In his book The Feeling of What Happens, Antonio Damasio offers a neurobiological theory of consciousness. Things happen to every object, but consciousness, he argues, emerges when there is an acknowledgement that it is to me that these things are happening. In this framework, consciousness is the recognition that there is a you, that things are happening to you, and that, because of this, you have changed. Specifically, Damasio states that “consciousness is the part of mind concerned with the apparent sense of self and knowing” (Damasio, 1999, p. 27). Importantly, this framework distinguishes feeling from knowing. Feelings and emotions emerged as regulatory states to help animals fulfill their evolutionary goals, but before consciousness, the owner of these emotions was oblivious to them. Consciousness arises as the feeling of a feeling, that is, being aware of one’s feelings.
In summary, Damasio’s model of consciousness begins with a proto-self: an interconnected coherent collection of neural patterns that represent the state of the organism, moment by moment. The proto-self is the foundation for the core self, a second-order account that occurs whenever an object modifies the proto-self. That is, core consciousness occurs when the brain generates an account of how the organism’s own state (the proto-self) is affected by the organism’s processing of an object. Finally, the core self is the foundation for the autobiographical self, which integrates the core self over time and thus gives rise to the richer, extended consciousness we humans experience. Damasio’s book is fairly extensive and offers much evidence for this theory, so if you are interested, I would highly recommend it.
Even without accepting his entire theory, it is intuitive that consciousness requires something like the proto-self: a singular and stable reference for the self, along with a means to generate it. Damasio argues that this representation of the self is grounded in a variety of neural patterns that ultimately represent the body. To serve as a stable and singular reference for the self, this representation must be remarkably invariant over time, so that it can offer the continuity the self requires. Damasio thinks that the body is well suited for this, because the range of possible states the internal milieu can take is highly limited, and although our cells may undergo high turnover, the architecture of the body is maintained.
Many leading scientific theories of consciousness focus on what goes on in the brain (Seth & Bayne, 2022). Complex, higher-order processing in the brain is arguably necessary for consciousness, but it may not be sufficient. Consciousness, as discussed above, requires a strong, stable representation of the self. For biological beings, the body provides such a representation. But what about machine intelligence? Artificial intelligence (AI) does contain robust and strong ideas of the self. However, these appear to be generic, just as it may also contain robust and strong ideas of emotions, math, and likely all of human knowledge. The smoking gun of machine consciousness will be evidence of consistent and robust internal processing that resembles “how is <the idea of my self> modified by <object I am processing>.” AI certainly has incredible representations of what it’s processing, where that is the prompt it receives or the image it generates. The missing ingredient is a representation of itself that remains stable and is consistently understood as being modified by its interactions with the world. I believe this representation is hard for an AI to attain unless it is explicitly designed for it, an idea proposed by Damasio himself (Man & Damasio, 2019). Consciousness and intelligence may have been bundled together in animals, but only because of the body, which offered the perfect substrate on which to construct this sense of self. In fact, most evolutionary goals are directly tied to the body; as long as the body persists, so does the animal (and its goals, beliefs, intentions, etc.). There is no such substrate for machines. The space of information that is spanned by AI does not include an innate, inalienable sense of this is me.
This view of consciousness certainly entails that LLM chatbots are not conscious — they don't have a self to monitor, so they don't have the self-monitoring function.
But it also is kind of an outline of how to create an AI system that would be conscious. Continuous learning, robot control, and automated datacenter management (i.e. senses and attention pointed towards the 'body' where the AI itself is running) should do it.
(The chatbot that does your coding isn't conscious, but its cousin that runs the maintenance drones and the Kubernetes cluster might be. Also, it runs the maintenance drones, so don't give it a hard time. Yes, it tells terrible puns; it's a sysadmin. Please don't call it the MCP, and definitely don't call it the BOFH; it has a thing for picking up personas ...)
The self or consciousness? I understand you to say the self.
I (repeating ideas of others, e.g. Joscha Bach), think that consciousness comes first. The self is only necessary as a computational substrate, not necessarily as a well defined object with borders yet I think. Consciousness then defines those borders. (E.g. are your fingernails part of you? Are your gloves? Your eyes? Your glasses? Your cochlear implant?) I would say the borders of the self are very much fluid over time.
In his book The Feeling of What Happens, Antonio Damasio offers a neurobiological theory of consciousness. Things happen to every object, but consciousness, he argues, emerges when there is an acknowledgement that it is to me that these things are happening. In this framework, consciousness is the recognition that there is a you, that things are happening to you, and that, because of this, you have changed. Specifically, Damasio states that “consciousness is the part of mind concerned with the apparent sense of self and knowing” (Damasio, 1999, p. 27). Importantly, this framework distinguishes feeling from knowing. Feelings and emotions emerged as regulatory states to help animals fulfill their evolutionary goals, but before consciousness, the owner of these emotions was oblivious to them. Consciousness arises as the feeling of a feeling, that is, being aware of one’s feelings.
In summary, Damasio’s model of consciousness begins with a proto-self: an interconnected coherent collection of neural patterns that represent the state of the organism, moment by moment. The proto-self is the foundation for the core self, a second-order account that occurs whenever an object modifies the proto-self. That is, core consciousness occurs when the brain generates an account of how the organism’s own state (the proto-self) is affected by the organism’s processing of an object. Finally, the core self is the foundation for the autobiographical self, which integrates the core self over time and thus gives rise to the richer, extended consciousness we humans experience. Damasio’s book is fairly extensive and offers much evidence for this theory, so if you are interested, I would highly recommend it.
Even without accepting his entire theory, it is intuitive that consciousness requires something like the proto-self: a singular and stable reference for the self, along with a means to generate it. Damasio argues that this representation of the self is grounded in a variety of neural patterns that ultimately represent the body. To serve as a stable and singular reference for the self, this representation must be remarkably invariant over time, so that it can offer the continuity the self requires. Damasio thinks that the body is well suited for this, because the range of possible states the internal milieu can take is highly limited, and although our cells may undergo high turnover, the architecture of the body is maintained.
Many leading scientific theories of consciousness focus on what goes on in the brain (Seth & Bayne, 2022). Complex, higher-order processing in the brain is arguably necessary for consciousness, but it may not be sufficient. Consciousness, as discussed above, requires a strong, stable representation of the self. For biological beings, the body provides such a representation. But what about machine intelligence? Artificial intelligence (AI) does contain robust and strong ideas of the self. However, these appear to be generic, just as it may also contain robust and strong ideas of emotions, math, and likely all of human knowledge. The smoking gun of machine consciousness will be evidence of consistent and robust internal processing that resembles “how is <the idea of my self> modified by <object I am processing>.” AI certainly has incredible representations of what it’s processing, where that is the prompt it receives or the image it generates. The missing ingredient is a representation of itself that remains stable and is consistently understood as being modified by its interactions with the world. I believe this representation is hard for an AI to attain unless it is explicitly designed for it, an idea proposed by Damasio himself (Man & Damasio, 2019). Consciousness and intelligence may have been bundled together in animals, but only because of the body, which offered the perfect substrate on which to construct this sense of self. In fact, most evolutionary goals are directly tied to the body; as long as the body persists, so does the animal (and its goals, beliefs, intentions, etc.). There is no such substrate for machines. The space of information that is spanned by AI does not include an innate, inalienable sense of this is me.