Aristo and the Friendly Forest Squirrels
On September 5, 2019, The New York Times splashed a brash headline on the front page of its business section:
A.I. Can Now Handle this 8th-Grade Test. Can You?¹
Once again, we are dragged into a race between man vs. machine, with artificial intelligence charging hard up the back side. Sheepishly, we complete the sample questions in the Times, knowing that Aristo, a machine learning program in Seattle, has already bested our score in the 8th grade science test. It will soon master a 12th grade science test, or so they claim.
DeepMind’s AlphaZero program defeated world Go champion Lee Sedol. IBM’s Watson defeated Jeopardy! champions Ken Jennings and Brad Rutter, and Deep Blue defeated grandmaster Garry Kasparov in chess. And now the Aristo program knows more 8th grade science than I do.
Following the headlines, we have to conclude that the human mind is a puny dishrag up against the Machine Matrix Masters of artificial intelligence. As a species, we should throw in the towel and take our place in the back row of history’s failures with the dodo and the dinosaur.
Except. The headlines make the wrong assumption. A race with artificial intelligence is a myth. Human intelligence and artificial intelligence are not the same. Deep Mind’s AlphaZero chess and Go playing derive from impressive deep learning neural networks, making 60,000 calculations and more per second.(Another less successful chess algorithm makes 60 million calculations per second.)² Human beings could never do that.
But, why would we? What evolutionary challenge did we ever face that required instantaneous numerical calculations? We evolved to work together in our chosen groups — families, tribes, and communities. Our instant calculations are fight or flight in the face of danger. We never staked our futures on the deliberative calculations of rook, pawn, and knight. Our survival never depended on being the first to buzz in with “Who’s buried in Grant’s tomb?”
As psychologist and neuroscientist Gary Marcus points out, IBM’s Watson won at Jeopardy! because:
…About 95% of the answers in Jeopardy! turn out to be the titles of Wikipedia pages. Instead of understanding language, reasoning about it and so forth, it was mostly doing information retrieval from a restricted set…³
We don’t try to outrun electrical current. So, why should we be impressed because electrical current can zoom through Wikipedia pages ahead of us?
But this 8th grade science test is different.
Like other computers, Aristo, the mechanical champion of the 8th grade science test, used Watson-like information retrieval on most of the test. However, Aristo didn’t just recall facts. It made a foray into human reasoning and logic. Here’s one of the logic questions that Aristo solved:
Which change would most likely cause a decrease in the number of squirrels living in an area?
(1) a decrease in the number of predators
(2) a decrease in competition between the squirrels
(3) an increase in available food
(4) an increase in the number of forest fires ⁴
Aristo named #4 as the correct answer: squirrel population will decline with an increase in forest fires. It’s impressive for a machine to figure that out. It requires knowledge of squirrels, animals eating squirrels, squirrels eating nuts, and squirrels living in trees. It may be 8th grade science, but it’s nothing that your average 4th grader doesn’t know.
But before we make Aristo forest ranger of the year, consider that Aristo was only trained to answer multiple-choice questions. Any questions with graphics or diagrams were excluded. And no essay questions.
Aristo cannot answer a general question like “What happened to the squirrels?” It cannot visit the forest and draw any conclusions about the dearth of squirrels. And, of course, Aristo has never seen a squirrel. Nor, can it mentally bounce from forest squirrels to squirrely kids to Rocky the Flying Squirrel.
Aristo cannot observe the world to gain new insights. A child notices that squirrels like to be in trees and concludes that squirrels must live in trees. But, Aristo has to be taught that squirrels live in trees — over and over, thousands of times.
Aristo can only answer based on the information fed into it. So, why not feed all the information in the world into Aristo or GPT-2 or another algorithm?
Then, Aristo will outscore us on a 12th grade science test, a PhD test, a personality test, or any other test imaginable. Aristo-type algorithms will have “read” every book in the world. It will know everything, but not in the human sense.
Certainly, no human “knows” everything. In fact, the most frequent human mental function is forgetting, dismissing information. We take in a constant stream of information in every moment, but we choose to remember very little. So, the algorithm seems to surpass humans again.
But to what end? What good is it to know everything? The value of information is only in how you use it in life, how you connect disparate information to achieve goals. In short, it’s not knowing everything, but knowing what you need and knowing how to connect between disparate ideas. Consequently, knowledge doesn’t come from information, it comes from experience.
In theory, the algorithm that gorges on endless information — reading every book ever written and digesting every morsel on the internet — will also know every experience. But how will it know what is true? How will it know what is beneficial or destructive to human society? How will it know what to remember and what to forget?
Without priorities, the algorithm gives equal weight to Hitler’s Mein Kampf and Martin Buber’s I and Thou. Microsoft’s chatbot, TAY, had to be shut down within hours after its launch because it was taught by trolls to be racist, anti-Semitic, and misogynistic.⁵ TAY learned, but what it learned was evil.
Humans rely on teachers, mentors, and elders to identify what to remember and what is important. And, pushed by curiosity, we learn by testing our observations in the world.
Like a student, artificial intelligence appears to be climbing the ladder from 8th grade science in 2019 to high school science in 2020. The ladder leads to what artificial intelligence developers call “the holy grail” — machines that have an intelligence equivalent to humans, artificial general intelligence. And from there…
Geoffrey Hinton, cognitive psychologist and one of the first designers of artificial neural networks that allow machines to learn, asks about artificial general intelligence: “Why do we need it?”⁶
We don’t. Yet, billions of dollars are being spent to achieve artificial general intelligence, to make artificial intelligence the power steam drill that laid human thinking to rest with John Henry.
We won’t collapse from too much thinking. Quite the opposite: failure to use parts of the brain leads to mental atrophy. Unused neurons, synaptic connections, and neural networks are modified, redirected, or simply pruned by the brain. In people who lose their sight, most of the visual network is taken over by the auditory network. Other parts of the visual network — unused — atrophy. If you were once fluent in a second language, but you don’t use it anymore, those neurons are re-directed.
As artificial intelligence takes over human thinking tasks, our brains become idle and, thereby, enfeebled.
So, let Watson win at Jeopardy. Let DeepMind’s algorithms rout humans from the gameboards of chess and Go. Let algorithms calculate the value of Pi to 60,000 places.
Meanwhile, we can hypothesize, generalize, and produce conjectures. We can live with ambiguity; and we can handle the unexpected with our mental flexibility. We can walk into the forest and deduce the fates of squirrels without the benefit of multiple-choice questions or endless information.
As economist Gary Smith pointed out in his book, The AI Delusion:
[T]he real danger is not that computers are smarter than us, but that we think computers are smarter than us.⁷
Machines are useful. But they do not equal human intelligence. And, if they do someday, we have to ask, like Geoffrey Hinton, “Why do we need it?
________________________________________
¹ Cade Metz, “A Breakthrough for A.I. Technology: Passing an 8th-Grade Science Test”, The New York Times, Sept. 4, 2019, https://www.nytimes.com/2019/09/04/technology/artificial-intelligence-aristo-passed-test.html
² David Silver, Thomas Hubert, Julian Schrittwieser, Demis Hassabis, “AlphaZero: Shedding new light on chess, shogi, and Go” DeepMind.com, Dec. 6, 2018, https://deepmind.com/blog/article/alphazero-shedding-new-light-grand-games-chess-shogi-and-go
³ Gary Marcus, interviewed by Martin Ford, Architects of Intelligence, Pakt Publishing, 2018, p.312
⁴ Metz, Ibid.
⁵ Gary Smith, The AI Delusion, Oxford University Press, 2018, p.37
⁶ Cade Metz, “With $1 Billion From Microsoft, an A.I. Lab Wants to Mimic the Brain”, The New York Times, July 22, 2019, https://www.nytimes.com/2019/07/22/technology/open-ai-microsoft.htm ?module=inline
⁷ Smith, Ibid. p.237
Dan Hunter is an award-winning playwright, songwriter, teacher and founding partner of Hunter Higgs, LLC, an advocacy and communications firm. H-IQ, the Hunter Imagination Questionnaire, invented by Dan Hunter and developed by Hunter Higgs, LLC, received global recognition for innovation by Reimagine Education, the world’s largest awards program for innovative pedagogies. Out of a field of 1200 applicants from all over the world, H-IQ was one of 12 finalists in December 2022. H-IQ is being used in pilot programs in Pennsylvania, Massachusetts, Oklahoma, North Carolina and New York. He is co-author, with Dr. Rex Jung and Ranee Flores, of A New Measure of Imagination Ability: Anatomical Brain Imaging Correlates, published March 22, 2016 in The Frontiers of Psychology, an international peer-reviewed journal. He’s served as managing director of the Boston Playwrights Theatre at Boston University, published numerous plays with Baker’s Plays, and has performed his one-man show ABC, NPR, BBC and CNN. Formerly executive director of the Massachusetts Advocates for the Arts, Sciences, and Humanities (MAASH) a statewide advocacy and education group, Hunter has 25 years’ experience in politics and arts advocacy. He served as Director of the Iowa Department of Cultural Affairs (a cabinet appointment requiring Senate confirmation). His most recent book, Atrophy, Apathy & Ambition,offers a layman’s investigation into artificial intelligence.