The term artificial intelligence tends to evoke less-than-pleasant images: Terminator, Skynet, HAL, M-5 (if you’re a Star Trek fan), Cybermen, and Daleks (if you’re a Doctor Who fan). Stories of intelligent automatons can be traced back to antiquity: Galatea, the Automata of Al-Jazari, the Prague Golem, and others; AI has been, and will likely remain, a staple theme in science fiction and film. In reality, AI is a wide and diverse field with many subfields, ranging from the creation of machine intelligence that equals or surpasses human intelligence to problem solving, natural language processing, machine perception and learning, robotics, etc. There is a trend today for researchers in AI to call their work by other names, such as cognitive systems or knowledge-based systems or computational intelligence. This may serve to distinguish their own field from the rest of AI or to avoid being viewed as wild-eyed dreamers. At times, AI has made overly ambitious and optimistic claims, a failing that continues to haunt those in this field.

While there are attempts to replicate the human brain and body, most AI has less ambitious goals; it’s really about making computers or machines do things that human beings already do, things like recognizing speech, driving cars, translating languages, controlling the temperature in houses, or even removing the bone from a ham. AI is used in aviation, finance, hospitals and medicine, heavy industry, customer service, and on the Web.

Because AI is often used in subtle ways, it can be hard to recognize, even though many of us use some form of it every day. There is a tendency for AI to lose the label “AI” after it goes mainstream. Technological tools in your home, like your DVR, washing machine, and thermostat, aren’t AI or robots (one exception being the Roomba). They’re smarter, or easier to use, or just better. The thermostat is better at keeping your home comfortable. Websites are getting better at recommending things you like based on your search habits and can do helpful things like translate web pages to English.

This report deals with a few specific applications of AI—natural language processing and conversational user interfaces, commonly known as chatbots. To give you a better understanding of where these fit in AI in general, let’s look at a few basic ideas in the AI world.

Artificial intelligence is often classified as “strong” or “weak.” Strong AI is also called “artificial general intelligence.” Strong AI may be defined as machine intelligence that matches or exceeds human intelligence. A computer with strong AI could perform any intellectual task that a human being could. It’s associated with human traits such as self-awareness and consciousness. This is the type of AI that’s common in science fiction and is often the topic of discussion for futurists like Ray Kurzweil.

Weak AI is also known as “applied” or “narrow” AI. Weak AI is the use of software to study or accomplish specific tasks, like sweeping floors, driving cars, or understanding natural language. Chatbots and NLP fall within this division. In fact, library chatbots can be considered a narrow application of an already narrow application, as their purpose is to assist with the use of library resources and not to chat at length about any possible topic.

A detailed history of artificial intelligence is beyond the scope of this report. Readers who are interested in this topic may wish to read Artificial Intelligence: The Basics by Kevin Warwick.⁸ AI as we know it really started with the birth of computers in the 1940s and 1950s and with the work of Alan Turing.

Alan Turing is widely regarded as the father of computer science and was a pioneer in artificial intelligence. In his 1948 report “Intelligent Machinery,” Turing explored the possibility of creating machines that could show intelligent behavior and proposed a forerunner of what would become the “Turing test.”⁹ A few years later, in 1950, Turing published his seminal paper “Computing Machinery and Intelligence.” Turing begins, “I propose to consider the question, ‘Can machines think?’”¹⁰ The traditional approach to such a question would be to first define the terms machine and intelligence. Turing bypassed such definitions, noting that intelligence is difficult to define (and continues to be difficult to define today). Taking a new approach, Turing proposed a test inspired by a party game known as the Imitation Game. In the Imitation Game, a man and a woman go into separate rooms. Party guests then try to tell them apart by submitting a series of questions to each and reading the typewritten responses. As the guests try to tell who is who, the man and woman both try to convince the guests that they are each other. In Turing’s test, a human interrogator asks a series of questions to a human and a computer, both separated from the interrogator, and tries to distinguish the human and the machine based on their responses. If the interrogator fails to tell the machine from the human, the machine has passed the test. Although AI has made great strides since Turing’s day, no computer has yet passed the Turing test.

An interesting side note to the Turing test is the CAPTCHA. (The term was coined in 2000 by researchers at Carnegie Mellon University. It is an acronym based on the word capture and standing for “completely automated public Turing test to tell computers and humans apart.”) CAPTCHAs are a reverse Turing test: a simple visual test that a sighted human can complete without much difficulty but that an automated program cannot understand.¹¹ CAPTCHAs are intended to keep bots and other automated programs from signing up for e-mail addresses, violating privacy, cracking passwords, or sending spam e-mail.

Another interesting side note to the Turing test is the annual Loebner Prize competition. Started in 1990 by Hugh Loebner, an inventor and businessman, this has become one of the biggest challenges in the world of chatbots.¹² Each year the judges at the competition conduct conversations with four finalist chatbots and some human surrogates. Conversation partners are ranked from the most to least human-like. The chatbot with the highest overall ranking wins. Chatbots using the two programming languages discussed in this report, AIML and ChatScript, have performed well in the Loebner Prize competition. The competition has been subject to criticism, remains controversial within the AI community, and has been avoided by serious researchers and commercial AI developers.

At the time of publication a number of companies are vying to create virtual agents for public sector companies. In addition to these company-specific resources, there is a new wave of “personal assistant” agents—like Ultra Hal (2012) for Windows computers, Siri (2012) for the iPhone, and Evi (2012) for Android devices—that will perform metasearch functions.²¹ With the success of these programs, virtual agents are almost certain to become more prevalent in the future.

Virtual agents are starting to gain popularity in libraries. From the German bots created in the 1990s to Emma, and Pixel, bots are enhancing services within physical libraries and providing assistance to offsite users. What is more, they are providing this assistance in an extremely cost-effective manner. Emma, for example, answered a total of 4,774 library-related questions during 2011. The calculated cost of providing this service was $0.14 per use. As library funding continues to erode and chatbots become more intelligent, automated reference services will become an increasingly attractive option, if not a necessity.²²

With the availability of simple and inexpensive options for virtual agent creation, it’s easy and cost-effective for libraries to explore this opportunity to expand their information services.