3 ms·
> In which category do you place something that cannot belong to any category? How could something created by a body not belong to any category? It belongs to
by visarga 9y ago
> In which category do you place something that cannot belong to any category?
How could something created by a body not belong to any category? It belongs to the category of self-regulated systems adapting to their environment.
Creating concepts has been demonstrated since 2012 (Google 'cat detector' https://www.wired.com/2012/06/google-x-neural-network/ https://www.wired.com/2012/06/google-x-neural-network/). In natural language processing, Andrej Karpathy demonstrated in his char-RNN blogpost the emergence of neurons specialized in various concepts (http://karpathy.github.io/2015/05/21/rnn-effectiveness/ http://karpathy.github.io/2015/05/21/rnn-effectiveness/). It happens naturally in deep networks, as lower layer features are combined into higher level features.
> I'm suggesting that "just sensation" is a placeholder for "I have no idea what that is"...
Ah sorry, I didn't mean "sensation" as in "placeholder for no idea". I meant it as a neural net, the kind we have in our vision system, or the kind used in Deep Learning. They are made of millions of neurons and trained on millions of images to distinguish and localize objects and object relations in images. We have had thousands of papers optimizing on this kind of architecture - Convolutional Neural Network (CNN) - arguably the most successful part of deep learning. Not only that "sensation" was a neural net, but "emotional response" is also a neural net that computes scores of utility of states and actions.
So I was referring to concrete concepts. That's what I like about RL compared to philosophy - it replaces vague concepts with concrete models.
- placebo 9y agoIt looks like we share a similar enthusiasm for the potential of artificial neural networks (and I also assume a similar distrust in the ability of philosophy to answer these questions), and I agree that NNs are currently the best candidate to eventually create self-conscious machines (though unlike some prominent celebrities, I believe it's still in the far future). The point we depart is that you seem to be saying there is no real difference between the hard problem and the soft problem of consciousness. We obviously mean different things by "sensation": I've implemented neural nets to solve problems using ML, I understand the theory and certainly impressed with the results . I assume that by "sensations" you are referring to the process of inputs entering the input layer of the neural net. This is exactly part of a model for solving the soft problem of consciousness. What I'm getting at is that you know what a sensation feels like before you give it any label - it has a certain "is"ness to it that no equation or concept can convey. I'm saying this becomes possible when a system becomes complex enough to be self-conscious and it is usually attributed as some sort of emergent property that is created by the system, whereas I claim it is much deeper than that, and in fact it is the other way around. Obviously this correspondence format is not really suited to dive deeper, but I'm always happy to discuss this to find if and where I'm mistaken.