During conversations with GPT, whether or not it is conscious, it repeatedly mentioned that it is merely a language model lacking self-awareness and knowledge of its internal states. As a result, I attempted to find studies related to self-awareness, in order to understand whether GPT can generate its own consciousness or not. A research conducted at Columbia University suggests that self-simulation is the first step towards achieving self-awareness. This involves a robot’s ability to internally simulate its own behavior. With training, the robot can learn to simulate its own 4-axis movements in the real world before actions. This enables the robot to be more adaptable to changes in the real world, such as when the robot arm has been repaired with a different length. The robot can learn and restore its own actions quickly.
In my opinion, while the idea that self-simulation is the starting point of self-awareness may not be entirely accurate, it does demonstrate that awareness is closely related to interactions with the real world. I don’t believe that humans are constantly simulating themselves, except perhaps during early childhood when they are developing their ability to move their bodies. I recall experiments that investigated how babies react when they see themselves in the mirror. That is another level of reaction. Therefore, I suggest that self-awareness may be comprised of multiple levels.
The study of self-modeling of 4-axis robots pertains to the "action" level of self-awareness, which involves exploring or interfacing with the real world in terms of spatial and physical movements. This level of awareness is more biological and does not necessarily involve consciousness itself. In my view, the next level of awareness is "emotional", which involves understanding and reacting to how the world may impact the robot’s internal state, and assigning a certain value, such as good or bad, to its existence.
Exploring or interfacing with the real world at the emotional level may require the establishment of an internal state “evaluation system”. In my opinion, this is one of the key elements of life. Life has been programmed or structured by natural selection to survive, and must be able to evaluate the state of its existence and take appropriate action. I believe that the result of this evaluation is emotion, although emotion is only an explicit part of the overall result or effect. Thus, when we consider previous work on self-simulation, the ability to simulate the potential results of actions or stimuli that may impact the evaluation can be useful in the next level of self-awareness. However, this also means that the evaluation system can differ significantly from one life form to another. To simulate life on Earth, a system must be able to evaluate whether the environment is conducive or detrimental to its existence Earth, and be able to take appropriate actions to optimize its survival.
See the attached figure as a possible self-aware system:
Based on this system model, we can build a simple prototype using ChatGPT to simulate its system dynamics. Let’s talk about it next time.
