VICTOR Socio-Emotional AI
2ai Labs — Tim Barber, Mark Changizi
Requirements for socio-emotional AI
(1) Negotiation: Able to engage in (quite possibly non-linguistic) discussion in search of a compromise.
(2) Etiquette: Understands how and when to be polite, impolite, aggressive, trash talk, etc.
(3) Emotion-Recognition: Comprehends emotions.
(4) Status-Awareness: Sensitive to its own and others’ ranks within the social community.
(5) Stakes-Grasping: Understands the manner in which signaling emotions commits one to bets of reputation (or “cool”). (E.g., a trash talker who loses loses more cool than a polite opponent who loses.)
(6) Other-Awareness: Recognizes social boundaries.
(7) Salutations: Addresses individuals in a manner appropriate to their rank.
(8) Socio-Emotional Engine: Possesses underlying machinery allowing the full range of emotional reasoning.
(9) Personality-Coherent: The socio-emotional engine should allow a tremendous variety of personality types, each characterized by the manner in which it engages in discussions via emotional signals, and modifiable in a socio-emotionally coherent fashion.
(10) Emotion-Production: Able to go from outputs of its socio-emotional engine to emotional signals.
True intelligence is not about solving math and puzzles. What people really mean by intelligence is… social intelligence. Dope, fool, idiot, imbecile, moron, simpleton, etc. These imply social incompetency. An “idiot-savante” has genius level skills in some area, but lacks social competency. Intelligence is our capacity to be polite, to negotiate, to strategize, to raise the stakes, to push back, to compromise.
At 2AI Labs, the VICTOR project will bring “other-awareness” to AI. With true social intelligence, humans are able to cooperate, compete, and even lead. We all have to know our boundaries; how far we can “push” each individual in each kind of interaction, and we have to determine this quickly on the fly. Learning these boundaries is what gives us an awareness of others. In the future, AI’s fate – like ours – will be to navigate an ever-buffeting sea of crafty beings, and will accordingly need the other-awareness utterly missing in AI today.
Key to human social intelligence is innate knowledge of an evolved, sophisticated, ancient, universal “language” for communicating with others. That “language” is… the emotions. As with any language, emotions are expressed (“spoken”) to others, and are read (“comprehended”) by others. And just as we reason about the sentences someone may utter, and formulate a linguistic response, humans have innate machinery for socio-emotional reasoning, i.e., for determining how to emotionally respond given another person’s emotion signal.
We’re socially intelligent to the extent that we can partake in this language for social communication. It’s therefore not enough to have shallow face-recognition attribution and a veneer of human facial expression. AI must fully grasp the language of emotion: understand what each emotion means, reason about what reply to make, and know which emotion signals that reply. In fact, any AI not doing all this is, by definition, a sociopath.
We at 2ai Labs have spent the last decade deciphering the hidden language of emotions, and have developed a unifying theory of emotions, of their meaning and the machinery underlying emotional reasoning. The theory emanates from a first-principles approach to the fundamental issues pre-linguistic animals face in communicating so as to settle disagreements. It has two key pillars.
I. Comprehension of emotions: The technology understands the specific meanings of emotional signals, including (a) what I want, (b) how compromising I feel I am being, (c) my hand strength in a situation, (d) my opinion about my opponent’s hand strength, and (e) my confirmation receipt of my opponent’s signals.
II. Emotional reasoning machinery: The technology consists of an “emotion engine,” and allows the construction of AI personalities to suit one’s needs. The machinery allows personalities to vary along socio-emotionally sensible dimensions, and so one can, for example, vary their level of aggression, willingness to compromise, and degree of hospitality. Social intelligence also requires understanding social currency, or cool, and that a person’s emotions, and an AI’s emotional response, are bets; this is part of VICTOR’s machinery.
VICTOR tech in fact satisfies the ten criteria listed earlier for what AI needs in order to be socio-emotionally intelligent.
2aiTM Labs was founded by Dr. Tim Barber and Dr. Mark Changizi to push the research frontier so as to get “to AI”. http://2ai.org
Tim Barber graduated from Princeton with a Ph.D. in mathematics, and after working for the NSA and IDA started his own internet companies, Clickbank and Kount, the latter which uses his own AI developed to outwit cunning credit card thieves and which recently had a majority share sold for $80 million. http://www.timbarber.org/ , tim@2ai.org
Mark Changizi graduated from the University of Maryland with a Ph.D. in applied mathematics and theoretical computer science, held several university posts including Sloan Swartz Fellow in Theoretical Neurobiology at Caltech and Assistant Professor in Cognitive Science at RPI, and has become a well-known cognitive scientist with discoveries on topics such as the structure of the brain, visual perception, the origins of language, and principles underlying complex behavior, ideas covered in his books such as Vision Revolution, and Harnessed. http://www.changizi.com , mark@2ai.org