I am a neuroscientist and a pragmatist. As we learn more and more about the brain consciousness will become less and less mysterious. In the end we will be left with nothing but the grin of the Chesire cat.

The membrane potential of neurons is an analog value, just like the voltage of a transistor gate. There is indeed a close physical analogy between the two and neuromorphic engineers can design chips to run at voltages near threshold that have many of the properties of neurons, such as conductance changes and voltage-dependent currents.

The answers to the questions about Hopfield ANNs are in The Computational Brain, Churchland and Sejnowski. We used Independent Component Analysis (ICA) to blindly separate voice and music.

I had to cut out the section of my talk on schizophrenia because of time limitations. There is increasing evidence that schizophrenia is an early developmental failure of neural circuits to mature properly, and especially the inhibitory interneurons that regulate activity in the cerebral cortex.

For a review paper look up "Sejnowski Behrens" in PubMed: http://www.ncbi.nlm.nih.gov/pubmed?term=sejnowski%20behrens

Autism starts to develop during the first year of life and one of the earliest indicators is an enlarged head circumference, which reflects an overgrowth of neurons. For a review paper see:

Courchesne E, Pierce K, Schumann CM, Redcay E, Buckwalter JA, Kennedy DP, Morgan J., Mapping early brain development in autism, Neuron. 2007 Oct 25;56(2):399-413

A good way to take to get started is to take a course on Computational Neuroscience. Other courses to explore are on Neuromorphic Engineering and Social Robots. Social robots are being used by psychologists to study child development. For a review see:

Meltzoff, A. N., Kuhl, P. K., Movellan, J., Sejnowski, T. J., Foundations for a new science of learning, Science 325: 284-288 (2009).

See above.

At a superficial level the cerebral cortex does look uniform, but on closer inspection the neurons in each area are specialized for different functions. In the visual cortex of primates there are even extra layers, and in the rat somatosensory cortex there are "barrels" for each whisker. The neural circuits are also specialized, and in the prefrontal cortex they are specialized for working memory, in comparison with primary sensory areas that is continually updated with new information that streams in from the sensors. Despite these differences there also common algorithms that allow the cortex to process and store enormous amounts of information, but even here different areas have different time scales and different neural codes.

We don’t know for sure what the computational power of the brain is since we mainly look at the electrical signals and ignore the biochemical signals and genetic regulation. What we study may just be the tip of the iceberg. Nonetheless, if the exponential increase in computational power continues it will eventually reach a point at which it will be possible to simulate brain functions. But then we will have to program the computer, which requires that we know how the brain works. More likely there will be an iterative process, where we simulate what we know at any given time and use that to make predictions and help design new experiments to improve what we know about the brain. Regarding consciousness, we can already see the wide variety of conscious states in humans and in other species, so I have no doubt that machines can become conscious too but probably not in the same way that we are. You would not be the same if you uploaded your consciousness into such a machine.

Noise is what is left in a measurement after you have accounted for all the signals you are interested in. This means that one man’s noise may be another man’s signal. There are sources of noise at the molecular level, such as the spontaneous release of neurotransmitter packets, but these are likely to have an important function. The so-called "spontaneous" activity that occurs during sleep is also likely to have an important function, which was what I alluded to in the talk with reference to the Boltzmann machine, which needs a "sleep" phase during which the internal correlations can be calibrated.

Artificial neural networks are most efficient when the inputs produce a pattern of neural activity that is sparse, that is, only a few of the many neurons are active. A new generation of artificial neural network models are being developed that take this one step further and uses random sparse projections of the inputs based on compressed sensing.

Yes, real neurons "wriggle" and are dynamic on a wide range of time scales. The spines on dendrites use actin, the same protein in your muscle, to change their shape. Even dendritic branches come and go, along with new synapses, even in the adult brain. The proteins themselves that make up synapses turnover on a time scale of hours to days. The turnover of proteins and synapses may have an important function in maintaining homeostasis – keeping all of the parts of the neuron in a balanced state that is sensitive to inputs. And homeostatic synaptic plasticity has been shown to improve the flexibility of neural network models.

Information can be cryptic. Take the human genome, for example. Having the sequence of base pairs is not the same as knowing what it means. The sequences greatly help biologists do experiments, but interpreting the sequences and their variants in different people will take a lot more work, and mistakes will be made along the way, as you suggest. What has changed is that we can collect a lot more information a lot more quickly than we could in the past, but we are still slow on interpreting the information. We need to speed up this process.

A wide range of brain differences have been identified in the brains of patients with autism spectrum disorders (see above). Some of these changes compensate for the primary disorder, which we suspect occurs during the early development of neurons in the brain.

Here is a source of more information at NIH:
http://www.nichd.nih.gov/publications/pubs/sos_autism/sub6.cfm

If you deprive the brain of sensory input, it goes into a free running state that is not normal. So the environment is necessary, and different environments can lead to profoundly different brain states. The environment and nutrition during the first few years of life are especially important in shaping a brain and preparing it for life. The recent earthquake and tsunami will leave its mark on the brains of the Japanese who lived through that experience, especially the children. Fortunately brains are resilient and can bounce back from even the worst environments.

It is easy to fool humans with words, so passing the Turing test is not a great achievement. Learning from experience is more difficult to program. An even more stringent test is to duplicate the sensorimotor capability of animals, which depend as much on the body as the brain. I don’t know how long it will take to achieve these capabilities in machines, but I am confident on the order of difficulty. Early researchers in AI thought that sensorimotor systems were the easiest to program, and the Turing test would be the most difficult.