Consciousness API

A thought experiment.

Many accept that machines are now as intelligent as humans, perhaps more so. Whether or not you agree would depend on your definition of intelligence. That said, there are many definitions of intelligence for which some machines are more intelligent than most humans.

With intelligence conquered, the new frontier for machines is consciousness. Consciousness is even harder to define that intelligence. You know that you are conscious, but you can’t know if I am conscious.

This article is a thought experiment speculating what a conscious machine might look like. More specifically, if we could build a system that was conscious, what would its API look like?

For the uninitiated, API means Application Program Interface, or roughly what one would input into a program, and what output one would get out of it.

Inputs

You would give it sensor information to the outside world, like video, audio, touch, smell, taste etc.

Outputs

The system would take the said sensor information and process it in several stages.

• Feelings and emotions are at the heart of a conscious being. So, in the first stage, it would react emotionally to the sense information; positively, negatively, or somewhere in between. For example, it could react to the video of an advancing army with fear.
• In the second stage, it would convert the sense information and its emotions into abstract concepts; like “advance”, “enemy” and “weaknesses”.
• In the third stage, it would string together multiple abstractions into complex thoughts; thoughts like “the enemy is advancing; I fear it; I must destroy it”
• In the final stage, the machine outputs its decision. For example, “Should we destroy the enemy with a nuclear missile? — Yes”

class Consciousness:
    def sense(self, sensings):
        emotions = self.get_emotions(sensings)
        concepts = self.get_abstractions(sensings, emotions)
        thoughts = self.get_thoughts(sensings, emotions, concepts)
        decisions = self.get_decisions(sensings, emotions, concepts, thoughts)
        return decisions

Variations

Here are a few variations on the design above:

• The machine could output not just decisions, but some of the intemediate information. For example, an emotional machine, might share some of its emotions with the outside world.
• The output could be asynchronous. In which case, we could have a callback function passed into sense, for the outside world to get outputs as and when they are ready.
• There could be many looks within the logic above — not merely simple linear steps.

Concluding Reflections

The interesting question is whether a machine could satisfactorily simulate emotions, abstractions, thoughts, and decisions.

And more interestingly, whether humans could satisfactorily tell if the simulation is satisfactory enough.

Such a consciousness would probably want to create more consciousnesses and thus starting an interesting chain of events.

DALL·E (openai.com)