Artificial Intelligence

views updated

ARTIFICIAL INTELLIGENCE

Artificial Intelligence (AI) is the science and technology that seeks to create intelligent computational systems. Researchers in AI use advanced techniques in computer science, logic, and mathematics to build computers and robots that can mimic or duplicate the intelligent behavior found in humans and other thinking things. The desire to construct thinking artifacts is very old and is reflected in myths and legends as well as in the creation of lifelike art and clockwork automatons during the Renaissance. But it was not until the invention of programmable computers in the mid-twentieth century that serious work in this field could begin.

AI Research Programs

The computer scientist John McCarthy organized a conference at Dartmouth College in 1956 where the field of AI was first defined as a research program. Since that time a large number of successful AI programs and robots have been built. Robots routinely explore the depths of the ocean and distant planets, and the AI program built by International Business Machines (IBM) called Deep Blue was able to defeat the grand master chess champion Garry Kasparov after a series of highly publicized matches. As impressive as these accomplishments are, critics still maintain that AI has yet to achieve the goal of creating a program or robot that can truly operate on its own (autonomously) for any significant length of time.

AI programs and autonomous robots are not yet advanced enough to survive on their own, or interact with the world in the same way that a natural creature might. So far AI programs have not been able to succeed in solving problems outside of narrowly defined domains. For instance, Deep Blue can play chess with the greatest players on the planet but it cannot do anything else. The dream of AI is to create programs that not only play world-class chess but also hold conversations with people, interact with the outside world, plan and coordinate goals and projects, have independent personalities, and perhaps exhibit some form of consciousness.

Critics claim that AI will not achieve these latter goals. One major criticism is that traditional AI focused too much on intelligence as a process that can be completely replicated in software, and ignored the role played by the lived body that all natural intelligent beings possess (Dreyfus 1994). Alternative fields such as Embodied Cognition and Dynamic Systems Theory have been formed as a reply to this criticism (Winograd and Flores 1987). Yet researchers in traditional AI maintain that the only thing needed for traditional AI to succeed is simply more time and increased computing power.

While AI researchers have not yet created machines with human intelligence, there are many lesser AI applications in daily use in industry, the military, and even in home electronics. In this entry, the use of AI to replicate human intelligence in a machine will be called strong AI, and any other use of AI will be referred to as weak AI.

Ethical Issues of Strong AI

AI has and will continue to pose a number of ethical issues that researchers in the field and society at large must confront. The word computer predates computer technology and originally referred to a person employed to do routine mathematical calculations. People no longer do these jobs because computing technology is so much better at routine calculations both in speed and accuracy (Moravec 1999). Over time this trend continued and automation by robotic and AI technologies has caused more and more jobs to disappear. One might argue, however, that many other important jobs have been created by AI technology, and that those jobs lost were not fulfilling to the workers who had them.

This is true enough, but assuming strong AI is possible, not only would manufacturing and assembly line jobs become fully automated, but upper management and strategic planning positions may be computerized as well. Just as the greatest human chess masters cannot compete with AI, so too might it become impossible for human CEOs to compete with their AI counterparts. If AI becomes sufficiently advanced, it might then radically alter the kinds of jobs available, with the potential to permanently remove a large segment of the population from the job market. In a fully automated world people would have to make decisions about the elimination of entire categories of human work and find ways of supporting the people who were employed in those industries.

Other ethical implications of AI technology also exist. From the beginning AI raised questions about what it means to be human. In 1950 the mathematician and cryptographer Alan Turing (1912–1954) proposed a test to determine whether an intelligent machine had indeed been created. If a person can have a normal conversation with a machine, without the person being able to identify the interlocutor as a machine, according to the Turing test the machine is intelligent (Boden 1990). In the early twenty-first century people regularly communicate with machines over the phone, and Turing tests are regularly held with successful results—as long as the topic of discussion is limited. In the past special status as expert thinkers has been proposed as the quality that distinguishes humans from other creatures, but with robust AI that would no longer be the case. One positive effect might be that this technology could help to better explain the place of humans in nature and what it means for something to be considered a person (Foerst 1999).

The ethical responsibility that people have toward any strong AI application is a matter that must be taken into consideration. It does not seem moral to create thinking minds and then force them to do work humans do not want to do themselves.

Finally because AI technology deals directly with human operators, people must make decisions regarding what kind of ethics and morality are going to be programmed into these thinking machines. The scientist and fiction writer Isaac Asimov proposed in his writings three moral imperatives that should be programmed into robots and other AI creations:

A robot may not injure a human being or, through inaction, allow a human being to come to harm.
A robot must obey the orders given it by human beings except where such orders conflict with the first law.
A robot must protect its own existence as long as such protection does not conflict with the first or second law.

These imperatives make for good reading but are sadly lacking as a solution to the problems presented by fully autonomous robotic technologies. Asimov wrote many stories and novels (The Robot Series [1940–1976] and I,Robot [1950]) that used the unforeseen loopholes in the logic of these laws, which occasionally allowed for fatal encounters between humans and robots. For instance, what should a robot do if, in order to protect a large number of people, it must harm one human who is threatening others? It can also be argued that AI technologies have already begun to harm people in various ways and that these laws are hopelessly naïve (Idier 2000). Other researchers in the field nevertheless argue that Asimov's laws are actually relevant and at least suggest a direction to explore while designing a computational morality (Thompson 1999).

This problem is more pressing than it may seem, because many industrial countries are working to create autonomous fighting vehicles to augment the capabilities of their armed forces. Such machines will have to be programmed so that they make appropriate life and death choices. More subtle and nuanced solutions are needed, and this topic remains wide-open—widely discussed in fiction but not adequately addressed by AI and robotics researchers.

Ethical Issues of Weak AI

Even if robust AI is not possible, or the technology turns out to be a long way off, there remain a number of vexing ethical problems to be confronted by researchers and technologists in the weak AI field. Instead of trying to create a machine that mimics or replicates exactly human-like intelligence, scientists may instead try to imbed smaller, subtler levels of intelligence and automation into all day-to-day technologies. In 1991 Mark Weiser (1952–1999) coined the term ubiquitous computing to refer to this form of AI, but it is also sometimes called the digital revolution (Gershenfeld 1999).

Ubiquitous computing and the digital revolution involve adding computational power to everyday objects that, when working together with other semismart objects, help to automate human surroundings and hopefully make life easier (Gershenfeld 1999). For instance, scientists could imbed very small computers into the packaging of food items that would network with a computer in the house and, through the Internet perhaps, remind people that they need to restock the refrigerator even when they are away from home. The system could be further automated so that it might even order the items so that one was never without them. In this way the world would be literally at the service of human beings, and the everyday items with which they interact would react intelligently to assist in their endeavors. Some form of this more modest style of AI is very likely to come into existence. Technologies are already moving in these directions through the merger of such things as mobile phones and personal data assistants.

Again this trend is not without ethical implications. In order for everyday technologies to operate in an intelligent manner they must take note of the behaviors, wants, and desires of their owners. This means they will collect a large amount of data about each individual who interacts with them. This data might include sensitive or embarrassing information about the user that could become known to anyone with the skill to access such information. Additionally these smart technologies will help increase the trend in direct marketing that is already taking over much of the bandwidth of the Internet. Aggressive advertisement software, spying software, and computer viruses would almost certainly find their way to this new network. These issues must be thoroughly considered and public policy enacted before such technology becomes widespread.

In addition, Weiser (1999) argues that in the design of ubiquitous computing people should work with a sense of humility and reverence to make sure these devices enhance the humanness of the world, advancing fundamental values and even spirituality, rather then just focusing on efficiency. Simply put, people should make their machines more human rather then letting the technology transform human beings into something more machine-like.

A last ethical consideration is the possibility that AI may strengthen some forms of gender bias. Women in general, and women's ways of knowing in particular, have not played a large role in the development of AI technology, and it has been argued that AI is the fruit of a number of social and philosophical movements that have not been friendly to the interests of women (Adam 1998). Women are not equally represented as researchers in the field of AI, and finding a way to reverse this trend is a pressing concern. The claim that AI advances the interests of males over those of females is a more radical, yet intriguing claim that deserves further study.

AI continues to grow in importance. Even though researchers have not yet been able to create a mechanical intelligence rivaling or exceeding that of human beings, AI has provided an impressive array of technologies in the fields of robotics and automation. Computers are becoming more powerful in both the speed and number of operations they can achieve in any given amount of time. If humans can solve the problem of how to program machines and other devices to display advanced levels of intelligence, as well as address the many ethical issues raised by this technology, then AI may yet expand in astonishing new directions.

JOHN P. SULLINS III

BIBLIOGRAPHY

Adam, Alison. (1998). Artificial Knowing: Gender and the Thinking Machine. New York: Routledge. Adam is an AI researcher who writes from experience about the difficulties encountered by women working in AI. She also argues that if women were more involved in AI it would significantly alter and improve the results of this science. Accessible, informative, and includes a select bibliography.

Asimov, Isaac. (2004). I, Robot. New York: Bantam Books. Classic science fiction account of the social impact of robotics. The stories are very entertaining while at the same time exploring deeply philosophical and ethical questions.

Boden, Margret A., ed. (1990). The Philosophy of Artificial Intelligence. New York: Oxford University Press. A collection of the best papers written on the philosophical debate regarding the possibility of creating AI. The research included stops at the early 1990s but this is a good place to start one's study of the subject. Includes a select bibliography.

Dreyfus, Hubert L. (1994). What Computers Still Can't Do. Cambridge MA: MIT Press. This is a famous counter argument to the very possibility of achieving any results in strong AI. Largely dismissed by the AI community, it is still useful as an example of how one might argue that AI misses the important insights into human intelligence and consciousness that phenomenological psychology and Heideggerian philosophy might add.

Dreyfus, Hubert L., and Stuart E Dreyfus. (1986). Mind over Machine: The Power of Human Intuition and Expertise in the Era of the Computer. New York: Free Press. This book was written with the author's brother who worked with the AI technology questioned in the book. They claim that human intelligence is not completely captured by AI technology and give numerous examples to illustrate their claims. Many of their claims, such as the inability of AI to program a computer to beat a chess master, seem to have been incorrect but this book is an important first step in Dreyfus's critique of AI.

Foerst, Anne. (1999). "Artificial Sociability: from Embodied AI toward New Understandings of Personhood." Technology in Society 21(4): 373–386. Good discussion of the impact AI and robotics has on complicating the notion of personhood, or what are the conditions under which something is understood to be a person.

Gershenfeld, Neil. (1999). When Things Start to Think. New York: Henry Holt. Good general overview of the attempt to add intelligence to everyday technology. Gershenfeld is involved in this research and speaks with authority on the subject without being overly technical.

Idier, Dominic. (2000). "Science Fiction and Technology Scenarios: Comparing Asimov's Robots and Gibson's Cyberspace." Technology in Society 22: 255–272. A more up-to-date analysis of Asimov's prognostications about robotics in light of the more dystopian ideas of the science fiction author William Gibson.

Luger, George F., ed. (1995). Computation and Intelligence. Cambridge, MA: AAAI Press/MIT Press. A collection of many of the best early papers written on AI by the pioneers of the field. These papers have shaped the field of AI and continue to be important. Includes an extensive bibliography.

Moravec, Hans. (1999). Robot: Mere Machine to Transcendent Mind. New York: Oxford University Press. One of the most successful AI researchers unabashedly predicts the future of humanity and the emergence of its robotic "mind children." Deftly presents very technical arguments for the general reader.

Nilsson, Nils J. (1986). Principles of Artificial Intelligence. Los Altos, CA: Morgan Kaufmann. Good introduction to the more technical issues of AI as a science.

Searle, John R. (1992). The Rediscovery of the Mind. Cambridge, MA: MIT Press. The American philosopher John Searle argues for the inability of AI to capture semantics (the meaning of words), with machines that can only manipulate syntax (the grammar and symbols of a language). If there is no mechanical way to create semantic understanding then there is no way to create thinking machines. This argument has not gone unchallenged by AI researchers and many of the entries in this bibliography attempt to refute Searle.

Thompson, Henry S. (1999). "Computational Systems, Responsibility, and Moral Sensibility." Technology in Society 21(4): 409–415. A discussion of how humans should be more sensitive to the moral implications of the computational systems they build.

Weiser, Mark. (1993). "Hot Topics: Ubiquitous Computing." IEEE Computer 26(10): 71–72. Also available from http://www.ubiq.com/hypertext/weiser/UbiCompHotTopics.html. Weiser introduces the concept of ubiquitous computing. This remains a very important defining moment for the field.

Weiser, Mark. (1999). "The Spirit of the Engineering Quest." Technology in Society 21: 355–361. Weiser argues that those who engineer devices that mediate the human-to-world interface should be especially careful that their work enhances human life rather than detracts from it.

Winograd, Terry, and Fernando Flores. (1987). Understanding Computers and Cognition: A New Foundation for Design. Boston: Addison-Wesley Publishing. Winograd and Flores argue for a form of AI that takes into account the important counterarguments to traditional AI, and instead of ignoring them, incorporates them into a better science that takes the embodiment of cognitive agents seriously.

Encyclopedia of Science, Technology, and Ethics