Fluid Intelligence

An example question from the Raven’s Progressive Matrices standardized fluid-intelligence test (the correct answer is D4). Credit: Ken Forbus

As I have been talking about progress in artificial intelligence in the previous posts (and I will further discuss AI in tomorrow’s post) it seems appropriate to report a news coming from Northwestern University where a team of researchers have built an AI system showing “fluid” intelligence.

They have created an AI based program that performs in the 75 percentile of Americans taking intelligent logic tests based on visual problem solving (that means that the program is better than 75% of Americans taking the same tests).

Most of AI programs focus on recognizing objects, shapes and are now pretty good. Little attention has been put in creating a program that understand the relations among objects and that can infer rules out of these relations. This is what visual problem solving is about (see figure). You first need to detect the various shapes, than understand their mutual relations and then abstract these relations into rules that can be applied to different shapes. This “abstraction” is an important aspects of intelligence, it is what let us transfer what we learn into a field to concept that can be applied to a different field.

It also requires to understand the concept of analogy (to know when you can apply something to a different field (e.g. you can learn to walk on grass and then you apply that skill to walk on sand, but you cannot apply that knowledge to walk on water…).


This kind of intelligence is known as “fluid” intelligence since it has to twist, morph and adapt to different situations.  It is something that humans are pretty good at, whilst many other animals are not.

Author - Roberto Saracco

© 2010-2018 EIT Digital IVZW. All rights reserved. Legal notice