Glossary
What is artificial superintelligence?
In short
Artificial superintelligence (ASI) is a hypothetical AI that surpasses any human in almost every field: science, invention, strategy and communication.
AI basics · artificial superintelligence (ASI)
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a new student in your class who solves every problem first, draws better than an artist and invents more fun games than all the other kids together. People like that do not exist. But scientists discuss whether a computer smarter than all people at once could appear one day. For now it is only an idea, and adults argue about how to make such a machine help people.
In plain words
Today’s AI is strong at separate tasks, and the next step is usually called strong AI, an AI equal to a human. Superintelligence goes further: a machine that thinks faster, deeper and wider than any expert in any field. It could make scientific discoveries and improve its own code by itself. No such system exists, and nobody knows if it will. People discuss it for two reasons: some expect new medicines and technologies from it, others fear that people could not control it.
For business
Superintelligence is a topic of forecasts, not budgets, but it already affects the market. Some labs openly name it as their goal, and their race drives the prices of models and computing. The practical advice for an executive is simple: buy for today’s tasks, with a measurable return within months, not for the promise of a future mind. If a vendor sells “a step toward superintelligence”, ask for a calculation of the money effect on your own process. The second point is about rules: regulators discuss limits for powerful models, and this can change access to foreign AI services. Such changes are tracked in our monthly AI market review.
For a specialist
N. Bostrom, in the paper “How long before superintelligence?” (International Journal of Futures Studies, 1998), defined superintelligence as an intellect much smarter than the human brain in practically every field, including scientific creativity, general wisdom and social skills. In 1965 I. J. Good named a possible path to it: the intelligence explosion, where an ultraintelligent machine designs even better machines. From this comes the central research problem: control and alignment of a system whose decisions people cannot check directly, known as scalable oversight. Researchers also discuss limits on computing and tests for dangerous abilities before a model is released. The company Safe Superintelligence Inc. calls safe superintelligence its only goal and its only product. There is no agreed test that would place a system in this class.
Example
At a strategy session, a director proposes to postpone AI adoption “until superintelligence appears”. The finance director shows that an assistant for the purchasing department will pay back on current models in a few months, and the project starts.
Related terms
Questions and answers
How is superintelligence different from strong AI?
Sources
Not sure whether your company needs this?
AI solutions map