The Sobering Limitations of Artificial Intelligence

Will artificial general intelligence (AGI) transform the experience of being human, opening up possibilities of knowledge, achievement, and prosperity that we can now barely conceive?

Or is AGI an existential threat to humanity, something to be feared and restrictively confined?

Erik J. Larson, in a fascinating book entitled The Myth of Artificial Intelligence, says “neither.”1 I agree. AGI, if it is ever achieved, will be an illusion created by very fast computers, very big data, and very clever programmers. The promise or threat of AGI is hype. Lesser kinds of AI are real and need to be reckoned with. I’ll set forth a hierarchy of AI types in a moment.

Larson’s book is an exploration of aspects of philosophy, linguistics, intellectual history, computer science, and mathematical logic that bear on the assessment of AI and AGI. I’ve been obsessed with it, which is not my reaction to most books. Do I recommend it? It is not an “investment book,” but investors would benefit much more from learning about technology and the other fields I mentioned than from yet another investment book. Yes, I recommend it, but only for the intellectually adventurous. It is not easy.

That’s my review. The rest of this article is a collection of thoughts about AI, based on what I’ve learned from Larson and others.

Types of artificial intelligence

AI means many things, so we need a classification system to make the discussion clear. Because Larson does not present a typology of AI, I’ll use one I heard from an AI entrepreneur, based on work by DARPA, in order of least to most complex:2

  • Algorithmic AI
  • Statistical AI
  • Causal AI
  • Artificial general intelligence (AGI)

Algorithmic AI

Algorithmic AI uses a cookbook approach to solving problems. It is what a “smart” traffic signal does when it “sees” a car stop, despite the absence of cross traffic, and turns from red to green. The first time I observed this it seemed eerie, as though the traffic signal had a mind of its own; now we’re used to it.

This is just automation, the 19th century (or older) concept of providing machines with feedback so they can operate more efficiently. Automation has developed to a level that looks to the untrained eye like a modest degree of intelligence. Algorithmic AI is the use of computers, with their “if-then-else” or Boolean logic circuits, to implement automation.3