The Emergence of Machine Learning as a Rupture Technology for Artificial Intelligence

James L. Crowley, Professeur Emeritus Grenoble INP MIAI AI

Institut Chair on Collaborative Intelligent Systems
14h00 Friday 11 Feb 2022

Abstract

Turing defined intelligence as human-level performance at interaction. After
more than 50 years of research, Machine Learning has provided an enabling
technology for constructing intelligent systems with abilities at or beyond
human level for interaction with people, with systems, and with the world.
In this talk I will review the emergence of Machine Learning as an enabling
technology for building systems with human-level intelligence.
 
Starting with a historical review of the of multi-layer perceptron, I will
describe how back-propagation combined with massive computing power and
planetary scale data have created the rupture technology known as deep
learning, and how this technology enables not only pattern recognition but
also signal generation and universal function approximation. I will trace
the emergence of deep learning as an enabling technology for computer
vision, robotics, and speech recognition and review recent advances such as
Generative Adversarial Networks and Deep Reinforcement Learning.
 
I will then describe how the auto-encoder, originally invented as a
distributed algorithm for independent components analysis, has recently
empowered the emergence of a revolutionary technology for natural language
processing known as Transformers. Transformers, such as Google's BERT and
OpenAI's GPT-3, have the potential to unlock all recorded literature as a
source for self-supervised machine learning.  I will discuss how
transformers can be used to build realistic systems for vision, robotics,
and natural language interaction and can potentially enable the emergence of
collaborative intelligent systems for scientific discovery.