Serendip is an independent site partnering with faculty at multiple colleges and universities around the world. Happy exploring!

The New Meaning of AI

Sarah Schnellbacher's picture

If you answered, "What/who is WATSON?" you are correct.
Though WATSON might have beaten you to this conclusion, the mechanism for his "thinking" is very different from that of humans. While we rely on a gut feeling from reinforced neural connections to tell us the validity of our answer, WATSON does a comprehensive search connecting the dots to find the "needle in the haystack answer" (Jeopardy) from all the information available in published works saved on his hard drives. Our nervous systems, unlike those of computers, allow us to focus only on relevant input to the brain rather than every sensory input. A computer search for a keyword or phrase on Google may produce too narrow or too broad a selection of irrelevant information. WATSON differs from the ordinary PC in that he is programmed to use probability algorithms to rule out irrelevant information and is programmed to associate words with similar words as we do in natural language. In emulating human neural connections through word association, WATSON's makers at IBM have created the closest successful T3 Turing Machine yet with regard to the human mind. 
Today we think of the age of computers for the layperson dawning in the eighties with the iconic super bowl release of the 1984 Macintosh PC; however, the foreground for computers like WATSON lies in Turing's 1950 paper on his "Turing Machine" and "Turing Test". A Turing Machine (TM) is a form of artificial intelligence that has the ability to pass a Turing Test (TT). Turing believes that a proper way to test whether a machine is really artificially intelligent is to have the machine attempt to trick a human into believing that it is a human male playing a female. If a machine can conquer this feat of fluid ambiguous gender, then it is really artificially intelligent (See link to Rapaport below). 
Turing Machines are gradated (T1-T5) to account for technological advances and the fact that some Turing machines come closer to the definition of a true Turing Machine than others. The T1 level of artificial intelligence (AI) is a machine's ability to trick a human into believing that the machine is human on a single instance. T1 machines such as "SmarterChild" on AOL Instant Messenger can provide what seem like insightful responses at the onset; however, their lack of ability to learn and follow a conversation make them easily detectable after a few minutes of conversing. The T2 Turing Machine is a pen pal. It can maintain a conversation for a lifetime through text but lacks functional "organs" to interpret pictures and to speak. T3 machines are functionally the same as humans and display real intelligence; however, their organs are physiologically and anatomically different from humans. T4 machines are virtually human in all regards but they are composed of synthetic materials even though their organs can be swapped with those of humans. T5 machines are indistinguishable in every sense and are composed of biological molecules differing only in how they were created. Most of our robots today are only T1 TMs. WATSON is not a full T3 machine, as he cannot solve word puzzles that involve complex puns and therefore his "mind" is not equivalent to that of a human with respect to natural language. He lacks the ability to hear and see so we could easily detect that he is a robot and thus he is only T3 with regard to his mind.
Turing’s description of artificial intelligence seems to differ in its goal from what we value today in robotics. The focus of early AI as presented by the Turing Test aims at creating an artificial human indistinguishable from a real one. Today I believe this goal has changed. We no longer fear a hostile takeover by machines that will begin to think and decide to revolt against humanity. WATSON could easily possess a simulated human face as his avatar on Jeopardy; however, what good is an android really to us? The developers of WATSON described their goals for the system in medicine and business during the three-day Jeopardy challenge. None of these goals involved competing with human capabilities but rather using the advantages of computer systems in instantly sorting through massive loads of information that is irrelevant to us.  Daniel Dennett uses the anecdote of a robot attempting to rescue it's power supply from a room with a bomb in Minds, Machines & Evolution to show how the human ability to function without extraneous sensory information from the autonomic nervous system allows us to do what machines can't: we can tune into only what is important at the moment. The battery the robot seeks is on a wagon with the bomb. The Robot pulls the wagon out of the room but does not realize that the bomb also came with the wagon in a system where a robot's programming is minimal. In a system where the robot observes all of its surroundings, the overload of information would prevent the robot from every finding its battery. Though we can distinguish what is relevant from the irrelevant, we often miss important information in the skimming process with texts and can waste hours searching through documents for meaning where it is lacking. The goals in robotics today are not to create artificial humans but rather tools to make our lives easier and connect us with one another.
In his article “How to Pass a Turing Test” William Rapaport describes how not only has our idea of a computer changed but the term itself means something entirely different today from when it was first used. A computer was originally a person who computed. What we think of as a computer today was originally called a computing machine. Just as our notion of what it means to be a computer has changed, so too has our notion of flying. Rapaport points out that we do not question whether or not a plane flies; it does, but it doesn’t fly in the same manner as a bird flies. A plane is a T3 Turing Machine emulating the function of a bird even though it is constructed differently and contains none of the same materials as a bird. Thus we “think”, but can we deny that WATSON "thinks" too when he is able to reach the same conclusions as we do? WATSON may think, but we should not fear that he will become HAL from 2001 Space Odyssey; our concept of intelligence like flight is what is changing.



Anne Dalke's picture

Into the uncanny valley

several students in my other class, on gender and technology, have also been reflecting on the implications of Watson's win. You might find of particular interest an essay on the "gendering" of that computer, which twines together two very interesting and complementary arguments. The first has to do w/ the gendering of robots--the use of "gender markers" that enable humans to interact more comfortably with machines. But the second has to do w/ the intriguing reverse activity--the caution against creating robots that are "too-humanized," and so "unsettling" to humans (according to the theory that we feel revulsion when robots too closely resemble us, the "uncanny valley" is a negative dip in the graph reflecting our reaction).

Your paper doesn't explore that terrain, or any such "edges" that might be evoked in the inexorable (is it inexorable?) progress from T1 to T5. What it does explore--and what I'd like to understand more and better--is the evolution of our thinking both about what it means to be a computer and what it means to think: "our concept of intelligence" is changing, you say: now that's thinking evolutionarily!