Reading 11: All Hail Our AI Overlords
Okay, maybe it’s too early to welcome our AI overlords — although Roko may disagree — but the rise of AI is certainly occurring. Artificial intelligence has continued to improve over the last few decades, and there is no indication of it slowing down any time soon.
What is AI?
Artificial intelligence, quite simply, is the sub-field of computer science that attempts to develop algorithms that seem to resemble intelligence somewhat as we as humans recognize it. Now, as the article from Computerworld discusses, there are different types of AI. There is strong versus weak AI and there is narrow versus general AI. Strong AI accurately simulates human reasoning and thinking while weak AI simply works but does not attempt to use the same methods of thinking as humans. Narrow AI is developed to achieve a specific task while general AI can accomplish many tasks because it has the ability to reason in general.
Feasibly, a strong and general AI would resemble a human’s intelligence quite well, but we certainly are not there yet technologically. So currently, artificial intelligence looks a lot different than human intelligence. However, that certainly does not mean it isn’t intelligent.
AI Beating Humans At Our Own Games
Today’s AI is definitely intelligent. In fact, it’s often beat us at our own games. DeepBlue beat Garry Kasparov in chess in 1997. IBM’s Watson beat Ken Jennings and Brad Rutter at “Jeopardy!” in 2011. Google’s AlphaGo beat Lee Se-dol at Go in 2016.
As a avid “Jeopardy!” fan myself, the win by Watson over Ken Jennings and Brad Rutter was extremely impressive. From watching both of them compete — and steamroll over the very talented competition —in the show’s Battle of the Decades Tournament, it’s incredibly obvious that both have really done their research and are extremely good at the game. However, in Double Jeopardy, it didn’t seem that they really stood a chance against Watson.
While sure it seems like an AI should be really good at trivia since many of us look up answers using Google’s search engine AI, there is so much more to “Jeopardy!” than that. The clues on the show frequently use lots of linguistic devices such as puns, metaphors, and rhyming. Some of them take a refined natural fluency in English in order to excel at, especially when competing against people like Ken and Brad. Watson wasn’t perfect, and certainly offered up some head-scratching answers, but showed off some stellar skills at natural language processing.
These competitions certainly do not mean that the AI revolution is here — the AIs are relatively narrow — but they do show the consistent improvement that our AI has made.
Turing Test and Chinese Room
The Turing Test is a quite typical test for artificial intelligence to indicate where it is in regards to human intelligence. Proposed by Alan Turing, the test requires that the artificial intelligence is able to have a conversation with a human without the human being able to tell that the AI is not human.
The Turing Test definitely has merit, and can indicate that an AI has reached a significant milestone, but is not without its faults. As the BBC article indicates, an AI passing the test can depend heavily on the person who is testing it. Someone asking simple trivia or basic questions will give a much easier Turing Test than someone else asking philosophical and existential questions.
Furthermore, as the Chinese Room illustrates, an artificial intelligence passing the Turing Test does not indicate that it can think like a human or is even as intelligent as a human. The person in the example certainly did not understand Chinese. If expected to hold a normal conversation in Chinese or read and comprehend a Chinese novel, there’s nothing to indicate they would have been able to do so successfully.
Potential Dangers of AI
Some of the most intelligent and accomplished people in our society, like Stephen Hawking and Elon Musk, believe that the inevitable AI revolution could spell doom for humanity.
While we are nowhere near such a disaster now in 2018, we should express some kind of general concern for the dangers AI could impose on our society. Theoretically, an AI that achieves superintelligence would be a major threat to humanity, especially if it were to reach a point in which its intelligence surpasses that of humanity collectively.
At such a point, humanity would have very little chance of stopping it, controlling it, or even influencing it. As a result, the AI would likely do one of two things. On one hand, it may see us as some kind of threat to either itself or the Earth — we are — and decide to drive us to extinction. Or, the AI may just see us as a relatively harmless subservient species (maybe as we see primates or even dogs or horses) and decide to use us in some way. Regardless, it would not be a great outcome for humanity.
So, we don’t necessarily have to be tinfoil hat paranoid about the day that AI destroys the human race, but we probably should think about taking some safeguards to that possibility now while we have the chance.
Could Computers Ever Be Human?
Will computers ever really mirror human thinking? Can they have the same emotions and morality that humans do?
These are some very philosophically tough questions, and unfortunately we don’t currently have enough information on how our own brains work to really have any answers to them.
However, embracing some of my more materialist beliefs, I want to believe that yes they can. If our mind is actually just made up of billions of connected neurons that transmit signals to one another, then there’s no reason that we can’t eventually fully replicate it with sufficiently sophisticated technology. Additionally, if we were able to fully replicate the human mind, then there’s no reason that computers shouldn’t be able to express everything we can too.
This being the case would give us so much insight into the human mind and all of its components. It would revolutionize psychology and philosophy by completely understanding how we think and why we think the things we do. Of course, we have to learn much more about how we think to ever reach this point in the first place, but it’s absolutely fascinating to ponder these future developments.
But, if we created artificial intelligence that perfectly mirrored our own methodology of thinking, we would probably have to treat it the same as one another. It would have the same propensity to feel and express emotions the way we do and be intellectually indistinguishable from us.
Perhaps, we wouldn’t even really be able to call it “artificial” intelligence at all.
