The Artificial Intelligence Historian

CMU tribute

Pamela McCorduck passed away last month. The New York Times obituary notes her interactions with many builders of the field of Artificial Intelligence from its infancy to its present state.

Today we remember her and talk about AI’s near future.

I knew McCorduck through her late husband, Joe Traub, who we memorialized in 2015. He became the head of the CS department at Carnegie Mellon University in 1971. She also moved to CMU where she became an English teacher. Per the above quote by Brown, she helped make AI real by writing a number of books on its history.

The NYT obit quotes something McCorduck wrote in her 2019 memoir, This Could Be Important: My Life and Times With the Artificial Intelligentsia.

“For 60 years, I’ve lived in AI’s exponential. I’ve watched computers evolve from plodding sorcerer’s apprentices to machines that can best any humans at checkers, then chess, then the guessing game Jeopardy!, and now the deeply complex game of Go.”

It is hard to project the future of an exponential, however. The best way I can try is to align it with my own field.

AI Movers and Shakers

At CMU McCorduck got to know the AI pioneers like Turing Award recipients Herbert Simon and Allen Newell and Raj Reddy. She already knew Edward Feigenbaum who said:

She was dumped into this saturated milieu of the great and greatest in AI at Carnegie Mellon—some of the same people whose papers she’d helped us assemble—and decided to write a history of the field.

The book was Machines Who Think: A Personal Inquiry Into the History and Prospects of Artificial Intelligence. Said Simon:

She was interacting with all the movers and shakers of AI. She was in the middle of it, an eyewitness to history.

I wish I knew more of her thoughts on the movers and shakers in Theory. She was well-versed in complexity of the dynamical-systems kind, and her husband’s work bridged to “our kind” of complexity. The title of her third novel, Bounded Rationality, speaks to both kinds of complexity from its setting at the Santa Fe Institute. This has led me to musing on the difference between AI and Theory.

AI Beats Theory

I never have worked on AI problems of any kind. The closest I ever came is I took a class at CMU as graduate student from Newell. He was a fun lecturer and the class was interesting. But I always worked on Theory. I must reflect a bit on why AI is so successful and Theory is less so.

Pinterest source

Let’s start by saying that a field of research is determined not by who works in the field. Not by the tools that the field uses. It is determined by the problems that the field works on. AI is different from Theory because of the problems that it studies. This is the fundamental difference:

AI looks at whole problems; Theory looks at sub-problems.

What do I mean? AI studies problems that are concrete, that are big, that are as close to real problems as possible. For example, how to play Go or how to recognize images of faces. AI looks at problems that humans actually wish to solve:

What move to make in this Go position? Or is this an image of X or Y?

Theory looks at sub-problems. We look at a real problem and then identify some part of the problem that is hard to solve. We then try to invoke clever methods that show that this sub-problem can be done more efficiently that was previous known. This is hard in general. Is fun to work on in general. And leads to a beautiful field of study. One that is deep and rewarding.

But Theory loses to AI. The issue is that no one really may wish to solve the sub-problem. This is real demand to solve the whole problem, but not the sub-problem. This is the fundamental advantage that AI holds over Theory.

AI Future

Public figures such as Stephen Hawking and Elon Musk have expressed concern that full artificial intelligence (AI) could result in human extinction. The consequences of the technological singularity and its potential benefit or harm to the human race have been intensely debated.

For our own part, we have wondered whether an AI can take over in theory research. This puts a second light on possible meanings of “problems” in another quotation by McCorduck from her memoir, as related here:

“We can’t now say what living beside other, in some ways superior, intelligences will mean to us. Will it widen and raise our own individual and collective intelligence? In significant ways, it already has. Find solutions to problems we could never solve? Probably. Find solutions to problems we lack the wit even to propose? Maybe. Cause problems? Surely. AI has already shattered some of our fondest myths about ourselves and has shone unwelcome light on others. This will continue.

…

When people ask me my greatest worry about AI, I say: what we aren’t smart enough even to imagine.”

Well, we can only talk about things we can imagine now. We can discuss facets of life that already outsource decisions to technology, such as high-speed stock trading and several flash–crashes it has caused.

But looking ahead, what is one near-term application area as a litmus test for the impact of AI? We think many will agree with our looking to self-driving cars. In taking over driving decisions, the AI expressly aims to reduce the evils of impaired or aggressive drivers. There have been mishaps during development, sure, and the algorithms have not yet demonstrated robustness against possible deceptions. That is to say:

• We can already see teething problems with this tech and imagine more along the same lines.

• But can we project structural problems with the driverless paradigm whose concrete forms we have not imagined?

Open Problems

What are your thoughts on the near future of AI?

Our condolences go out to Pamela’s family and associates.