Artificial intelligence extends what human beings are able to do

The argument

The case that artificial intelligence is good for humanity rests first on capability. Proponents argue that the recurring pattern of beneficial technology is the extension of a human faculty beyond its natural limit: writing extended memory, the lens extended sight, the engine extended strength. Artificial intelligence extends the faculties of pattern recognition, inference and language, and proponents argue that these are the faculties on which almost every form of human progress depends, so extending them raises the ceiling on all of it at once. The mechanism proponents emphasise is the removal of scarcity in expert attention. Diagnosis, legal explanation, tutoring, translation and technical advice have always been limited by the number of trained people available and the cost of their time, which is why most of the world has gone without them. Proponents argue that a competent system can supply such attention at a cost approaching nothing per use, and that the people who gain most are those who previously had no access at all rather than those who could already afford a specialist. Proponents extend the argument to discovery. Much of science consists of searching spaces too large for human inspection, such as the space of possible molecules, protein shapes, materials or proofs, and proponents argue that systems able to search those spaces compress work that would otherwise take generations. They accept that the benefits are not automatic and depend on distribution, and argue that this is a question about who gets the technology rather than about whether the technology is good. On their account a tool that enlarges human capability in medicine, learning and science is beneficial, and the task is to spread it rather than to regret it.

Premises

Counter-arguments

Critics argue that the analogy with earlier tools is what carries the conclusion and that it does not hold. Writing and the lens extended faculties while leaving judgement with the person using them, whereas these systems substitute for judgement, and a faculty that is not exercised atrophies. They add that near costless use describes the technology, not its ownership: the systems are built and controlled by a small number of firms, so the gains accrue there first and reach others only on terms those firms set. Scarcity of expertise, on this account, is replaced by dependence.

Rejecting the premises

[Rejecting P1] Earlier tools extended faculties while leaving judgement with the user, whereas these systems substitute for judgement, so the analogy that carries the premise fails. [Rejecting P2] The scarcity of expert attention is a problem of training and funding, and a technology that bypasses it may entrench the shortage rather than relieve it. [Rejecting P3] Near costless use is a property of the technology and not of its ownership, so the benefit reaches others only on terms set by those who control the systems. [Rejecting C] The premises establish that the technology has large potential benefits, which is consistent with its overall effect being harmful.