www.Conscious-Robots.com - Re: Robot Mirror Test

Lo Último
13 Conf. Finlandesa de Inteligencia Artificial (STeP 2008) En el futuro, podremos crear máquinas conscientes Searching for Spirit Competición RoboChamps Microsoft Robotics Developer Studio 2008 CTP Abril Teorías de Orden Superior sobre la Consciencia Proyecto Feelix Growing

Lo más visto
Foro de debate PSYCHE-D Análisis I-Cybie Un Robot en Cada Casa MobileRobots Pioneer 3 (P3 DX) Robótica Cognitiva IA Fuerte y Conciencia Artificial Insectos Robóticos

Conscious-Robots.com Forum

Ruta :: Lista de foros

Machine Consciousness Forums

Test for Consciousness

Raúl

Moderador

Mensajes: 325

Karma: 4

Mirror Test - 2007/02/28 20:10 The mirror test is another classical test for consciousness generally applied to animals and infants. This test was originally developed by Gordon Gallup in 1970.

Instead of saying that this is a test for consciousness, I would say that is a test for self-consciousness because it is based in self-recognition. Certain animal could be aware of its environment, but not aware of itself, therefore it won't be successful in this test.

Scientists use the mirror test in order to determine whether or not animals are self-conscious. Have a look to this article for a recent related work done with elephants:

http://www.conscious-robots.com/en/neuroscience/mammals-brain/elephants-recognize-themselves-in-the- m.html

It is not clear to me whether or not an equivalent test could be devised to be applied in the field of machine consciosness. Raúl Arrabales Moreno. conscious-robots.com

El administrador ha deshabilitado la escritura de mensajes a los usuarios no registrados.

Plato Demosthenes

Usuarios

Platinum Boarder

Mensajes: 50

Karma: 5

Re:Mirror Test - 2007/05/08 23:16 Interesting idea...a sort of mirror program as a test for self-awareness. If you had a "sociable" artificial intelligence - one that could interact with others - then I suppose the test would be having it confront its own programming. Would its reactions be in the same steps as the elephants do it?

El administrador ha deshabilitado la escritura de mensajes a los usuarios no registrados.

Raúl

Moderador

Mensajes: 325

Karma: 4

Re:Mirror Test - 2007/05/10 01:23 Yup, I've thought about that a couple of times... The thing is that I can’t envisage a way to implement it. What I’ve been trying so far is to confront a machine consciousness implementation to a set of axioms for the presence of consciousness. However, this is done at the design level and completely depends on the truth of the axiomatic base.

An automatic test for consciousness oriented for artificial agents would be ideal in order to use it as part of the fitness function in an evolutionary environment. Taking the mirror test as the base idea for a test for self-consciousness in agents could be a good stating point. Thanks for the suggestion by the way

Let’s think about how we could ground to computational design the concepts of mirror and reflection (and of course, self-reflection). Raúl Arrabales Moreno. conscious-robots.com

El administrador ha deshabilitado la escritura de mensajes a los usuarios no registrados.

Raúl

Moderador

Mensajes: 325

Karma: 4

Re:Mirror Test - 2007/05/11 18:53

Social interaction seems to be key in chimps in order to pass the mirror test. Gallup work with chimps shows that isolated animals didn't pass the test initially, it was after interacting with other chimps when they passed the test. This could indicate that self-consciousness and consciousness of other are actually based on the same underlying mechanism.

Applying this idea to the field of artificial agents, the same mechanisms that allow an agent to be 'sociable' could provide it with some degree of self-consciousness (being able to recognise itself as a self).

[1] Gallup, G.G., Jr. (1977). Self-recognition in primates: A comparative approach to the bidirectional properties of consciousness. American Psychologist, 32, 329-337. Raúl Arrabales Moreno. conscious-robots.com

El administrador ha deshabilitado la escritura de mensajes a los usuarios no registrados.

Raúl

Moderador

Mensajes: 325

Karma: 4

Re: Robot Mirror Test - 2007/05/11 19:41 I've found a work by Takeno et al. where they claim that a robot can recognise itself in a mirror [1].
They define four steps:
1) The robot imitates the action of its own image reflected in a mirror.
2) The robot imitates an action taken intentionally by another robot as imitative behavior.
3) In an environment where the other robot is controlled completely via cables from the self-robot to imitate the behaviour.
4) The robot imitates the random actions of another robot.

The authors propose a evaluation method for discriminating the first step from the others.

Coming back again to the social dimension of self-awareness, it's clear that a requirement exists to recognise others' intentionality in order to be able to recognise yourself in a mirror. If the person (or robot) you see has no different intentions/behavior from yours, then it's you.

[1] Junichi Takeno, Keita Inaba, Tohru Suzuki. Experiments and examination of mirror image cognition using a small robot. Proceedings. 2005 IEEE International Symposium on Computational Intelligence in Robotics and Automation, 2005. CIRA 2005.

Post edited by: Raúl, at: 2007/05/11 19:46 Raúl Arrabales Moreno. conscious-robots.com

El administrador ha deshabilitado la escritura de mensajes a los usuarios no registrados.

Ruta :: Lista de foros Machine Consciousness Forums Test for Consciousness