Scitech | Are search engines smart?

Founder of Harvard Robotics Laboratory lectures on different types of intelligence, and how to test for them.

What is intelligence and who has it? These have been essential questions addressed by the fields of philosophy, psychology, and more recently artificial intelligence. Roger Brockett, founder of the Harvard Robotics Laboratory, had a succinct and definite answer. “Intelligence is the ability to partner with the environment.”

Brockett was invited to speak at the annual Beatty Memorial Lecture at McGill on October 29.  The objective of the lecture series – covering topics in the sciences, humanities, and social sciences – is to arouse curiosity and dialogue around a contentious topic.

During the lecture, Brockett elaborated various theories of intelligence. In 1950, Alan Turing, a logician and computer scientist, proposed a test to differentiate those with and without intelligence. The so-called “Turing test” involves a human communicating via computer, with either another human or a computer. If the tester communicates with a computer, and cannot detect it as such, then the computer is intelligent. This test offers a very narrow understanding of intelligence, according to Brockett. It emphasizes language, and it assumes that the ability to communicate with others is essential to intelligence. This definition precludes animals – many of which are capable of complex tasks – from being considered intelligent.

According to Harvard psychologist Howard Gardner’s theory of multiple intelligence, Brockett explained, humans have at least eight unique ways to learn and manipulate information. Domains of intelligence range from classically valued disciplines such as linguistics and mathematics to areas like bodily-kinesthetic and interpersonal skills. Brockett thinks that the Turing test might be more useful if it is expanded to assess these domains of intelligence as well.

One way Brockett discussed expanding the Turing test involves robotic handshakes – a proper handshake requires not only fine motor skills, but also the ability to adapt to the environment. If the robotic hand’s partner grips harder, then a realistic robot’s response must reciprocate with an equally strong grip: handshaking requires sensory input, feedback, and communication between different parts of the machine.

A team of researchers led by Amri Karniel, a professor at Ben Gurion University in Beersheba, Israel, is holding a competition in January 2011, in which robotic arms will compete in a Turing-like test: robotic arms must fool human testers into not being able to differentiate between when the robotic arm is controlled by humans in another room and when the robotic arm acts on its own. Brockett challenged the audience to contemplate whether or not passing this new iteration of the Turing test would constitute robotic intelligence.

According to Brockett, the reason humans still contemplate whether or not machines are intelligent is because people have a tendency to devalue the abilities of robots in order to bolster the unique status of the human mind. Brockett even believes that Google’s search engine can be considered intelligent: it explores and brings together many complex aspects of the environment. By this standard, intelligent machines are ubiquitous.