Daisy, Daisy, Give Me Your Answer Do
March 13, 2008
Even the most successful attempts at artificial intelligence (AI) always seem to lack certain essential qualities of a living brain. It is a formidable task to create a robotic or computerized simulation of a human that seems to display original desires or beliefs, or one that truly understands the desires and beliefs of others in the way people can. This latter ability, often referred to as “theory of mind”, is considered an integral aspect of being human, and the extent to which it has developed in us may be one thing that sets us apart from other animals. Reproducing theory of mind in AI is difficult, but a semblance of it has been demonstrated before with physical robots (click here for an example). Until now, however, it has never been recreated in computer generated characters.
A group of researchers at Rensselaer Polytechnic Institute (RPI) have developed a character in the popular computer game Second Life who uses reasoning to determine what another character in the game is thinking. The character was created with a logic-based programming RPI calls RASCALS (Rensselaer Advanced Synthetic Character Architecture for “Living” Systems). The program involves several levels of cognition, simple systems for low and mid-level cognition (like perception and movement), and advanced logical systems for abstract thought. The group believes they can eventually use RASCALS to create characters in Second Life that possess all the qualities of a real person, such as the capacity to lie, believe, remember, or be manipulative.
Second Life is a life-simulating game, similar in some ways to the popular game The Sims. Unlike the The Sims, however, Second Life involves a virtual universe (metaverse) where people can interact with one another in real-time through avatars they create for use in the game.
The character created by the group at RPI, Edd, appears to have reasoning abilities equivalent to those of about a four-year old child. To test these abilities, Edd was placed in a situation with two other characters (we’ll call them John and Mike). Mike places a gun in briefcase A in full sight of John and Edd. He then asks John to leave the room. Once he is gone, Mike moves the gun to case B, then calls John back. Mike asks Edd which case John will look in for the gun.
Does this sound familiar? It's an actual psychological test developed in the 1980s, originally known as the Sally-Anne test. The Sally-Anne test plays out the same scenario described above, only with dolls and a marble or ball (since its inception the test has been done with human actors as well). A child watches the Anne doll take a marble from Sally’s basket and put it in her box while Sally is not in the room. If the child, after watching the interaction, can guess when Sally returns that she will look in her basket for the marble, it demonstrates he or she has begun to form theory of mind. The child is able to understand that other people have thoughts and beliefs different from his or her own. They realize that when Sally re-enters the room she is unaware the marble has changed positions, so she will look in the spot where the marble originally was. The ability to make these types of attributions of belief usually develops at around age three to four in children.
Edd, the character from Second Life, is able to do the same. When Mike asks him in which case John will look for the gun, he will say case A—the case John saw the gun placed in (for the demonstration click here). And Edd is not programmed specifically to make this choice. Instead he “learns” from past mistakes that, if John cannot see the gun being moved he will not know it is in the other briefcase.
The research group at RPI see Edd as a first step in the creation of avatars on Second Life that can interact with humans in a manner unlike that of any simulated characters before, being able to understand and predict the actions of others, and act virtually autonomously. They see potential benefits of this technology in education and defense, as well as entertainment. IBM, a supporter of the research, has visions of creating holographic characters for games like Second Life, which could interact with humans directly.
This is all pretty amazing stuff, but for some reason HAL singing “Daisy Bell” keeps eerily replaying in my head as I write it.