In 2017, Google DeepMind developed an artificial intelligence (AI) program called AlphaZero. It was programmed to play chess against an earlier developed program called Stockfish. The difference between the two, Stockfish, the then dominant program in chess, was programmed with all the moves that could be made in chess matches and it made choices from this database. AlphaZero was different, it used logic of its own informed by the ability to recognize patterns of moves across a vast series of possibilities, many not conceived by human minds. It learned from these patterns of possibilities, actually playing against itself to build this knowledge base. This is the new AI, filled with possibilities and consequences associated with independent unbridled computer “thinking”, seeing a wider option of possibilities than humans.

In early 2020, Massachusetts Institute of Technology (MIT) announced they had discovered a novel antibiotic that could kill all strains of bacteria, which until then had been resistant to previous antibiotics. Previously this work was done by human researchers through trial and error testing. MIT had a team of AI researchers develop an AI program which created a set of 2,000 molecules, looking at all the characteristics of each including such factors of types of bonds it contains and their unique ability to inhibit bacterial growth. AI then was fed a library of 61,000 molecules to encode which included FDA drugs and natural products. AI then predicted which would be effective as antibiotics, which did not look like any existing antibiotics, and which were predicted to be non-toxic to humans. One molecule they named halicin was identified as the only one to meet all requirements. With human researchers this process would take years, even decades. With AI, it takes a fraction of that time and expense.

AI can train itself hundreds of thousands to billions of times a day (or hour). At this speed, time for direct human feedback becomes impractical, so programmers often automate such functions. Programmers need to carefully define how AI operates and simulates reality. AI measures the quality of the outcome and provides a means to improve outcomes, allowing them to learn.

The challenges with AI include situations where we allow them to act on the outcomes. An extreme example involved Russia which at one time considered allowing using AI to determine its nuclear response to building military situations. Much like the movie War Games (1983), does AI recognize human factors or the concept of using nuclear threats as a deterrent to military actions of your foe? There lies one of the differences between human action and AI action.  

So we are at a time similar to the Enlightenment in 15th Century Europe. The printing press was developed, allowing knowledge to be shared widely throughout the world, instead of experience being lived and knowledge transmitted orally. For some, the intake of knowledge was faster than they could process and control, so they stuck with the old way. For others, they found ways to adapt to the increased flow of knowledge and figured ways to harness it to make their lives easier and more productive.

Leaders in this technological transformation have indicated there needs to be a global collaboration to set standards on how AI can be used. In what situations can it implement solutions with limited human interaction, and what situations must human intervention and supervision of solutions be required. And how did AlphaZero do playing chess against reigning chess champion Stockfish? AlphaZero won 28 games, drew 72, and lost none. AI will improve each of our lives in some way, but we must chart that path forward.

Dr. Steve Patchin is Superintendent of Hancock Public Schools. Programs he has contributed to creating include Mind Trekkers and CareerFEST, helping students explore their talents and associated careers in STEM. His research has focused on increasing development of self-efficacy in individual students.