Could Dreams be the Gateway to Understanding Consciousness in People and Chatbots?

 Article at Medium

The first truly sentient AI will be a chatty self-driving car that decides where to go on vacation, and dreams about the trip beforehand.

Recent strides in dream research may have quietly led to a simple and practical definition of consciousness.

Researchers investigating neural correlates of dreaming have discovered a hotspot (posterior cortical hot zone), which when activated is associated with dream recall on awakening. Conversely, suppression of this region is predictive of reports of dreamless sleep.

This finding is consistent with a theory known as Prefrontal Synthesis. In this model, percepts abstracted from our interaction with the environment are represented as distributed neural ensembles, circuits defined by the synchronous firing of constituent neurons according to the principle of ‘what fires together wires together’.

When an ObjectNE as these ensembles are known is activated, we become conscious of the corresponding percept. When the prefrontal cortex activates a sequence of ObjectNEs in the course of planning or problem solving, we become conscious of our own thoughts. And spontaneous activation of ObjectNEs by the posterior cortical hot zone (primarily during REM sleep) leads to the more freewheeling experience we know as dreaming.

The interesting thing here is that a concise definition of consciousness just falls out of this approach. In a nutshell, consciousness becomes synonymous with experience. We say we are conscious when we are interacting with the environment, albeit always indirectly through our stored percepts:

1) If we are experiencing the formulation of new ObjectNEs through the consolidation of sensations, we call this being conscious of our surroundings.

2) If a sequence of ObjectNEs is activated by the prefrontal cortex, this is experienced as reflecting on something or being conscious of our thoughts.

3) And if the ObjectNEs are spontaneously activated and recombined during sleep by the posterior cortical hot zone, we call this (altered) state of consciousness dreaming.

All this has interesting ramifications for the recent debate over whether language-parsing algorithms should be called conscious. There is no question that such programs build up the equivalent of ObjectNEs, as weighted neural networks arise through interaction with big data. By analogy, one could argue that during the training process, the program is converting sensation to perception and is thus conscious of its (informational) surroundings. Likewise, as these percepts are referenced and organized during external queries, one could argue that the chatbot is potentially conscious of its own (millisecond-long) train of thought.

Taken together, these two phenomena may be responsible for the impression popularized by some researchers that AI is already ‘a little conscious’.

If we accept the idea that consciousness equates to ‘interaction with the environment’ (always mediated by internal percepts), additional mechanisms would be required before AI could be called truly conscious in the ordinary sense. Specifically:

1) Before a program could be considered conscious of its surroundings (‘awake’), it would need to be in continuous training mode, consistently forming new perceptions and consolidating them with previous learning. Not impossible by the way: think the marriage of a self-driving car with a chatbot.

2) There would need to be a revolution in the understanding and coding of executive functions (volition) to replace the user query system with a self-querying design. The resulting potential for autonomous reflection would be one step towards an ongoing stream of consciousness…and maybe an awareness of self.

3) Spontaneous reactivation of ObjectNEs during downtime, perhaps through some Darwinian analog to random mutation and natural selection (with an eye to increased storage efficiency?) would lay the groundwork for ‘electric dreams’.

Note that equating consciousness with experience does nothing to explain qualia (the redness of red or the taste of an apple). A thorny question for another day…

For more reading…Siclari F, Baird B, Perogamvros L, Bernardi G, LaRocque JJ, Riedner B, Boly M, Postle BR, Tononi G. The neural correlates of dreaming. Nat Neurosci. 2017 Jun;20(6):872–878. doi: 10.1038/nn.4545. Epub 2017 Apr 10. PMID: 28394322; PMCID: PMC5462120.