AI definitions, and narrow vs. general vs. super intelligence

This article discusses definitions of Artificial Intelligence, and the distinction between narrow, general and super intelligence.

AI definitions, and narrow vs. general vs. super intelligence

Definition of AI

Google "definition of AI" - go ahead - here is the link - you'll get pages and pages of definitions, some humorous, some silly, and some that are impossible to understand. Here is a small sample of them and I'll close with the one that I like the most as somebody who approaches new technology from a practical perspective.

Wikipedia Definition (edited): "In computer science, artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans and animals. Leading AI textbooks define the field as the study of "intelligent agents": any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals."

Tesler's Theorem: "AI is whatever hasn’t been done yet."

Personally I like this one from D. Fogel: "Any system...that generates adaptive behavior to meet goals in a range of environments can be said to be intelligent."

My goal with this series (and blog and podcast) is to stay pretty grounded and practical and to leave the navel-gazing and philosophical debates to others and focus on what is practical and practicable in the here and now. So, for our purposes that will be as good a definition as any!

Artificial Narrow, General, or Super Intelligence

This is a much simplified version of the growth curve that was originally illustrated over at WaitButWhy in their excellent post on AI.

It is widely assumed that there will be at least two, if not three major eras in artificial intelligence and for the most part they are labelled Artificial Narrow Intelligence (ANI), Artificial General Intelligence (AGI), and Artificial Super Intelligence (ASI).

In a field full of disagreement, surprisingly most people do agree that we are in the first major era of Artificial Narrow Intelligence (ANI), meaning machines that have access to lots of compute cycles and vast amounts of data, can be trained to recognize patterns and identify states (dog vs. non-dog, person vs. non-person, etc.) This set of skills turns out to be useful for a lot of things like converting speech to text, or text to speech, translating between languages, computer vision, etc. and that in turn can be used to improve production lines, do back-office type work, process documents, provide customer service, drive autonomous vehicles, read legal case law, and generally do a number of specific tasks better, faster, and more completely than humans.

Once we get past narrow AI, it is assumed that the next phase will be Strong or General AI - a system that can work across many domains (as opposed to narrow single domains) and that can do things such as construct its own problems, and do its own planning and coordinating.

Kai-Fu Lee, a leader in the field of AI, has said in many of his talks that he believes that "we've only had 1 really major breakthrough in the first 60 years of AI (deep learning), and that it could take 10-20 more big (and unknown) break-throughs to get to Artificial General Intelligence." (BTW, many people also equate this to Human Level AI but there is debate on if those are the same or different.) There are many out there who believe that once AIs are as smart as humans, that things will seriously escalate, leading to...the era of Artificial Super Intelligence. This is a wild crap-shoot and is likely in Skynet / Overlord territory. Humans will be like pets or ants comparatively speaking on the intelligence scale. AIs at this level may spend no more time thinking about humans than we do about insects. We won't know in my lifetime so I'm not going to worry about it.

For the purpose of this series of articles and indeed this blog and podcast, I'll be focusing squarely on Artificial Narrow intelligence - and what we can do with it today and tomorrow to address real-world problems.

[This was first published on on March 9, 2020 and on LinkedIn on March 27, 2020 and was moved here in June 2020 to the new site.]