Which Cognitive Skills AI Need to Reach General Intelligence

Algorithms are still missing conceptual flexibility, resilience and common sense

Photo by Possessed Photography on Unsplash

Through academic contests these past 20 years(Imagenet, Alphago challenge, Word2Vec…), AI has shown huge successes in specific fields.

Whether object, voice or word recognition, deep learning algorithms from major tech companies have shown accuracy close to perfection.

But these results hide imperfections that could soon hinder their progress. According to Melanie Mitchell in A Guide for thinking humans, AI lacks the basis of a generalist intelligence that understands and adapts to situations, and that could significantly slow down data learning.

Here are four cognitive skills that machines lack, and the skills they need to pass these obstacles.

Object recognition lacks conceptual understanding

Based on the perceptron model, the first object recognition algorithms were fairly simple systems: neurons that each input one pixel of the image and indicate 0 or 1 depending on the value of the image selected. They were algorithms that learned to define the pattern of a specific image (dog, cat, human…), by learning the relevant values to recognize them…