What is AI anyway?
In the 1960s (1970s*) advanced "automatic train control" was implemented for the Bay Area Rapid Transit (BART) commuter rail system. The newly coined term "artificial intelligence" was then only a few years old, having first been used by Dartmouth professor John McCarthy in 1956. But from the perspective of the 1950s, ATC was clearly an example of artificial intelligence.
Automatic train control moved a task to a computer that had previously required human cognition -- starting and stopping a train safely and making sure that it stopped at precisely the right spot on a platform to allow passengers to get on and off. This is a useful illustration of the problem we have with the term "artificial intelligence." As McCarthy initially used the term, the idea of applying computation to any task that we think of as requiring human cognition (such as train control) can be called artificial intelligence. BUT once the task has been proven to be one that can be done by a computer, we can see that it is a task that doesn't require human cognition... so is it artificial intelligence after all?
It's difficult to be rigorous about whether a machine really 'knows', 'thinks', etc., because we're hard put to define these things. We understand human mental processes only slightly better than a fish understands swimming. - John McCarthy
One direction that people have tried in defining "artificial intelligence" is to equate these computational approaches to human thinking, saying that "true" AI is one that can do all of the things a person would do, in the way a person would do them. And some AI systems today (large language models for example) are starting to exhibit those kinds of human level behaviors. But this is also likely to fail as a definition -- computational "thinking" may never work the same way as our human faculties, and this may not even be desirable. Should we try to exactly emulate human thinking in a machine? Or are there more productive non-human ways to think?
领英推荐
In the end the term "artificial intelligence" may only be useful in the sense that it points toward tasks that we still require humans to perform but hope that machines might someday perform. The term may simply define a perpetually aspirational state.
How we got here: The phrase artificial intelligence is now over 50 years old and has been used over that time to describe a series of different computational approaches to performing tasks. These tasks, before a computer did them, required human cognition but as we develop computer programs that can do those human tasks, the definition of artificial intelligence has shifted to the remaining tasks that still require humans. It is easy with 20/20 hindsight to say that the automated train control that BART implemented in the 1960s is not "artificial intelligence" but it surely would have been so from the vantage point of 1950. Likely we will continue to debate what is "true" AI until one day our computers are explaining the concept of human intelligence back to us...
*while the design of the BART system began in the 1960s, the system opened to the general public in the early 1970s (Thanks for the fact check James M. Spitze )
HFS Research Executive Research Leader | Generative and Agentic AI, AGI & Automation | Web3 | Metaverse | HFS Generative Enterprise & Ecosystem
2 年Could apply to the 'intelligent' element of 'intelligent automation', no?
Emeritus Professor
2 年"The idea of applying computation to any task that we think of as requiring human cognition (such as train control) can be called artificial intelligence. BUT once the task has been proven to be one that can be done by a computer, we can see that it is a task that doesn't require human cognition... so is it artificial intelligence after all?" Nicely put!
Emeritus Professor
2 年"The idea of applying computation to any task that we think of as requiring human cognition (such as train control) can be called artificial intelligence. BUT once the task has been proven to be one that can be done by a computer, we can see that it is a task that doesn't require human cognition... so is it artificial intelligence after all?" Nicely put!
What’s next?
2 年With the dawn of each era the first thing we do is apply the tech breakthrough that defines the new era to things we were doing in the previous ones. So the machines of the industrial were applied to war, textiles and agriculture. Info tech was applied to calculating ballistic trajectories to make cannon fire more accurate. Etc. With AI we’re applying computers to perform tasks that previously could only be done in ‘wetware.’ Where each era gets interesting is when the new tech is applied to things we hadn’t even considered doing in the previous era. I wonder what AI will be used for in 20 years??
Managing Partner, Levreg Partners LLP, Chief Revenue Officer, Board of Directors
2 年Great article Ted. Seems humans are more interested in discussing just how "smart" AI has to be to actually be considered AI vs. asking if there "are more productive non-human ways to think".