Rob Conscious’ Day Zero
Richard Feynman used to say:
“What I cannot create, I do not understand”
I was preparing a lengthy article on "Robot Consciousness", including aspects of information abstraction, levels of awareness, sentience thresholds, etc. etc. …
Then I thought: "What if we'd try to make one?" It would be definitely be much more fun. Of course, while writing this, part of me (and just about everyone else on the team) is thinking: "This is totally crazy. It's never going to work. Just think of the sheer complexity of conscious experience. There are just too many open questions. How would one even know if a machine became conscious?".
On the other hand, why not?
Let's make a genuine attempt, and see how far we get. It will be interesting to see where we get stuck —for instance, is a non-von Neumann, possibly neuromorphic, hardware necessary, as it is widely believed?
And then look how we might go beyond.
Goal: Build a Conscious Robot.
We know where we're aiming at, but we don't know how to get there. No one knows yet. Hence it's going to be an iterative process, a creative exploration of deliberately concrete nature: We want to make something in order to understand it. We will use any tool and resource that we deem useful: Deep Learning, Bayesian inference, simulations, GPU, smartphones, motors, force-sensitive resistors, electrophysiological data, connectomes,… But we will start small and always keep it as simple as possible. The overarching philosophy will be that of trying to kill hypotheses (similar to the Google X Moonshot philosophy of killing projects): codify knowledge and devise simple experiments with the goal of refuting the hypotheses. Then iterate while refining the hypotheses. In this case, the null hypotheses we make are about implementing and interpreting state-of-the-art knowledge and raw data of consciousness research. Especially at the beginning, we expect the results to be underwhelming. And it will be definitely messy and complex all along.
Additionally, we will Design for Collaboration: Because of the dynamic nature of consciousness, we want to provide simple and intuitive ways of testing and experimenting with our progress, and let any constructive contribution in. For instance, we might start with a smartphone app with extensible elements such as an evolvable public genome, rearrangeable capabilities, BLE connectivity to Arduino-based motors, or cloud updates.
Rob Conscious will keep a public diary to provide context.