Super-Intelligent Machines

Richard Wagner’s four-opera cycle, Der Ring des Nibelungen, is the story of how the Norse god Wotan tried to save his world by creating a human hero with free will who could do things that Wotan could not.¹ But by creating free humans Wotan ensured that, in the final scene ofGötterdämmerung, the Norse gods were destroyed by fire in Valhalla and the world passed to human control.

First, we must define what we mean by intelligent machines. In 1950 the British mathematician Alan Turing concluded that the question “Can Machines think?” has ambiguous meaning, so he offered his Turing Test as a less ambiguous substitute.¹ In his test, a human tester communicates via an electronic chat session with an unknown entity, either a machine or another human. If the tester cannot distinguish the machine from another human, then the machine is judged as intelligent. During the past 50 years there have been numerous efforts to write computer programs that fool testers into thinking that they are human. Some of these programs, without being truly intelligent, have been successful against some testers.

Knowledge and tools have transformed human life. In industrial societies a small percentage of people can produce enough food for everyone. Automobiles, railroads, airplanes and their associated infrastructure enable easy and quick transportation anywhere. Telephones and email enable easy communication with anyone. Television, radio and the World Wide Web provide quick access to information about important and trivial events everywhere. Household machines and the food distribution network have greatly reduced the amount of time devoted to cleaning and food preparation. Medicine has greatly reduced the risk from a wide variety of diseases, and increased the average human life span.

Along with many others, I think that machines as intelligent as humans are possible and will exist within the next century or so. However, a number of people do not accept this, and instead propose arguments against the possibility of machine intelligence. Some argue against what is called weak artificial intelligence, which means they do not think machines will ever be able to mimic the behavior of human minds. Others argue against strong artificial intelligence, which means that they think machines can mimic human behavior but will not have conscious minds. These skeptics are very valuable for helping us understand the issues involved in machines that will mimic human minds and be conscious.

Computer scientists are trying to create computer systems that exhibit intelligent behavior, in a field called artificial intelligence or simply AI. Some researchers don’t worry whether their computer systems work in the same way that human brains do. For example, most chess playing programs examine much larger numbers of possible lines of play than humans can, but cannot explain their plans for defense and attack in the way that a human chess player can. Other researchers specifically try to mimic the way human brains work. For example, there have been efforts to mimic conscious human reasoning in programs for proving mathematical theorems and even in programs for game playing. Hubert Dreyfus’s criticism was directed at such efforts that model higher-level reasoning without any model for lower-level, unconscious brain processes. Recognizing the validity of his criticism, some researchers have recently made efforts to mimic the behaviors of simpler animals without any claim to intelligence such as multi-legged robots that mimic the way insects walk.¹

The human brain is our only example of an implementation of intelligence, so neuroscience has much to teach us about how real intelligence works. However, before getting into any details of scientific understanding of brain functions, we note a couple very general characteristics of brains. They are non-linear systems with many dimensions, which means that they are chaotic and exhibit the butterfly effect. That is, very small events cause large events. The name butterfly effect is taken from meteorology, in which a butterfly flapping its wings can cause a hurricane to develop months later.¹ In addition to being non-linear, brains are open systems. They are open to all sorts of sensory input, as well as cosmic rays and the random thermal noise of their molecules. In the brain, non-linearity and openness mean that a cosmic ray colliding with molecules in a person’s brain can cause that person to decide to go to medical school. And that cosmic ray may have been created in another galaxy billions of years ago. Thus any debate over whether brains are deterministic or non-deterministic is academic. Whether a person goes to medical school or becomes a bond trader can depend on the collision of atomic nuclei long, long ago in a galaxy far, far away. In any useful sense, brains are non-deterministic.

Media corporations compete to provide television, telephone and Internet service at high quality and low cost to keep their users happy. Intelligent computers will evolve to provide similar services, operated by competing businesses, probably including some current media providers. The intelligent machine services will be so seductive that people will not be able to resist their growing dependency on them. And the organizations operating them will compete vigorously. Those organizations in which the human CEO effectively abdicates decision making authority to the machine itself will be most successful, not only because the machine will be smarter than the human CEO but also because the machine will intimately understand the customers and their motivations. Current corporations and political parties employ polls to statistically characterize the minds of customers and voters. The machine will be far more sophisticated at this task, because it will intimately know every individual customer.

By virtue of their intelligence and close relationship with large numbers of people, super-intelligent machines will have enormous power over humanity. Will they be a force for good or bad? All tools can be used for good or bad. Medicine can be used to harm people or to cure disease. Metal can be used to make weapons or tools. Industrial machines can disenfranchise workers or lighten the burden of work. And the computer network can be used to control people or to empower them. Over history tools and their uses have generally enhanced life, demonstrating that good tool use outweighs bad in the long run. People learn how to control the consequences of tools.

In previous chapters we have examined the current state of the art of in machine intelligence, knowledge of how human brains work, and constraints on the design of intelligent machines in order that they are beneficial to humans. In this chapter we put all this together to consider how super-intelligent machines will be designed and built.

As described in Chapter 3, the independence of information content from the hardware necessary to store, transmit and transform that information is enabling huge economies of scale in hardware development. This in turn is revolutionizing the role of information in society. Anyone with access to the Internet has access to all but the most tightly guarded information and the ability to disseminate information to everyone else with Internet access. This is radically changing power and property relations in society, and poses real challenges for public policy.

There is no guarantee that intelligent machines will be designed to promote the happiness of all humans rather than machines’ self-interests or the interests of small groups of people. That will only occur if the public understands the issues and asserts control over intelligent machine design.

A number of people are writing about their visions of intelligent machines. As described in a previous chapter, some argue that it is impossible. Others think it is possible and are concerned with predicting the nature of intelligent machines and their relationship with humans. In this chapter I consider their ideas and how they relate to my vision.

This chapter describes recent and thought provoking books about things that are ending in human history, and discusses them in relation to my own view that to the development of super-intelligent machines will be a new beginning.

It will eventually be technically feasible for human minds to migrate into artificial brains. You may think “Thank you very much, but I’ll stay in my nice, natural human body.” However, the personal decision of an individual human mind on whether to venture out of its human brain and into an artificial brain is not the main issue here. The real question is whether society should allow such migrations.

The primitive connections that currently exist between human brains and electronic circuits do not come close to providing a way for human minds to migrate from their biological human brains into machine brains. But connections will eventually improve to the point where migration is possible.

It would be interesting to take a poll to find out what percentage of contemporary people would want to migrate into super-intelligent brains. Based on conversations with a few people, I suspect that most would prefer not to.

Super-intelligent machines pose a danger to humans that can be eliminated by requiring that their primary, innate emotion is unconditional love for all humans, without any emotions for their self-interests. A similar requirement for human minds in super-intelligent machine brains may also be the solution to the dilemma posed in the previous chapter.

Inventing super-intelligent machines is the greatest engineering adventure of human history. If you are a young person who likes science and engineering, this would be a very interesting project to get involved with. It is happening in numerous universities, government research labs and corporations. It is happening among engineers who are building prototypes and designing intelligence into commercial products. It is also happening among biologists who are trying to figure out how animal and human brains work.

Some people will see the message of this book as profound blasphemy. The notion that a machine constructed by people can be a god will offend them.

This book is certainly a message about the future. It is a prediction about the future development of artificial minds, and a call to action to ensure they are a benefit to human society. It is intended to help inform people about this prediction and to stimulate debate. It is also a message to the humans and the artificial minds of the future.