Action selection

Motor control

• Location-based perception (can't be everywhere at once)

• Limited field of view/hearing (can's sense everything everywhere)

...Action

• Localized operations

• Limited reach at that location

Embodiment of an agent puts constraints on other agents:

Need to get agent's attention to communicate

Whereas it would be great to have no constraints on mobility and action, embodiment is an important ingredient in communication

Embodiment is limiting to the agent who is embodied, but it liberates other agents, because allows them to make predictions about...

• ...the agent's mental state

• ...the state of the dialog

• ...the mental capacity of the agent

• ...the perceptual capabilities of the agent

Examples:

• Why did he do that? Perhaps he didn't see me? [visual attention, reasoning]

• "Look here, I'll show you" [visual attention, real time monitoring]

• People tend to look where they're listening

• (At a cocktail party) "John, over here!" [auditory perception]

Embodiment works as a framework for communication and action

For task-oriented collaboration, embodiment allows coordination of the limited resources:

• The agents' attention

• Their communication

• The actions the task calls for

People practice face-to-face communication almost every day

- Makes sense to base intelligent interaction on this metaphor

- Not possible if you break the assumptions that people instinctively make based on embodiment

Physical makeup plays a part in making or breaking the fluidity and naturalness of face-to-face dialogue

Users’ expectations about a system’s performance are based on what they sea and hear -- a good interface does not break users’ expectations

"Baldie"

DECface

• Perceptual abilities
  agents equipped with today’s computer vision couldn’t possibly recognize more than a handful of everyday objects, yet users might mistakenly assume that it can “see” objects in the surrounding just like they themselves can.

• Linguistic
  Users of speech-recognizing computers are invariably found to think that computers have a larger vocabulary than they actually do.

• Motor abilities
  Giving computer agents human bodies, users may easily think their agility equals their own

• Mental
  Looking human makes people think you are as smart as humans

"ToonFace"

• Pros:
  Fast rendering, hence quick update rates and faster communicative response time
  Muscle models don’t need morphing
  Rendering errors easily forgivable
  Fewer false expectations

• Cons:
  Perceiving which direction a face is ‘facing’
  Perceiving direction of gaze

Embodiment: What is Your Purpose?

When choosing an embodiment, you have to ask: What is my purpose with this embodiment?

• If it is to scare people, go with monsters

• If it is to make photo-realistic pictures, good for you

• BUT IF it is to support and sustain effective communication the embodiment needs to match the control structure, otherwise you will end up with false users’ expectations....

For a machine this match would be a metaphor, and the only measure of its “correctness” is that it is beneficial for the communication.