I would like to mention a few points that the above question helps to clarify. To do this, it is first necessary to discuss the concept of consciousness, then it is about how artificial consciousness can be produced and finally about the question of whether one is dealing with a copy of human consciousness in the end.
1. If we divide (human) consciousness into two aspects, the subjective aspect of experience and the ontological as well as the externally measurable aspect, then this is first about the latter. So, it is about the question of what constitutes ontological consciousness and how can it be reproduced in some way. After all, consciousness, as the totality of all thinking and feeling, arises as a property of life in — as far as we know — neuronal form. So, consciousness seems to be a function of life in neuronal form.
2. Life arises from a specific structure of molecules that sustains itself, adapts to the environment and evolves. Theoretically, such a structure can be recreated. The prerequisite is the constant exchange of energy, which must be regulated autonomously. Recreating the entire complexity of a living organism, including its brain, is likely to go far beyond current capabilities. So, it has to be a matter of reducing this complexity by not recreating the structures, but by recognizing the principles behind the structures. What, then, are the principles by which life organizes itself?
a. First, it is those reaction cycles that make up life. So, the basis is an autocatalytic system. If we take artificial neural networks as a basis, they must be enabled to carry out autocatalysis.
b. The basis of the coupling of structures can be represented as an association. This applies to the coupling of structures from DNA to those of the brain. It is the same principle described in psychology as associative thinking. Neural networks must therefore be enabled to behave associatively. It is always about network-by-network structures.
c. Why do structures couple with each other? It is valences that establish an association. I use the term by analogy with the valence electrons, which provide a coupling of atoms. In the realm of the living, valences refer exclusively to structures. For neural networks, this means that they must be topologically compatible.
d. How does such a topology come about? Let’s take human perception. Neither the sensory nor the cortical apparatus is able to reproduce reality one-to-one. Let’s take a tree. All billions of pixels (or voxels) cannot be represented in their entirety, either spatially or temporally. Inevitably, therefore, there is a coarse-grained representation, a reduction in complexity by highlighting essential topological points from the noise. This results in a topology of mountains and valleys (noise).
e. The mountains form a pattern that I call a metastructure. Thus, the metastructure protrudes from the sea of noise and ontologically creates consciousness. It is the metastructures that are compared with other metastructures. Neural networks must therefore be able to form metastructures. Metastructures are combined into new metastructures. This process is best described in terms of Piaget’s assimilation and accommodation, by which assimilated knowledge at a certain point (when there are enough elements) tilts into a new regime (the elements combine to form new structures) or, as we like to say, more abstract patterns emerge.
f. Thus, there is always an information gradient (more generally: structural gradient), which has a controlling effect on lower information density in that there is a higher reaction density here. A higher reaction density means that valences have a stronger influence on the coupling.
g. The process of the formation of metastructures, their coupling, the formation of information gradients, as well as the formation of islands or centers of higher reaction density leads to a movement towards higher or higher levels of adaptation to the environment. In summary, it can be said that it is an adaptive random walk within the given spaces of possibility. In this process, the value of nodes and edges of the neural network is changed and adjusted, and constant feedback ensures that the integrity of the network is maintained. So, we are dealing with a valence-based adaptive random walk.
h. But what makes such a constituted network make decisions on its own? This refers to ‘intrinsic’ decisions, not those that logically result from algorithms. What has evolved genetically over billions of years is the formation of stable islands of information with high density, or rather: an information center that can be described as an I. This ego is the ‘core’ of self-organized activity, it unites everything that has been said above, it is the center of information, valence and control. It unites what we call consciousness and subconscious, intuition and rationality.
3. If it is possible to construct an artificial network with the described properties, it is possible to observe actions that are very similar to human behavior. The artificial agent combines the properties described under h. If one takes the Freudian instance model, the ego can be supplemented with the superego by implementing metastructures from the outset that contain general rules of social coexistence and thus limit spaces of possibility. The IT means the evaluation of technical functional processes and informs the "I" about possible problems. As indicated above, we have described the view ‘from the outside’, i.e. excluded the subjective side of experience. Whether such a construct may produce something similar to consciousness from the point of view of the experience side due to the absence of the somatic component can be described as irrelevant at this point. His actions are fed by evaluations, even without them arising from somatic aspects. To what extent the somatic aspect plays a role, I describe as unimportant in this context. Even if we were dealing with a zombie, the project would be successful in the sense that you would have a machine that moves according to human criteria, i.e. makes decisions according to certain evaluations, but which does not matter whether it shows the side that we call the emotional world.
The result would be an ‘intelligent’ machine that combines human intuition and rational and logical abilities generated by algorithms, and that would be controllable by the ‘implemented’ superego.