2 ms·
> No matter how precisely we could specify the mechanisms underlying, for example, the perception and recognition of tomatoes, we could still ask: Why is this p
by visarga 10y ago
> No matter how precisely we could specify the mechanisms underlying, for example, the perception and recognition of tomatoes, we could still ask: Why is this process accompanied by the subjective experience of red, or any experience at all? Why couldn’t we have just the physical process, but no consciousness?
The difference comes from context. It is one thing to have a feedforward network that detects tomatoes, another to have a reinforcement learning agent that optimizes for survival in nature.
The RL agent has a value function that assigns a predicted future reward to the current state and actions available. The value network, together with the sense data, is "what it feels like" for the agent to see the tomatoes. Maybe it knows that red tomatoes are good, or that red fruit are ripe, and they can give it a reward signal by reducing hunger. That makes perceiving red an experience colored by emotion. Maybe agents that didn't see red tomatoes as good, had less chance of survival and died off.
Also, the RL agent is not a simple feedforward net, but a loop: perception->judgement->action->effect in environment, and rewards. It is a dynamic process, evolving from moment to moment. In that continuous perception-action-effect loop, there is space for the inner world and all its complexities.
- SomeStupidPoint 10y agoWhy? All of those could be feedback/computation systems without any experience. Your answer only argues for why the system would have certain computational features, not a subjective aspect.
- visarga 10y agoMy answer explains why there would be an emotion attached to a perception. The subjective aspect is a result of the network being an agent, that has to fend for itself in the greater environment. The stake of the RL game is survival itself. Each new state it finds itself in opens some possible future paths and closes other paths. Some actions lead to rewards, other to losses. I think subjectivity lays in this space of values, options and actions. That places is where "it feels like something" to see red. And if you're still unconvinced, remember what we are: a collection of cells, a protein based dynamical system that has computing and learning abilities. Why would it feel like something to be a protein-based computer?
- SomeStupidPoint 10y agoYou've missed the question: why is there a perception at all for red to seem like that in?
- ppod 10y agoWhat do you mean by 'perception'?
- SomeStupidPoint 10y agoThat there's any subjective experience, that there are things I perceive or experience. The question is how do I explain having a subjective experience at all, and do toasters have one?
- abiox 10y ago> that there are things I perceive or experience you have vision and other senses, and physical phenomena that interact with them cause physiological changes. > how do I explain having a subjective experience at all isn't experience necessarily subjective? i'm not sure what non-subjective experience would even mean.
- visarga 10y agoToasters are not self-replicating systems like living things, so they don't share the same problems. An organism has to fend for itself and reproduce in order to exist. Perception appeared as a mechanism that allows organisms to adapt to changing external conditions. With experience, they learned complex behaviors by reinforcement learning. Every moment, it perceives the world, judges the value potential of possible actions (instinctively), then selects the action and executes it. This loop requires the agent to have an evolving internal state. This is 'subjective experience'. It's a mechanism by which organisms behave in a more adaptive way.
- visarga 10y agoYou mean, why can an agent that sees an apple also recognize the color red itself? That's trivial, it's because the red color is a low level pattern, learned by the agent even before the apple pattern. If you look at the weight maps of convolutional neural nets, in the first layer you see lines, in the second layer small shapes, then parts of objects, then whole objects as you go up layer by layer. The red color is probably somewhere in the first layer.