018-Creating an AI or repper - awareness problems and issues
when we begin to think about an AI in terms of it's structure and processes we must confront the problems of representation and awareness directly. They are the first principles, they cannot be treated as second order objects in the construction of an AI.
in simple terms, representation is the mapping of some "sensory" data into an information structure. the less "mapping" by programmers that is done the better. because it is "mapping" that is the act of representation. we want the AI to develop representations, not us. otherwise the AI will not be an intelligence, it will be a program.
the second issue is how to make the AI aware?
what is awareness? we must answer this question to know when our AI is aware.
Representation making is awareness.
all the issues of making representations, the act of representing, reveal that representation making is inseparable from awareness. If there is no representation making, there is no awareness. When representation making occurs, we mean explicitly that there are objects that have representational relationships to each other. And to know those relations exist requires awareness.
when sensory data becomes "a thing" itself, that is the instantiation of awareness. When a representation becomes a thing itself, that is when the object "represented" is separate from the representation, which means there are two representations (the object itself and the representation of it). it also means that a possible simulation exists where the object (which is a representation) and the representation of the object exist within a larger representation.
when we speak about awareness and it's properties, we can speak about representation making and it's properties in exactly the same way we speak about representation. expansion of awareness is expansion of representations. altered states of consciousness are where different kinds of representations and representation making occur. the diminishment of awareness is exactly the diminishment of the ability to make representations. single minded (of one awareness) is where representation making is the constrained and does not vary. attention to detail, is representation making of the details of experience. big picture awareness is making larger connections, larger representations, about experience. ignorance, or being unaware, is being unable to make representations about something.
awareness can be so arbitrary and contradictory and confusing and wrong, precisely because representation making is arbitrary, contradictory, confusing, and erroneous.
in practical terms, and for the construction of an AI/repper, awareness is representation making.
it is not the storing or existence of representations (memory) it is the making of representations. Memory is about the content of representations, it is not about the making of connections the making of representations. breadth and depth of awareness is equivalent to breadth and depth of representation making.
there is one difference between awareness and representation making. An awareness exists in the face of contradictory or conflicting representations. the primary challenge of awareness is to represent divergent experiences (representations) into whole and larger structures of representation. But awareness is exactly that state where conflicting representations can co-exist. These contradictions may be categorized, or there may be some psychological defect, or there may be a dialectic of representation making to resolve contradictions, but awareness differs from representation making in the fact that contradictory objects can exist in awareness.
This fact induces representation making to expand in all directions at once (to resolve a contradiction for instance). It is awareness which functions to manage this expansion. the contents of awareness are the contents of representation making. The constraints of representation making are the constraints of awareness.... Or perhaps it is the reverse, the constraints of awareness are the constraints of representation making.
This seems to be exactly what we see in the study of neurological defects. If we describe awareness as representation making, then we see awareness limited precisely when representation making is limited. eg. http://en.wikipedia.org/wiki/Henry_Molaison
The fact that some representational content is unavailable, new long term memories for instance, does not change the act of representation making.
Or, as has been shown by Vilayanur S. Ramachandran's mirror box therapies, the making of representations changes the awareness of pain and experiences of the body. The presence of the mirror box experience opens up new representation making and this is a change to the contents of awareness. The representational change via the mirror box appears to reduce phantom limb pain in some studies.
Split Brain studies show how the contents of sensation have different representational results and it is the representation making that produces the awareness. Especially when the representation making is "incorrect", it is the representation making itself that produces the difference in awareness [Damasio - Descartes Error]
Synesthesia shows this same characteristic of representation making corresponding to awareness. that some experiences, some objects produce awareness because of the representations made about them. Synesthesia is special in this context because the representation making is automatic. But even in the Bouba/Kiki test, it's clear that associative representation making extends the awareness or richness of object experience.
While synesthesia and the bouba/kiki effect appear to be non-arbitrary representation making, we mean only that they are immediate. These experiences are not rationally decided but are felt immediately. There is little doubt that the synesthesia associations and the bouba/kiki effects are profoundly arbitrary in that it is the existence of neural connections which produces the experiences.
The works of Damsio, Cytowic, Sacks and others shows that the brain is the place where representation making occurs. Representations are stored, or embodied in the brain. What we see is that even when a memory is stored, the memory representation does not exist "outside" of awareness. For a memory to be experienced, it must be part of awareness, it must be put into the flow of representation making so that it can be ordered and structured and fit into the representations being made.
All the memory and demonstrations of cognitive loss show an alteration to representations, or to representation making. And these changes produce corresponding changes to awareness, to experience itself. This embodiment of representations, particularly in neurological structures is how representations become inputs into human experience. we remember and thus memories are inputs to representation making. remember, relive, recall, recite, revive. These are all words that mean "experience again". An AI must have a similar experience. It's memories are experienced again, experienced as inputs.
The idea of memory is all about representations and objects. The contents of memories, however they are stored and managed, but correspond to objects of awareness to be experienced. But this does not mean that memories, or even representation making must be conscious. Obviously there are memories that we experience that occur subconsciously or that produce subconscious behaviors or that have non-conscious antecedents. What is key is that memory formation is, in fact, representation making. memory recall is also representation making.
See also: Optogenetic stimulation of a hippocampal engram activates fear memory recall: Xu Liu, Steve Ramirez, Petti T. Pang, Corey B. Puryear, Arvind Govindarajan, Karl Deisseroth & Susumu Tonegawa
Npas4 Regulates a Transcriptional Program in CA3 Required for Contextual Memory Formation: Kartik Ramamoorthi, Robin Fropf, Gabriel M. Belfort, Helen L. Fitzmaurice, Ross M. McKinney, Rachael L. Neve, Tim Otto, Yingxi Lin
Can Memory Bias be Modified? The Effects of an Explicit Cued-Recall Training in Two Independent Samples
Janna N. Vrijsen, Eni S. Becker, Mike Rinck, Iris van Oostrom, Anne Speckens, Anson Whitmer, Ian H. Gotlib
Dynamics of dendritic spines in the mouse auditory cortex during memory formation and memory recall:
Kaja Ewa Moczulska, Juliane Tinter-Thiede, Manuel Peter1, Lyubov Ushakova, Tanja Wernle, Brice Bathellier, and Simon Rumpel
Parental olfactory experience influences behavior and neural structure in subsequent generations: Brian G Dias & Kerry J Ressler