September 19, 2016

DAVIES | Illusory Intelligence

Print More

The technical possibilities of tomorrow are just as incredible as those of the 1950s because they are real. Simultaneously everything is within reach and nothing. We use new technologies but few people understand their function. Machines, programs and devices on the horizon, rushing towards us, will be far less widely understood than would those of the 20th century, had they come to pass.

It is conceivable that most people, with a modicum of study, could understand the functioning of a color TV or a flying car depicted in a pulp science fiction book. In contrast, now, and into the future, many of the technologies we create will be beyond the understanding of the average person. With the advent of machine learning and artificial intelligence we may only be able to understand how we built something, not how it functions.

In the past human understanding was shared across the group — everyone within the tribe understood pretty much everything the others did. After the spread of metal tools, most members of a village could likely tell you how to go about fashioning an axe, even if none of them understood the science explaining quite how the process proceeds. The total sum of human knowledge was a fraction of what it is now but any individual could easily understand a greater proportion of it.

Today, no one could describe the function of all technologies they use or explain all phenomena they witness, though the knowledge of most of them exists within some human mind. If artificial intelligence continues to develop, in the future we will create systems that generate knowledge of their operation, independent of us. An artificial intelligence could focus its computing power on understanding the operations of the complex, mutating interlinkages of which it is composed. Such technologies would understand themselves better than we do.

They will also likely develop knowledge of human interactions and societies that is separate from human understanding. Depending on the AI’s task, such information could be beneficial to its goal — a new financial market model for an AI directed to forecast stock prices for instance. Because of its source within an AI such knowledge may be beyond the capability of the human brain to fully comprehend.

Thoroughly and forcefully centralized understanding has always presented a threat to human freedom. An entity, be they man, party or machine, professing to be better placed to select the ends which an individual should pursue results in tyranny. A belief in the superior strength and accuracy of their knowledge of the individual’s ends provides a justification for the compulsion of action contrary to the individual’s own desires.

Artificial intelligence will usher technology through even greater leaps but its power over our decisions must be tightly circumscribed. If AI is deployed as some central planner, akin to the economic planning that the communist experiments of the last century engaged in, with power over the distribution of resources within society, we are likely to permit it to define its own goals in achieving our public distributional ends. In so doing, we threaten to subjugate the individual’s understanding of their own ends to the inhuman planner’s understanding of the correct means to achieve its goals.

For society to heed the decisions of such an AI rather than deferring to individuals’ choices would deal a great blow to human freedom. An artificial intelligence possessed of both greater knowledge and power, direct or indirect, that would likely result in the enforcement of its own logical conclusions should be resisted by all.

Alex Davies is a senior in the College of Arts and Sciences. He can be reached at [email protected]. Have I Got News For You? appears alternate Tuesdays this semester.