Artificial intelligence (AI) and humans are learning from one another as the two begin to collaborate more frequently together. While this may sound like a futuristic concept, experts agree it’s all part of the natural progression of interactive technology.
Why it Happens
Humans learn by interacting with one another, a process we commonly call socialization. Since AI and other smart technologies are interactive (meaning humans interact with them in ways other than giving commands) the two learn from each other’s successes and failures when working together.
According to Briana Brownell, the founder and CEO of Pure Strategy Inc., humans learn from AI by finding “novel” strategies. “A famous example of this is when Lee Sedol played Go against Google DeepMind’s AlphaGo,” she tells Parentology. “During the game, the AI played several moves that were unconventional and looked like a mistake from the perspective of commentators. But they weren’t mistakes.”
The AI was able to find moves that human players would not have. Because of this, the technology was able to win the game.
Learning to be Practical
In the 1950s, Arthur Samuel created an AI program to play checkers. Since the computational requirements to learning the game were extensive, he decided to use an approach called “machine learning” to teach the technology.
“Samuel created a scoring function that could be used to estimate the chance of winning for each board position,” Brownell explains.
Since there’s no exact function to teach the game, Samuel created a shortcut. “He picked six things he thought were relevant to winning: number of pieces on each side, the number of kings on each side, and proximity of pieces from being kinged on each side.”
Samuel would then have the computer repeatedly play against itself to get a better understanding of how those six factors were linked to winning. When the AI finally played a human opponent, it chose moves that would give it the highest chance of winning using just those six variables.
“For human checker players, this is especially relevant because we can also use those six variables as a shortcut to improve our own play.”
Brownell explains that humans can use this same technique in many other areas to improve decision-making skills.
Challenges of Learning from AI
One challenge with learning from AI is that our memory is completely different from an AI memory. Brownell says this makes it much harder for us to memorize exact positions and probabilities.
Another issue we face is that AI and humans are simply different, according to Dr. Tim Lynch, a Robopsychologist and Computer Psychologist. “By learning too much from AI, we lose our humanity and creativity, which is our advantage over the machine,” Lynch tells Parentology. “Let’s not forget that the goal of general AI is to make the machine more human not the other way around.”
Lynch explains that every self-learning AI based on neural nets is trained by subjecting it to large data sets propagated by humans. “These data sets can show human bias or prejudice which is then passed on to the AI,” he explains. “Amazon was using an AI trained on hiring data to screen applicants and ensure only the best most qualified humans were hired.”
As a result, the AI picked all-male candidates because the data it was trained on was biased against female workers. Lynch gives another example where an AI profiling program used by police was biased against certain racial, economic and neighborhoods and often ignored more affluent suspects.
A Brave New World
In the future, the applications that may benefit the most from an AI/human collaboration will probably be business-based. Brownell says marketing, customer experience, market analysis, supply chain management, scheduling, planning and hiring are all areas in which a well-trained AI could shine.
And that may happen sooner than we think. “I believe within the next decade, most people will have both human and AI coworkers,” Brownell says. “Right now AI is more prevalent in our personal lives than in our work lives, but that won’t last very long.”