FAQ: - superintelligence

Nick Bostrom (bostrom@ndirect.co.uk)
Wed, 5 Aug 1998 01:26:13 +0000

What is superintelligence?

A superintelligence is any intellect that greatly outperforms the best human brains in practically every field, including general wisdom, scientific creativity and social skills. This definition leaves open how the superintelligence is implemented. For example, it could be a classical AI or a neural network, or a combination of the two. It could run on a digital computer, a network of interconnected computers, a human brain augmented with extra circuitry, or what have you. The definition also leaves open whether the superintelligence is conscious and has subjective experiences.

Sometimes a distinction is made between weak and strong superintelligence. Weak superintelligence is what you would get if you could run a human-like brain at an accelerated clock speed, perhaps by uploading a human onto a computer [see "What is uploading?"] If the upload's clock-rate is a thousand times that of a biological human brain, it would perceive reality as being slowed down by a factor of thousand. This means it could think a thousand times more thoughts in a given time than its natural counterpart.

Strong superintelligence denotes an intellect that is not only faster than a human brain but also qualitatively superior. Not matter how much you would speed up a dog brain, you would not get a human-equivalent brain. Similarly, some people think that there could be strong superintelligence that no human brain could match no matter fast it runs. The distinction between weak and strong superintelligence may, however, not be at all clear-cut.

Many (but by no means all) transhumanists think that superintelligence will be created in the first half of the next century. This requires two things: hardware and software.

When chip-manufacturers plan new products, they rely on a regularity called "Moore's law". It states that processor speed doubles about every eighteen months. Moore's law has been true for all computers, even going back to the old mechanical calculators. If it continues to hold true for a few decades then human-equivalent hardware will have been achieved. Moore's law is mere extrapolation, but the conclusion is supported by more direct considerations based on what is physically possible and what is being developed in the laboratories today. Increased parallelization would also be a way to achieve enough computing power even without faster processors.

As for the software problem, progress in computational neuroscience will teach us about the computational architecture of the human brain and what learning rules it uses. We can then implement the same algorithms on a computer. Using a neural network approach we would not have to program the superintelligence: we could make it learn from experience exactly like a human child. A possible alternative to this route is to use perhaps genetic algorithms and some methods from classical AI to create a superintelligences that may not bear a close resemblance to human brains.

The arrival of superintelligence will clearly deal a philosophical blow to any anthropocentric world-view. Much more important, however, are the practical ramifications. Creating superintelligence is the last invention that humans will ever need to make, since superintelligences could themselves take care of further scientific and technological development more efficiently than humans could. The human species will no longer be the smartest life-form in the known universe.

The prospect of superintelligence raises many big issues and concerns that need to be thought hard about now, before the actual developments occur. The big question is: what can be done to maximize the chances that the arrival of superintelligences will benefit humans rather than harm them? The range of expertise needed to address this question extend far beyond that of computer scientists and AI researchers. Neuroscientists, economists, cognitive scientists, philosophers, sociologists, science-fiction writers, military strategists, politicians and legislators and many others will have to pool their insights in order to deal wisely with what may be the most important task the human species will ever face.

Transhumanists tend to want to grow into and become superintelligences themselves. The two ways in which they hope to do this are: (1) Through gradual augmentation of their biological brains, perhaps using nootropics, cognitive techniques, IT tools (e.g. wearable computers, smart agents, information filtering systems, visualization software etc.), and, in the future, neuro/chip interfaces and bionic brain implants. (2) Through mind uploading.

Nick Bostrom
Department of Philosophy, Logic and Scientific Method London School of Economics