Page 172 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

162 CHAPTER 8 The New AI: Basic Concepts, and Urgent Risks

1. INTRODUCTION AND OVERVIEW
1.1 DEEP LEARNING AND NEURAL NETWORKS BEFORE 2009e11
For many years, the majority of computer scientists believed that the field of artifi-
cial intelligence would never live up to its initial promise, or have a major impact on
technology or the economy. There were periodic efforts to abolish research
programs in that area at the National Science Foundation, based on the view that
all those lines of research never panned out and that other stakeholders should get
the money. Even within artificial intelligence, the field of machine learning was
mostly viewed with disdain until the US National Science Foundation (NSF)
mounted a major cross-cutting initiative to support it [1] as part of a larger initiative
which grew still larger for a few years [2]. Sergey Brin, cofounder of Google, has
reported [3] that the leaders of computer science assured him that neural networks
could never do anything very interesting.
From 1988 to 2008, mathematical neural network research was led primarily by
a partnership between engineers (IEEE) and the International Neural Network
Society (INNS), who organized the International Joint Conferences on Neural
Networks (IJCNN). Another important conference, smaller in that period, was
the Neural Information Processing Systems (NIPS) conference, led by Terry
Sejnowski, a prominent neuroscientist. Substantial advances and substantial appli-
cations in real-world engineering challenges were made in this work [4],but
because of tribalism and vested interests the progress was not well reported in
general education in computer science. The number of inaccurate statements
made about the field outside the field were far too numerous and extreme for us
to address directly. Even today, much of the information sent to high-level policy
makers and to the press comes from the same type of “expert” who provided
false information in the past.

1.2 THE DEEP LEARNING CULTURAL REVOLUTION AND NEW
OPPORTUNITIES
The deep learning revolution of 2009e11 changed the story dramatically. At the
2014 World Congress in Computational Intelligence (WCCI) in Beijing, LeCun
called this revolution “the second rebirth of neural networks,” as important as the
first dramatic rebirth back in 1986e88 which crystallized in the first IEEE
International Conference on Neural Networks in 1988 in San Diego, California.
Immediately after his talks and mine at that conference5, at a time when I still
ran the neural network research at NSF, some people at Beijing whisked me to
the office of a very powerful dean at Tsinghua University and from there to the
National Science Foundation of China (NSFC), which proposed a dramatic new
research push for joint US-China research. I was deeply disappointed when new
management at NSF did not respond well to the proposal, but China did create a
major research push on its own, and I submitted retirement papers to NSF effective
on Valentine’s day, 2015.