Terminology: Superhuman narrow AI vs AGI
Consider again an image captioning AI which can tell you which animal, bird or plant is in a given picture. Despite being narrow, by ‘training’ on vast amounts of labelled pictures, this AI can become extremely good at guessing the species, including some very obscure ones. It can easily beat the best human zoologist (who studies animals), ornithologist (who studies birds) or botanist (who studies plants) combined. But this AI will never discuss a football game or poetry with you, because it has no concept beyond the questions it is trained on.
In recent years, we have seen some truly impressive narrow AI systems that demonstrate superhuman performance in one task, such as beating the world champion in the board game of Chess or Go. Yet, these systems are completely incompetent at other tasks, such as interpreting a simple question in natural language.
No matter how good a narrow AI system becomes at its narrow task, it can never spontaneously exhibit general intelligence. But it can still exceed human performance on the narrow task. When we say an AI is ‘superhuman’, we mean it is better than humans at a given task. This is not to be confused with Artificial General Intelligence (AGI), which surpasses humans in everything.
This isn’t a new phenomenon. A pocket calculator can perform arithmetic much faster than you, but it cannot do anything else. As a specialized tool, the pocket calculator is very useful in the hands of a human who knows how to use it. The same can be said for machinery, such as power tools that help us fasten screws quickly, or diggers that help us move dirt without breaking a sweat. They are physically more powerful than us, but are themselves very narrow.
When armed with numerous narrow, yet superhuman AIs, powerful actors in the world—states, corporations, armies—can become ever more powerful. In the same way that the world’s leading universities, companies and intelligence agencies are empowered by the collections of bright human minds that they employ, these organizations can also aggregate human ingenuity with superhuman AIs to create what management theorist Thomas Malone calls ‘superminds.’
References
Goertzel, B. Artificial General Intelligence: Concept, State of the Art, and Future Prospects. Journal of Artificial General Intelligence; Vienna 5, 1–48 (2014).
Malone, T. W. Superminds: The surprising power of people and computers thinking together. (Little, Brown Spark, 2018).