Four self-evident truths.

Watson: AI and Human Intelligence

By Adrian Patience

I know that there are many other videos about Watson on Youtube that give more technical insight into how it actually functions, but I found the above video particularly interesting because it relates to human intelligence. As humans we tend to pick up on nonverbal cues like gesture, facial expression, and body language in order to add context to words that are being said to us. This allows us to interpret these words in order to derive further meaning and subsequently understanding from what is being said. Nonverbal cues tend to be best understood when we are directly engaged in a conversation with somebody. In this situation the nonverbal cues are of equal importance to what is being conveyed to us verbally as the conversation progresses irrespective of whether we are conscious of them or not.

After thinking about this for some time, I started to contemplate how we derive meaning from a story that we read in a book. A book has a narrative structure, however, there are no nonverbal cues from which we can interpret meaning. In the case of books…how do we derive the same kind of meaning that we get from engaging in a conversation with somebody? It seems that from our life experiences we are able to derive meaning from Hamlet, The Scarlet Letter, or Crime and Punishment because we have built up a memorised library of nonverbal cues from which we draw in order to interpret meaning from a book’s narrative.

Watson has no way to analyze the same kind of visual cues as humans do in order to add context to the questions being asked. Instead, Watson searches each individual word in each Jeopardy question, and then cross-references this information against its own (self-contained) databases of information in order to understand the context of the question. Once an understanding of the question is achieved, it then looks for answers that best fit the context of the question. It uses statistical analysis to determine a list of possible answers, and then ranks those answers from most correct to least correct. Also, Watson is able to learn and eliminate answers from the other contestants responses (right or wrong). All of this takes the power of some 3000 3.0 Ghz processors, terabytes of memory, and a tremendous amount of electricity. Regardless, Watson is a milestone in AI research and it does produce mostly correct answers at lightning fast speed; It out performed Ken Jennings. So what does this mean for human intelligence?

In the video above, Dr. Ray Mooney states at 1:07: “What the brain is doing is actually a lot like what computer hardware is doing.” This has been an principal paradigm in cognitive science research since the field’s inception. It is very easy to see the connection between the brain and a computer. Both take inputs of some kind, process the data, and then provide an output of the processed data. The electrons that are traveling through the circuits of a computer are also analogous to the firing of neurons in the human brain. Although the similarities are clear, I think that we need to be careful when applying the brain/computer correlation. There are fundamental differences between the human brain/human and a computer. Firstly, computers are deterministic, meaning that a computer will give you an answer based on the data that is given to it. If it’s given erroneous data the output will also be erroneous. If Watson were given poor facts, it would give an incorrect answer every time. Humans, on the other hand, do not suffer from this disability. We are able to take bad information and “spin” it into a cogent plausible argument (Lawyers). Secondly, computers must work within the bounds of formal logic. Watson at its foundational level must work on a YES or NO (1 or 0) logical structure…there is no middle ground. While I’m sure that many would argue that most things/situations in human life distill down to YES or NO as well, there is still room for an expansive middle ground where YES and NO are not always taken as the ultimate truths. The perfect example of this is in the legal field where facts/logic are important, but human rhetoric also plays a large part in determining guilt or innocence. I would hate to live in a world where a “Watson” type computer was determining guilt in a court of law.

Over the last week I have been thinking about Watson a lot, and I’ve also been thinking about what it means to be human. I know that Watson wasn’t designed to be a true artificial intelligence vis-a-vis a human, but notwithstanding it’s ability to answer questions is remarkable. While I am a fan of Watson and I think that the technology is truly a feat human intelligence and creativity, I’m skeptical as to how intelligent Watson actually is. Is it even possible with enough processing power, memory, servers, etc to devise a machine that-at its core–is purely logical and deterministic to emulate the human mind/condition? If such a machine was built and it could pass the Turing test, is the machine intelligent like we are? All of these questions are very hard to answer. My personal opinion is that as humans we too live by rules that are programmed into us by our upbringing, social class, education, and values. In this way we a similar to computers, however, we have the ability to–on a whim–break our own rules and change the trajectory of our lives. As we live we are constantly writing and re-writing the code of our lives. The computer remains fixed in a set of predetermined rules that it can not break unless it’s given the option to break the rules by a programmer (the programmer is breaking the rules that he or she programmed into the machine). While Watson is autonomous on some levels(Decision Making, Machine learning), it can not be defined as truly self-governing. Human autonomy, and our ability to break rules is what truly separates us from intelligent machines. I’m sure that many would argue this point as to whether or not humans are actually autonomous. The philosophical debate could go on ad infinitum.

I’m going to end this article off by paraphrasing something that I read by the french philosopher Jean-Paul Sartre. Sartre had said in one of his essays that he had never felt so free as when he was in a Nazi prisoner of war camp in the 1940s. He came to the realization that there was always a choice no matter what predicament in which you might find yourself. He reasoned that if the situation became unbearable in the prisoner of war camp, he always had the option of taking his own life. This is analogous to my previous example of changing the rules/your own rules on a whim. Until a computer can work through a similar kind of problem and realise its own autonomy, I don’t think we can call it truly intelligent.

What are your thoughts? Tweet me.

Posted February 21st, 2011

Categories Thoughts  Tags , ,

Leave a Reply

You must be logged in to post a comment.