Jason Lim is a Washington, D.C.-based expert on innovation, leadership and organizational culture.
Can AlphaGo cry?
.jpg?w=728)
By Jason Lim
Growing up in an era where “go” was the epitome of an outdated game from a bygone era played by old men with long pipes emitting thin trails of smoke, I never imagined the day when this ancient game would become chic again and engage in a “mano a mano” with one of the most advanced Artificial Intelligence (AI) engines in the world.
According to many, Lee Se-dol’s loss to AI in go was a watershed moment in the evolution of homo sapiens itself, marking the turning point in which human beings would quickly become anachronistic and useless in the new world where AI will replace humans in most jobs. This is naturally leading to alarming warnings against Terminator-like scenarios, caution against technological elites in the shadows ruling over the rest of the humanity, or the emergence of AI-enhanced humans more akin to androids than people.
While others are asking these fundamental questions about the survival of human species, I want to ask a simpler question: Can AlphaGo cry?
While AI’s performance in go against a human player was impressive, it needs to be seen in its proper context. Go is a specific game governed by specific rules within a specific space. In other words, it can be finitely defined. In this finite space, risks can be calculated and placement of stones can be optimized.
Brown University computer scientist Michal L. Littman tells Tech Insider: “If it’s a set of fixed rules, with players acting according to preexisting behaviors, then it becomes sufficiently well-defined problems that you can optimize against and the machine will do better than people. If it’s fluid and changing and contextual, it’s much harder to say what it means to do well.” Which means, “If there’s a social component with players playing together, it’s not clear.”
Logic prevails in the go space; AI can do logic well. But what about spaces where emotions rule the day? How about morality? Ethics? Contextual decision-making in which the correct answer might not be the right answer? What about where human beings spend most of our time and energy on: relationships with other human beings?
These are far more interesting questions for me when talking about AI. We know from neuroscience that emotions are more powerful than logic.
The limbic system of the brain ― also termed the reptilian brain ― is the brain’s processing center for the brain. It’s far older than the neocortex ― which gives us our reasoning capability and separates us from other animals ― the logical part.
We like to think that we rationally think through situations and weigh various options according to some logical cost benefit analysis before making a choice. This is the assumption behind all economic studies ― that we optimize utility in our decision making. Well, anyone who lived through the Great 2008 Recession and witnessed how banks make their risk decisions might want to think differently. Emotions don’t come into play in this model.
We know that this is wrong. We actually make emotional decisions and then use our rational faculties to justify them. When a piece of information enters into our system, it actually gets processed more quickly by the limbic system before it registers in the neocortex. This means that your “rational” decision is already pre-colored by your more “primitive” emotional context. In other words, there are no purely logical decisions. It’s all emotional.
In fact, Antonio Damasio, professor of neuroscience at the University of Southern California and an adjunct professor at the Salk Institute, famously observed that people with brains with damaged emotional processing centers can’t actually make even simple decisions. They can tell you in minute details the pros and cons of various restaurants, but they can’t decide which restaurant to go for dinner. In other words, emotions provide the social construct that is necessary for a person to make a decision.
How does this apply to AI? It means that AI can’t make a decision that is meaningful in a social context. It can make a decision to place a Go stone in a specific location to optimize its probability to end up with more stones at the end of the game. But it can’t make a decision to go to a French restaurant because your wife is feeling nostalgic about that trip she took to Paris when in college and you want to let her know that you love her.
This also means that AI can’t yet be truly “creative,” in that creativity involves a logical leap beyond predefined borders to view things in a different social context. AI can optimize, but it can’t imagine possibilities that are outside the logical set within the existing finite boundaries.
In other words, we know that AI can be Spock. Can it also be Kirk? Not yet, and we know who the captain was.
Jason Lim is a Washington, D.C.-based expert on innovation, leadership and organizational culture. He has been writing for The Korea Times since 2006. Reach him at jasonlim@msn.com, facebook.com/jasonlimkoreatimes or @jasonlim2012.