By the way, to understand what you mean by consciousness, do you consider yourself conscious when sleeping?
Consciousness is that thing that allows me to make a sandwich when I get up in the morning, so as not to die of hunger. How do you like this short and concrete definition?
The magic of consciousness happens when there are: 1- an agent, 2- the world around it, 3- self replication and evolution, that give the main thrust of purpose to the agent, the purpose of self replication being recurrent, it is just more self replication.
For example:
1. Why loop in RL produces consciousness and other kids of loops don't?
2. In the "perception -> judgement -> action -> reward" loop, how these different parts are qualitatively different. If you remove action, does consciousness stop?
4. Following that definition even the simplest LR robot of few hundred lines would be conscious and more complex industrial robot with advanced machine vision would not be.
Because it helps in selecting and carrying out actions, that are necessary for maximizing rewards. In real life, reward == more life, so, by natural selection, consciousness is that function that preserves the body.
> If you remove action, does consciousness stop?
If you remove actions, there would be no purpose to be learned by the agent, so it would not form values, so it would not have emotion. It would be just a feed-forward perception system that does not develop any response to anything.
> Following that definition even the simplest LR robot of few hundred lines would be conscious and more complex industrial robot with advanced machine vision would not be.
I'd say consciousness depends on the game the agent is playing, but there is always a kind of consciousness as long as the agent can answer adaptively to changing situations and learn to act better with time.
Would plants be conscious under that definition?
In complex organisms, gene regulatory networks coordinate the development of the body and internal organs. So it has the complexity necessary for consciousness, it's just a small size chemical-based consciousness, but the main ingredients are present - sensing, acting, learning and a goal - to maximize its own life and replicate, and let's not forget the most essential one - it is part of the world, which guides its evolution.
(2) does anyone in epigenetics share this view?
If your position is that those questions are meaningless, that's fine, but since they are a well-accepted part of the 'problem', you have to call that out explicitly rather than just ignore them.
I think the best alternative is Reinforcement Learning. In RL, there is an agent which exists inside an environment. It can perceive the world around, move about and perform actions. The agent has a goal to achieve and receives reward signals from time to time. Such an agent can learn behavior that maximizes rewards.
That's what consciousness is. It is not an experience, it is a whole loop "perception -> judgement -> action -> reward" that defines life moment by moment. This also explains the role of consciousness - it is to choose actions that lead to survival of the individual and of its genes (reproduction). It is all a self replicating loop, in the end. The purpose of life is life, the purpose of consciousness is to guard life.
Consciousness is a set of four functions (perception, evaluating actions, acting, learning from reward signals) that work together to select the next action, to protect the body, on which consciousness depends - yep, full circle.
This whole "RL agent inside environment = consciousness" theory has the advantage that it is concrete and not supernatural in any way, and has promising applications (such as AlphaGo and self driving car).