Is Artificial Intelligence Truly Possible?
Posted by admin | Filed under Artificial Intelligence, Cognitive Science, Science, Technology
When we hear the term Artificial Intelligence, most people might think of a servant (R2D2 in Star Wars), a law enforcer (K.I.T.T. “Knight Rider”), a comrade (Lt. Commander Data in Star Trek; The Next Generation), an overlord (The Matrix) or an assassin (Terminator). These depictions of artificial intelligence are not based on real science, but the imagination of Hollywood screenwriters and science fiction authors. According to an article in the New York Times, the cultural definition of artificial intelligence is “the science of how to get machines to do the things they do in the movies”6. A more scientific definition may be better supplied based on a seminal paper written by Alan Turing asking the question; “Can machines think?”8. In Turing’s paper, he also provides a test to answer this question. Based on the results of this test thus far, artificial intelligence only mimics human thinking and cannot truly think autonomously.
To formulate this argument, I have organized this post into two main sections. In section one, I provide the current state of artificial intelligence, citing recent attempts at simulating human intelligence, most notably by the super computer called Watson from IBM who competed against humans on the game show Jeopardy!. I conclude this paper with the second section, discussing reasons why the current definition of artificial intelligence will never be possible, and what the real definition of artificial intelligence should be.
Current State of Artificial Intelligence
During the late 1980’s, a researcher from Carnegie Mellow University developed the first viable computer to play chess at a competitive level against human world class chess champions. It was however easily defeated in 1989 when it played two matches against World Chess Champion, Garry Kasparov. The project was moved to the IBM Research group and a few years later a new challenger named Deep Blue was pitted against Garry Kasparov. On May 11th, 1997, the machine was able to “outthink” its competitor with two wins and three draws. Kasparov accused IBM of cheating and demanded a rematch, but IBM refused and dismantled Deep Blue2.
Our interaction with artificial intelligence does not just include chess playing against super computers. Every day, millions of people use the Internet to find information a broad variety of topics. This information is typically accessed by typing a text query into a search engine which will find the most relevant results based on the text query entered by the user. In order to improve the results and ensure its relevancy, artificial intelligence is employed.3 One would assume that this is an easy task, however, while even the best artificial intelligence can beat a human chess player, it can have extreme difficulty identifying objects in photos or understanding sentences. (par. 1)
Enter the next generation of super computer to take on the Turing test and answer the question: “Can Machines think?” Between February 14th and 16th, the IBM’s DeepQA project team matched their latest artificial computer system named Watson, against previous champions of the television quiz show Jeopardy!8. Watson was able to defeat Brad Rutter, the biggest all-time money winner on Jeopardy!, and Ken Jennings, the record holder for the longest championship streak on Jeopardy!.
In the last twenty years, computers have become capable of defeating humans at specific tasks, defeating Grand Champions at chess, picking relevant results from billions of pages on the Internet and crushing the best of the best at trivia. This is, however, far from the necessary proof that a machine can truly think like a human. When one examines how all of three these tasks are handled, it can be determined that computers thus far are better at two specific functions.
The first function is the ability to store and retrieve enormous amounts of data in order to solve a specific problem. Humans also use memory to solve problems based on previous experiences and other cognitive functions, but we accumulate those memories through the process of learning. This is the first obstacle to true thinking for a system attempting to display intelligence: They must be able to learn independently.
The second function computers are very proficient at is calculation. Once a computer has searched and retrieved possible solutions, it computes the relevancy based on algorithms and available datasets to come up with a suitable answer. In the example of chess playing super computers like Deep Blue, future moves and counter moves can be searched to a depth of “between six and eight moves to a maximum of twenty or even more moves in some situations”1. This provides a disproportional advantage to the computer as it can calculate exponentially more potential outcomes, but it lacks a crucial element that is needed to go beyond mere chess moves: strategy. The lack of strategy is substituted in chess matches by using a brute force approach of predicting moves. Strategic thinking is the second obstacle challenging machines to truly display intelligence.
Why True Artificial Intelligence Is Not Possible
The common cultural definition of artificial intelligence, rather than the scientific one, is a portrayal of humanlike thinking and behavior, by human looking androids that can physically interact and display complex cognitive behaviors. Watson is far from an android that has the appearance and mannerisms of a human. The epitome of artificial intelligence right now is a room full of computers that are fed enormous amounts of data by scientists so that the computer may play a trivia game with limited scope.
During Final Jeopardy! in the category U.S. Cities, the question was asked: “Its largest airport was named for a World War II hero; it’s second largest, for a World War II battle”. Both Rutter and Jennings gave the correct response of Chicago, but Watson’s response was “What is Toronto?????”4.
In discussing decision making in another domain such as nursing, Jones and Beck states as follows: “Regardless the type of system, the accuracy of the outcome decision is only as good as the data” 5.The fact that a system is only as good as its data could be considered a universal truth that applies to artificial intelligence, nursing and many other domains and systems. Data is thereby a limiting factor that creates finite boundaries within a decision-making process has to be made, ultimately limiting the decision making process.
When IBM set out to create Watson, they had to consider the finite storage capacity of the hardware they used. This placed a limitation on the breadth and depth of information they could potentially store in a database for Watson to utilize when formulating answers. It also created a human imposed limit on the topics and facts that could be included in this finite data set. These confines ultimately placed a limitation on the range of answers that Watson could come up with in response to an unlimited number of questions. This demonstrates the absence of understanding, thinking and reasoning when the database is incomplete or the search algorithms fail in delivering a brute force answer to what should be an easy deductive reasoning problem, given the necessary data and search algorithms. Since the trend in computer hardware is to shrink in size and lower in cost, the availability of larger data sets and consequently better answers is only limited by economics of scale at this point.
It is my conclusion that current developments with computers, such as getting better at brute force and task specific functions aka. expert systems, will encourage further research and development to move away from generalized android-like artificial intelligence towards specific expert systems that excel at explicit tasks. The reasoning behind this is almost exclusively based on the financial aspect of continued funding of projects that have viable returns on the investments made. We see immediate business application for voice recognition artificial intelligence being used in call centers, or text recognition being used for web based search engines. There does not seem to be a viable business incentive for creating an android that looks like, act like and thinks like a human.
For that matter, I predict that funding will continue in areas of research where limited tasks can be accomplished with great efficiency and cost reduction, which is promising for specialized systems, but not general artificial intelligence. This will ultimately lead to a dead end for artificial intelligence and a prosperous future for specialized expert systems which will change our world in significant ways no one can truly imagine yet. As an example, no one can deny the great impact of GPS navigation which uses sophisticated artificial intelligence and large amounts of data to ensure accurate travel. Perhaps it is time to reconsider the science fiction inspired definition of artificial intelligence as a relic of that genre and accept the limited – but specialized – definition of what computers are capable off: being experts at specific tasks.
Resources cited:
- Campbell, Murray. “An Enjoyable Game”. In Stork, D. G.. HAL’s Legacy: 2001′s Computer as Dream and Reality. Cambridge, Mass.: MIT Press, 1998. Print
- Hsu, Feng-hsiung. Behind Deep Blue: Building the Computer that Defeated the World Chess Champion. Princeton University Press, 2002. Print.
- “Helping computers understand language”. The Official Google Blog. Google Inc., 19 Jan. 2010. Web. 23 Sep. 2011. http://googleblog.blogspot.com/2010/01/helping-computers-understand-language.html
- “IBM’s computer wins ‘Jeopardy!’ but… Toronto?” CTV News, Entertainment. CTV Television Network, 15 Feb. 2011. Web. 25 Sep. 2011. http://www.ctv.ca/CTVNews/Entertainment/20110215/watson-jeopardy-final-toronto-110215/
- Jones, Rebecca A. Patronis, Sharon E. Beck, eds. Decision Making In Nursing. Albany: Delmar Publishers., 1996. Print.
- “Smart Machines, and Why we Fear them.” New York Times (1923-Current file): A15. ProQuest Historical Newspapers: The New York Times (1851-2007). Mar 21 1998. Web. 28 Sep. 2011. http://search.proquest.com/docview/109953529?accountid=36334
- Turing, A.M. “Computing machinery and intelligence.” Mind, 59 (1950): 433-460. Web. 27 Sep. 2011. http://loebner.net/Prizef/TuringArticle.html
- “What is Watson”. IBM Innovation. IBM Inc., n.d. Web. 26 Sep. 2011. http://www-03.ibm.com/innovation/us/watson/what-is-watson/index.html
Intellectual Athletic ex-pat South African American Vegetarian Techie who loves his wife, parents and all creatures regardless of their current state of delusion.