There are many different definitions of Artificial Intelligence in the literature, all are suggestive and insightful. However, at the end of the day, I think there is really one simple enough to be understood and formalised rigorously.
This is John McCarthy’s original definition of AI from 1955: “the science and engineering of making intelligent machines”.
We know what machines are — the set of all computable functions that can be performed by a universal Turing machines will suffice as a definition for our purpose here.
But what qualifies as intelligent? This paper by Legg and Hutter has a collection of ~70 definitions of Intelligence: http://arxiv.org/abs/0706.3639 Towards the end of the paper, the authors asked whether a single definition of intelligence is possible, and proposed this (informal) definition: “Intelligence measures an agent’s ability to achieve goals in a wide range of environments.” The beauty of this definition is that it can be formalised rigorously using algorithmic information theory, as is done in this paper by the same authors: Universal Intelligence: A Definition of Machine Intelligence. Between the two papers, they have accumulated around 1900 citations at last count, so they can be considered authoritative.
Chaining McCarthy’s definition with the Machine Intelligence definition of Legg and Hutter, we arrive at this reasonable and simple definition of AI: “Artificial Intelligence is the science and engineering of making intelligent machines, where intelligence is measured by the machine’s ability to achieve goals in a wide range of environments.”
We largely know how to engineer safe and responsible AI that are narrowly intelligent (in that they can achieve goals in a narrow range of environments), but there is a lot of R&D still to be done for us to understand how we can engineer safe and responsible AI that are widely intelligent. Solving the latter problem will be my research focus for the next little while.
Mental Models 4 Life 