Eskildson: Artificial intelligence is coming to Arizona mostly from Lee’s ‘AI Superpowers’

China caught artificial intelligence fever after the world’s best Go player, China’s Ke Jie, lost to Google’s AlphaGo in 2017.

While Go’s rules can be laid out in nine sentences, the number of possible tile placements on a Go board exceeds the number of atoms in the universe.

Loyd Eskildson

Less than two months after AlphaGo’s victory, China’s central government issued an ambitious plan to lead the world in AI capabilities by 2030. The plan includes clear progress benchmarks for 2020 and 2025, and envisions AI playing major roles in improving and expanding Chinese healthcare and its urban security and traffic management.

Their plan is further supported by initiatives in quantum computing, computer chip R&D, and new AI education initiatives in its public schools. Chinese VC investors also stepped up, providing 48 percent of global AI venture funding in 2017 and surpassing the U.S. for the first time.

AI’s big technical breakthrough occurred in 1986 when British researcher Geoffrey Hinton discovered how to efficiently train neural networks modeled after the human brain.

It was called “backpropagation,” and used to calculate factor weights — the centerpiece of “deep learning” computer algorithms that are far easier to program and much more accurate than alternative rule-based “expert-systems” that were “guided” by loading a computer with what experts’ decision guidelines — a very time-consuming and problematic process.

Since Hinton’s discovery, researchers have learned how to “train” AI computers to recognize faces and images, translate voice to print in real-time, operate autonomous vehicles, translate between languages, interpret medical images, skin lesions, etc., trade financial instruments, automate equipment, colorize black-and-white images/videos, make individualized purchase recommendations, grade/correct grammar, assess pupil interest/comprehension/attendance, etc.

Experts now expect the number of applications to grow rapidly.

The AlphaGo victory differed importantly from IBM’s Deep Blue defeat of chess champion Garry Kasparov in 1997.

Deep Blue had largely relied on guiding heuristics from real-life chess champions, the board contained only 64 positions (361 in Go), and AlphaGo taught itself how to play and beat the then world champion.

Many believe the U.S. is far ahead of Chinese AI hardware and software, and that China is unlikely to soon catch up. However, AI development has reached the point where Chinese strengths (much larger, more integrated, faster-growing databases) are now more important than the U.S.’s relative strength in highly-complex R&D.

Dr. Lee also contends that Silicon Valley is far slower at implementation than Chinese entrepreneurs.

True, elite U.S. AI researchers provide the potential to push the field to the next level, but those advances have occurred rarely — “deep learning,” for example, was invented 32 years ago. Thus, the availability of large “training databases” will be the driving force behind AI disruptions of countless industries around the world.

Dr. Lee contends that an algorithm designed by a handful of mid-level AI engineers and fed lots more data usually outperforms one designed by a world-class deep-learning researcher with less data. Ergo, having more of the best and the brightest researchers isn’t as important as it used to be.

PwC forecasts AI-improved productivity will contribute $15.7 trillion in world GDP by 2030 — with China taking nearly double the North American share.

China’s central government has adopted a “Sputnik” response to the AlphaGo victory and is doing everything it can to tip the scales for itself.

Our government, however, is instead slashing funds for basic research, and even lacks goals for matching Chinese pupil achievement levels. (Shanghai 15-year-old pupils are estimated to be three years ahead of their American peers, and about 1.5 years ahead in science. Pupils in other large Chinese cities also perform far above U.S. levels.)

PwC estimates AI will eliminate an estimated 40+ percent of jobs at all levels in the U.S. by the early 2030s. Income inequality will substantially worsen, and tax revenues dive precipitously.

China’s populace is used to and accepting of surveillance, not likely to strongly oppose AI. Conversely, U.S. AI development is hindered by high-tech workers opposing participation in surveillance or AI applications with possible military application, Teamster union opposition to self-driving trucks, privacy concerns about healthcare and education data, and resistance to statewide pupil-testing programs.

The biggest potential threat to rapid U.S. AI development could come from Trump administration efforts to reduce immigration and H1-B etc. visas. Over half of Silicon Valley STEM workers with a bachelor’s degree or above are foreign born. Pew Research Center estimates nearly one million temporary non-agricultural foreign workers were employed in 2013.

What type of jobs are most likely to survive AI? Dr. Lee and others suggest physicians trained to empathetically deliver serious diagnoses such as cancer, teachers trained to support computer individualized instruction systems, AI professionals, skilled craftsmen, firefighters, police, and positions requiring creativity and/or cross-functional thinking (eg. lawyers).

A crisis is coming. Completion for AI supremacy and profits will change Arizona at a pace and breadth never before experienced. Most of today’s students will enter a greatly changed work environment. Educators must lead the preparations. Suggestions include providing students with an understanding of AI databases, AI utilization/programming, robotics, probability, genomics, and critical thinking courses.

Room for these new and/or enlarged offerings could be attained by reducing or eliminating requirements for advanced algebra, trigonometry, foreign languages, grammar/writing, and calculus courses that will be less relevant in an AI era.

The most obvious need is to substantially improve math and science achievement levels so that Arizona students can successfully compete against those elsewhere in America and China.

Education costs also must be drastically reduced — individualized computer-assisted learning offers and MOOCS provide a way to do that.

Greatly improved pupil-testing and data-base collection will also allow much more objective evaluation of education platforms, curricula, and personnel.

Cities should benefit from improved traffic management and policing (facial and license-plate recognition, more driver monitoring).

Finally, state level leaders must determine how to operate an economy with high unemployment and underemployment rates. Two prime candidates for cost reductions are education (about 10 percent of GDP) and healthcare (about 18 percent).

Editor’s note: Mr. Eskildson is a member of the Scottsdale Unified School District community and a Paradise Valley resident.

You are encouraged to leave relevant comments but engaging in personal attacks, threats, online bullying or commercial spam will not be allowed. All comments should remain within the bounds of fair play and civility. You can disagree with others courteously, without being disagreeable. Feel free to express yourself but keep an open mind toward finding value in what others say. To report abuse or spam, click the arrow in the upper right corner of the comment box.

Facebook Comment