Artificial intelligence is intelligent, human-like behavior exhibited by machines and software. While it used to be the subject of many popular twentieth century science fiction movies, now it seems like only a few decades from becoming an actual reality. Along with the word ‘intelligence’, AI was developed to not only mimic the human brain but also its natural structure. Artificial neural networks (ANNs) were inspired by their biological namesake, and are one of the many similarities AI and human intelligence share. ANNs play a vital role in machine learning and cognitive sciences through deep learning, which gives machines the “ability to learn without being explicitly programmed”. But while this may seem akin to human intelligence, there is a specific difference that makes it distinct: decision-making by AI today is bound by a single directive. Humans function much differently and are always constantly comparing and re-prioritizing goals in their everyday lives. This is not to mention the fact that we often create goals on our own. It’s obvious that there is still much work that needs to be done to bridge the gap between AI and human intelligence.
That’s not to say current advancement in AI is not stunning. AlphaGo, Deep Blue, and Watson are all well-known examples of artificial intelligence technology. Although each was originally designed for a specific, singular purpose (playing Go, Chess, and the game show Jeopardy! respectively) they demonstrated great mastery by handily defeating human professionals in their respective fields. While this may seem like passing entertainment to some, I’m sure fellow engineers have no problem imagining their use in other, more generalized commercial applications. In fact, derivatives of Watson are already currently ‘employed’ by the Memorial Sloan Kettering Cancer Center in New York as decision aides for lung cancer patients. In addition to that, AlphaGo’s triumph over 9-dan Go professional Lee Sedol just last March convinced the South Korean government to invest $863 million in AI research over the next 5 years! That should silence any AI viability naysayers.
But when can we definitively say that AI has reached the complexity of human intelligence? Alan Turing established a test in 1950 to measure a machine’s capacity to ‘behave intelligently like a human being’. The standard version involves a human evaluator’s ability to distinguish between another human and AI based solely on their responses. “How closely can the AI pass for being the human?” is the final question posed by this Turing test. Today, this test comes at odds with John Searle’s Chinese Room argument, which states that regardless of how intelligently a machine behaves, its program cannot give it a human “mind”, “understanding”, or “consciousness”. To visualize his point, he describes a locked room with a non-Chinese speaking person inside. If that room is filled with “to-Chinese translation books”, the non-Chinese speaker would be able to carry a fluent written conversation in Chinese with any Chinese-speaking person outside. In essence, any Chinese-speaking outsiders could be fooled into thinking the room resident knew Chinese when in they were simply following written translation instructions. In actuality, there was never any understanding of the Chinese characters at all!
I am of the belief that, with the machines we use today, artificial intelligence cannot hope to achieve a perfect human “mind”. While the end result of machine behavior and actions may mimic ours, I sincerely doubt that the reasoning behind them would be anything but the end result of massive computation. It’s difficult to imagine machines today acting out of more complex human emotions like regret, appreciation, compassion, or love. In that sense, I guess I would agree with Searle’s Chinese Room argument regarding today’s computers. But that’s not to say I don’t think a perfect AI can never exist. I just believe that our interpretation of modern computing would have to change drastically in order to make it happen.
dzaragoz 