The Quest For Artificial Intelligence

by Nils Nilsson

Nils Nilsson has been involved in artificial intelligence research more or less since the inception of the field in the late 1950s. (He’s probably best known for his work on Shakey the robot, A* search, and his idea of teleo-reactive planning.) This makes him ideally placed to write a history of artificial intelligence.

This book is a fairly straightforward history, picking out highlights from each of the distinct phases that AI research has gone through in the last 60 years, with a digression in the middle to talk about some more philosophical issues. The book is as comprehensive as is realistically possible in a work of this size, although it becomes a little more fragmentary as we move towards the present–as Nilsson himself says, historians have to cede to more immediate and current sources of information for current affairs, so he just shows a few examples of what’s been going on in the last few years.

A few things struck me quite strongly:

First, there really is nothing new under the sun. So many of the ideas that are important today in AI research were discovered or invented in the very early days of the field, but languished due to inadequate computing resources or other technological limitations or because of lack of funding. Much of modern AI research seems to be a rediscovery or reworking of these ideas in a setting where they can be exploited as their inventors had initially hoped.

Second, I was shocked (although in retrospect that was a product of my own naïvety) by how much AI research has been funded by the military. For most of the 1970s, it seems as though almost the entirety of the US AI research community was funded by DARPA or other DOD elements. It makes sense, but it seems something of a shame that what started as an intellectual quest for the foundations of human consciousness should be largely applied to warfare.

Third, much of the book dedicated to more recent periods is taken up with descriptions of things I would hesitate to call artificial intelligence. Very few people are working towards the goal of creating human level general artificial intelligence any more. There’s much more interest in applications based on machine learning and data mining–the huge volumes of data now available for training these systems makes machine translation and other natural language processing tasks, as well as machine vision, planning and other complex activities, amenable to analysis by relatively unsophisticated algorithms, algorithms that certainly don’t provide any insight into what “general intelligence” is.

For me, perhaps the most interesting part of the book is the philosophical digression about what AI is, and Nilsson’s comments on human level AI. The book is worth picking up for that alone. Although Nilsson is careful not to make any overly judgmental remarks, you can read between the lines to get a sense of what he really thinks.

One slightly odd thing: no mention of Douglas Hofstadter. I would have been interested to see Nilsson try to put Hofstadter’s work on analogies and knowledge representation into the context of more conventional work in that area.