Page 216 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

1. Introduction 207

FIGURE 10.1
Examples of mimicking human performance and intelligence during the past 500 years;
from Leonardo Da Vinci’s mechanical knight design (1495) through Kempelen Farkas’
chess playing robot the “Turk” (1795), to IBM Deep Blue (gray in print version) defeating
the world chess champion Garry Kasparov (1997), and Google AlphaGo defeating the Go
world champion Lee Sedol (2016).

neuromorphic computational theories and practical implementations [9]. Indeed, deep
learning uses multilayer neural network architectures trained by the backpropagation
learning algorithm, which has been developed over 40 years ago [25]. These results
provide firm theoretical foundations for a new wave of AI with neurally motivated
deep learning. The successes of deep learning in the past years were facilitated by
several main factors, including easy access to vast amounts of data, often referred
to as “Big Data,” and the availability of cheap computing technology based on very
powerful chips: Graphical Processing Units (GPU), Tensor Processing Units (TPU),
and neuromorphic chips manufactured by IBM, Intel, QualComm, Cambricon, and
other industry leaders. Deep learning can efficiently adjust billions of parameters
on massively parallel computer hardware, leading to superior performance in various
applications, such as image processing and speech recognition. Aiming at future
developments in the field, this chapter reviews some important lessons learnt from
the operation of human brains and human intelligence, which can be incorporated
in next generation AI algorithms and hardware implementations.