenetrable.

Symbolic-systems especially logic-based ones are exact there are no near misses.

Neural networks are restricted to their basic associative function they cannot deal with structures or compositions.

Each representation makes a trade-off between expressiveness and tractability.

 

Home.

© CKSoft 2007.