Kismet (robot)

• 1. Introduction

  ›

  Anger, extreme anger, disgust, excitement, fear, happiness, interest, sadness, surprise, tiredness, and sleep: Kismet could perform all of them with a single face. Dr. Cynthia Breazeal created it at MIT in the 1990s. The project was an experiment in affective computing, a field dedicated to machines that can recognize and simulate emotion. The name itself is a clue. Kismet comes from a Turkish word meaning 'fate' or sometimes 'luck.' Physical materials alone cost an estimated US$25,000. Breazeal described her relationship with Kismet in striking terms. She called it 'something like an infant-caretaker interaction, where I'm the caretaker essentially, and the robot is like an infant.'

• 2. Building the Face

  ›

  Four Motorola 68332 processors sit at the core of Kismet's computing setup, alongside nine 400 MHz personal computers and one additional 500 MHz machine. That hardware drives a face built from moving parts chosen to carry emotional weight: ears, eyebrows, eyelids, lips, a jaw, and the head itself. Each element has its own range of motion. Facial expressions emerge from the interplay of all of them together. Kismet also receives information through three distinct input streams: auditory, visual, and proprioceptive signals. Proprioception is the sense of one's own body position. For Kismet, tracking its own head position becomes a practical necessity that shapes how the whole system runs. Steering all of this hardware is a piece of software Breazeal's team called the synthetic nervous system.

• 3. Eyes and Ears

  ›

  When a large, close, rapidly moving object enters Kismet's visual field, the system reads it as a threat. This response comes from stereo cameras that estimate how far away objects are, giving the robot a sense of depth. The vision system can also perform eye detection and skin-color detection, though the last of these was noted as controversial. Motion detection pauses momentarily whenever Kismet moves its own head. Without that pause, the robot would flag its own movement as danger in the scene.

  Kismet's audio system was built with a specific listener in mind: a parent talking to a child. It detects affect in infant-directed speech, sorting utterances into five categories: approval, prohibition, attention, comfort, and neutral. The classifier was trained on recorded speech using features like pitch and volume variance. These were fitted to a gaussian mixture model using expectation-maximization, a method that iterates until the model best fits the data. Classification runs in two stages: first sorting an utterance into one of two broad groups, then doing finer work within that group. That approach improved accuracy on the hardest cases. Approval, as in 'You're a clever robot,' and attention, as in 'Hey Kismet, over here,' were among the most difficult categories to separate reliably.

• 4. One State at a Time

  ›

  At any given moment, Kismet can only occupy one emotional state. Breazeal was direct about what that means: Kismet is not conscious, and it does not have feelings. What the robot produces are performances of emotional states, not experiences of them. In Breazeal's own demonstrations, the facial expressions are described as communicating the robot's 'motivational state,' a term that stops short of claiming inner experience. The human-robot relationship, as Breazeal framed it, was explicitly educational. She described providing the scaffolding for Kismet's development. That framing implies Kismet's responses were not fixed at the factory but shaped through the act of interaction itself.

• 5. Speaking in Phonemes

  ›

  Kismet does not speak in words. Its vocal system generates a proto-language built from phoneme-like sounds, similar in character to infant babbling. DECtalk, the voice synthesizer it uses, shifts pitch, timing, and articulation to match the emotional state Kismet is performing. Intonation also carries grammatical meaning, allowing the robot to distinguish questions from statements in its utterances.

  'Simplicity is the secret to successful lip animation' was the principle Kismet's designers borrowed from animation. Lip synchronization mattered to them for realism, but they chose not to replicate the exact mechanics of human lips. Instead, they aimed to 'create a visual shorthand that passes unchallenged by the viewer.' That the shorthand passes unchallenged is as much a finding about human perception as it is about robot design.