Christophe emailed an interesting reply to address the question of robot consciousness and anxiety:

Robot evolution to consciousness

The subject of robot's anxiety as highlighted by steve on April 24th deserves some answer in terms of possible evolution of robots up to consciousness based on human evolution. The proposed process of evolution from animal to humans with creation of self-consciousness does indeed contain anxiety as part of an evolutionary engine, but other options may be possible for application to robots. Let's begin by reminding the proposed evolutionary scenario before addressing a possible application to robots. (Details in Evolution of Representations. From Basic Life to Self-Representation and Self-Consciousness )

1. Early primates (lemurians, say) carry representations of their conspecifics as well as "auto-representations" (representations of themselves with no notion of self-representation).
2. Pre-human primates were capable of some intersubjectivity that gave them the possibility for a limited identification with their conspecifics, like our today great apes (mirror neurons level in evolution).
3. This identification with conspecifics allows for a merger of representations that bring auto-representation to access the meanings associated to representations of conspecifics. So the auto-representation gets access to the meaning of an organism existing in the environment with an identity, and recognized as so by subject and by other conspecifics. I propose to consider that these new meanings give to the auto-representation some first elements of self-representation.
4. Performances resulting from this self-representation have some selective advantages in evolution and will participate to the evolution of self-representation up to self-consciousness. The role of anxiety in this process is with an evolutionary engine. (See Fig V.1). Anxiety is naturally generated/amplified by identification with endangered or suffering conspecifics (we are at the pre-human primate time frame where survival of the fittest was the law). Too much anxiety is difficult to live with and has to be limited. I propose that anxiety limitation for pre-human primates went thru the development of empathy, imitation, language and intersubjectivity. This provided a positive feedback loop on identification with conspecific which was the starting point of the process. So for the corresponding evolutionary engine.

Step 1) looks achievable as is, in principle, with robots.

For step 2), some work is needed to have it implemented within robots. We need there to make possible some kind of "inter-agency" between robots where a robot becomes capable of "understanding/guessing" what a conspecific robot is to implement (is like "intersubjectivity" among pre-human primates). This needs some clarification of the relations between "theory- theory" and "simulation-theory" in order to get a good enough understanding of the overall evolutionary process for application to robots (a). Some conceptual effort may be needed, but there should be no major obstacle.

Step 3) is the key event where auto-representation gets access to the meanings associated with representations of conspecifics. This would be the result of the merger of robot auto-representations with representations of conspecific robots, with associated meanings. It should be implementable in usage oriented robots. At this level, we would have the first elements of self-representation appearing in a robot where the meanings related to the representation of conspecifics ("entity existing in the environment") are attributed to the auto-representation. Auto-representations (initially devoid of any notion of self-representation) will then contain new meanings identifying a beginning of self-representation. This would be the toughest part to analyze and realize. A lot is to be done first in understanding what has been the exact content of this step for pre-human primates in terms of data processing, and then see how it can be implemented within robots.

Regarding Step 4), the evolution of robots from self-representation to self-consciousness may or may not utilize the same evolutionary engines that the ones that were used by pre-human primates. And the anxiety/anxiety limitation process is not mandatory for the evolution of robots. We can imagine/program robots with other evolutionary engines. It could be decided to favorise empathy between robots (for their group survival), as well as imitation, language and inter-agency, but there is no obligation to relate it to some anxiety limitation. Such choices will condition the content of the emotional world of the robot that will result from this evolutionary process. Robots emotional world can be close to ours or be very different.

There is also another item to take into account in addition to the above. It is about "free will" within robots, as it is generally accepted that self-consciousness and free will are tightly related ("free will" as conscious selection of action by an agent). Our self-conscious robot will have an autonomy up to the level of free will. But let's consider the question about robots capable of choosing their own deeds as beyond the subject of this post. It will be addressed later, understanding that it should not be eluded as it is closely related to robot self-consciousness. Comment are welcome.

Christophe Menant

(a) http://www2.unipr.it/~gallese/Gallese-Goldman%201998.pdf