Mind/Brain and ME theorising

[user9876]

Cognition is way ahead of neuroscience on these things - just look at Seidenberg and McClelland, who in the 1990's were developing computational models of how each individual learning episode contributes to, and subtly changes our knowledge base. Prediction, feedback loops, adjustment of weightings in response to unexpected feedback, all these things have modelled computationally without reference to the brain. But a lot of the neuroscience is being done by those who don't know this literature (and that's probably not the neurospsycs, that's the psychiatrists and the neurologists).

Some of the connectionist NLP work (such as McCellands) was interesting in the implications for the way processing happens in that rather than having long inference chains that would happen in a logical deduction system there are associative maps between representations that are formed from sets of activations (or really a point in a high dimensional vector space). Some of the feedback loops etc then can be handled quite naturally. It always felt like a much more natural model for cognition than more logic based approaches. But I never really saw how it mapped from neural networks onto a brain like structure. I suspect there was a bit of a naive view on that in the late 80's and early 90's but I suspect it will have evolved.

 

[Woolie]

@Snow Leopard, good suggestions on how to proceed!

On your first set of questions:

How do each of the imaging technologies work?

I'm not best placed to give you a rundown on the physics, but for our purposes, the most important thing to know is that fMRI measures oxygen uptake (the more active a brain region, the more oxygen it takes up). Therefore, the fMRI signal can be influenced by caridovascular factors (heart rate variability, etc.). This is a problematic confound when using it to study the physically ill.

The other thing to know is that fMRI isn't great at getting a signal from some brain parts. Cortically, these include the temporal and frontal "poles" (the anteriormost portions). Generally speaking, however, this issue is unlikely to be critical for the types of imaging studies we'll be looking at here.

What are the common biases?
Which are likely to be the most useful?

Not sure about biases, but the biggest problems are not with the technologies but with: 1) psychological task design; and 2) interpretation of results.
This is a big issue for another day (sorry!)

What new and novel technologies are being developed?

Not a new technology as such, but the big buzz right now is brain connectivity. This involves statistically analysing how activation (measured by fMRI) changes over time in different brain regions - to see whether they are "talking" to each other. So if an increase in activation in region X is quickly followed by an increase in Y - and this happens consistently - it would suggest the regions are influencing each other in some way. This is called functional connectivity.
Functional connectivity can be measured while a person is just resting under the scanner ("resting state" fMRI). Or it can be measured while the person is performing a particular active task. The second one tells you how brain regions communicate during that task, which could be very different from what happens at rest.

Structural connectivity is also big now. That's when you map the major white matter tracts in the brain using a specific type of fMRI called diffusion tensor imaging.

Another big noise at the moment is Multivoxel Pattern Analysis (MVPA), another statistical technique applied to fMRI data. If you have an fMRI scanner with enough resolution, you can not only examine which areas are activated during a particular task, but also build a map of the actual pattern of activation in that area. People produce different patterns in response to individual stimuli. For example, there is a distinct pattern associated with seeing a word like "cat" and another one for seeing a word like "lamp". People have claimed this is like mind reading because you can use the patterns to get an idea exactly what the person was thinking about at the time.

Other technologies:
MEG (Magnetoencephalography) is also big. It works differently from fMRI, as it detects changes in the magnetic fields that occur over the scalp as you perform different tasks. The magnetic changes occur when large groups of neurons fire at the same time. So its picking up on a really different activity from the one fMRI measures, which is changes in the blood. MEG's huge plus is that its very good at unpacking the time course of events during performance on a particular task. For example, if you hear a word, which cortical regions are the first to become activated, and how long does it take for the signal to be transmitted to other particular brain regions? (e.g., regions that process the word's meaning).
MEG isn't good for examining activity in deep brain structures because it detects changes that are detecable from th scalp only.

TMS (transcranial magnetic stimulation) is getting bigger. Here we're talking about its use to temporarily interfere with function in a particular brain area for a short period. We can then examine what effect this temporary "dysfunction" has on performance. Then hopefully, learn more about what the brain area contributes to the task.

 

[Christian Godbout]

If the author of "Bright air, Brilliant fire: On the matter of the mind", the late Gerald Edelman, were here, he would have loved this discussion...

 

[alex3619]

But I never really saw how it mapped from neural networks onto a brain like structure.

As part of my PhD candidacy I designed a neural network mapping system that enabled a wide variety of connectionist architectures, based on published equations used in neurological modelling. It could create a network of any size or complexity, providing you had the hardware to run it. Brain structures have layers with different functions and rules, small modules if you like, that can combine to create larger layers. I was not trying to model the brain, rather I was looking at using brain architecture to create better connectionist systems. There are ways of doing that, and I was working toward that, but then I lost my math and reading abilities as my ME kicked in big time.

The system was largely built by a friend of mine though, who was working on a related question. My brain was in serious decline at that point. Our use of the system was very different though, his question was on something to do with representational remapping, I forget the details. In other words, we were looking at different network structures and problems.

Each sub module could have a different purpose and function. It is possible, has been since 1988, to create modular connectionist systems. The technology existed even then to build them to any size. The hardware limitations were still an issue then though as an operating network was relatively small but the systems needed to build and train a big network were relatively large.

The technology to build more sophisticated neural networks has been around since 1988 but largely ignored.

If the author of "Bright air, Brilliant fire: On the matter of the mind", the late Gerald Edelman, were here, he would have loved this discussion...

Gerald Edelman was the scientist whose equations I was using as a starting point, and who figured out some of the cell surface markers that enable neurons to find their place and function in the brain.

 

[alex3619]

This is a problematic confound when using it to study the physically ill.

Especially for diseases with blood, vascular or lung issues, which might include ME.

 

[Woolie]

It always felt like a much more natural model for cognition than more logic based approaches. But I never really saw how it mapped from neural networks onto a brain like structure. I suspect there was a bit of a naive view on that in the late 80's and early 90's but I suspect it will have evolved.

@user9876, if you're interested, there's some great more recent stuff that uses PDP networks with various different learning styles to model the memory functions of the various temporal lobe memory structures:

http://compmem.princeton.edu/publications/normorei03_inpress.pdf

Its an extraordinarily hard read (I'm not kidding!), but @user9876, you just might have the background.

 

[user9876]

@user9876, if you're interested, there's some great more recent stuff that uses PDP networks with various different learning styles to model the memory functions of the various temporal lobe memory structures:

http://compmem.princeton.edu/publications/normorei03_inpress.pdf

Its an extraordinarily hard read (I'm not kidding!), but @user9876, you just might have the background.

Thanks looks interesting.

 

[Jonathan Edwards]

@Jonathan Edwards, I think good cognitive neuroscientists are clear on the levels of description thing, and they understand that some questions involve knowing where/how at a neural processing level and some questions involve knowing what those insights can - and can't - tell us about cognition more generally. I'm more worried about how others outside the field may be misled by the blurring of levels.

You will see that i put some stuff on interoception on the other thread Woolie, which I think is the place to continue that discussion because it is directly relevant to ongoing research.

Apart from talking to PWME on PR and going to meetings I spend almost all my time with cognitive scientists together with philosophers and I fear they do not have a clear idea of levels. The fact that they are continually arguing with each other at cross purpose more or less proves that I think!! Our multidisciplinary group does not include Mark Edwards but other local people involved in predictive coding and perception, like Patrick Haggard, are regular members.

 

[Marco]

I don't expect there'll ever come a time when we can localise any particular type of experience to a particular brain part. Where do emotions happen? Ar they in the bodily changes that accompany them? (e.g., heat rate increases)? Or are they in the brain responses that occur? Early psychology tried to pitch these as either/or possibilities, but of course, they are very likely to be all true. Emotional experience is likely to reflect the sum activity of a constellation of interacting and constantly changing brain and body states.

A critical reason why all this optimistic talk of 'human-like' artificial intelligence is pie in the sky. You need to inhabit a human body to develop human intelligence.

 

[user9876]

A critical reason why all this optimistic talk of 'human-like' artificial intelligence is pie in the sky. You need to inhabit a human body to develop human intelligence.

But I think a lot of the AI work would not claim to be human like. Much of it is sets of inference, search and pattern recognition techniques.

 

[user9876]

Not sure if this is relevant but its a debate about hysteria including Mark Edwards so may be of interest

I've not looked.

Published on Oct 20, 2015
'Is Freud's book Studies on Hysteria still relevant?' - Debate aimed at the general public. Chaired by Dr Tim Nicholson (Institute of Psychiatry / Maudsley Hospital, London). Motion proposed by Richard Kanaan (Professor of Neuropsychiatry, Melbourne University) & Stephanie Howlett (Psychotherapist, Sheffield, UK). Motion opposed by Mark Edwards (Professor of Neurology, St George's London) & Alan Carson (Senior Lecturer in Neuropsychiatry, Edinburgh University). Public event at the Freud Museum London & funded by the UK National Institute of Health Research.

 

[Cheshire]

Wow, just watched the beginning. I thought France was the last refuge for psychoanalysis, I was plain wrong. Freud is still revered by some in the UK.

The same old story: anecdotal evidence, “I’ve seen cases of true hysteria therefore everyone with functional neurological symptoms is hysterical”. “You have to dig deep into the psyche of these people, because it’s all about repressed memories. If you don’t find anything, it’s because it’s an intra-psychic conflict, the patient can’t handle the trauma, that’s why they can’t remember”…

Mark Edwards seems rational and listening to people compared to the others…

 

[Marco]

But I think a lot of the AI work would not claim to be human like. Much of it is sets of inference, search and pattern recognition techniques.

I agree and I'm not knocking AI - just utopian notions of replicating human intelligence rather than it's successful application to domain specific problems. But more generally I was just agreeing with @Woolie that supposedly higher (or at least accessible to the conscious 'mind') cognitive processes are likely to have their roots in the body.

 

[Hip]

Apart from talking to PWME on PR and going to meetings I spend almost all my time with cognitive scientists together with philosophers and I fear they do not have a clear idea of levels.

Would you have any views or biases regarding the so-called hard problem of consciousness, ie, the issue of explaining how we appear to experience internal mental phenomena (qualia), like the various hues such as red, green or blue, that we only seem to experience within our minds.

Generally people tend believe either that consciousness, and the qualia that appear within it, can be explained in terms of ordinary neural information processing, or that consciousness may be some fundamental, transcendent, or non-material aspect of the universe.

In their views on the hard problem, the personality of individuals seems to strongly influence their beliefs. For example, in my case, I was always interested in meditation and mystical aspects of the world (though unfortunately I can no longer "tune into" this as a result of brain fog); so I tend to be biased towards the idea that consciousness is a transcendent entity. I did Cog Sci MSc after a maths/physics degree, and I also think that you will need to bring in physics to explain consciousness. But these are just my personal inclinations.

I believe from prior discussions that @alex3619 holds the opposite view, and thinks consciousness will eventually be explained in terms of neural information processing in the brain — ie, will be explained at the regular material level of reality, without the need of invoking any transcendental aspects.

In any case, irrespective of personal bias or leanings, we still have to adhere to the scientific method, and ultimately have to validate any theory of consciousness by empirical evidence.

 

[Snow Leopard]

A critical reason why all this optimistic talk of 'human-like' artificial intelligence is pie in the sky. You need to inhabit a human body to develop human intelligence.

What would happen if the neural networks of a human brain were mapped 1:1 to an artificial substrate? Yes it's science-fiction, but philosophically speaking...

 

[Jonathan Edwards]

Would you have any views or biases regarding the so-called hard problem of consciousness,

In their views on the hard problem, the personality of individuals seems to strongly influence their beliefs. For example, in my case, I was always interested in meditation and mystical aspects of the world (though unfortunately I can no longer "tune into" this as a result of brain fog); so I tend to be biased towards the idea that consciousness is a transcendent entity. I did Cog Sci MSc after a maths/physics degree, and I also think that you will need to bring in physics to explain consciousness. But these are just my personal inclinations.

I believe from prior discussions that @alex3619 holds the opposite view, and thinks consciousness will eventually be explained in terms of neural information processing in the brain — ie, will be explained at the regular material level of reality, without the need of invoking any transcendental aspects.

I have had a busy day and have come home to far too many scintillating comments on this thread to know where to start. Why is it that intellectual levels on PR are so much higher than at scientific meetings (certain individuals excepted I admit)? Why do we have to do the scientific critique for these people? (Tongue slightly in cheek I suppose.) The video will probably keep me amused before bedtime - thanks for that.

But I love this last line

In any case, irrespective of personal bias or leanings, we still have to adhere to the scientific method, and ultimately have to validate any theory of consciousness by empirical evidence.

and I will try to respond to Hip simply because this is my most familiar ground and responding to Woolie and co will need some gathering of thoughts.

I have be trying to address the 'Hard Problem' since 2002 and have come to a firm view, part of which is that the hard problem as originally expressed is based on a false assumption. The key point relates to the conception of 'regular material level of reality'. What got forgotten in the twentieth century is that physicists from Newton and Leibniz in the 1680s onwards via Maxwell and Mach to JJ Thompson in the 1880s realised that there is no 'regular material level of reality' in physics. Physics has only ever been able to describe chains of causal connection that can end in experiences (observations). All the equations of physics tell you what experience you should get at a particular place and time if you start off with certain causal tendencies at other places and times (thus 'mass' is a causal tendency). Whether the causal links are 'material' is a matter for the man in the street and naive realist philosophers to consider. Physics says absolutely nothing about stuff, only about what happens and how one event can lead to another, and always judged in terms of ability to affect experience at the end.

So, as Leibniz worked out, physics can only be about how individual dynamic, or causal, units relate to the universe. Things got obscured because early physics dealt with aggregates, like Newton's gravity applying to planets, but modern physics is laid out in Leibniz's terms. The dynamic unit is a quantum mode (which does not have to be small; the earth can actually be a quantum mode but this gets very technical) usually called psi and the universe is a field of potentials usually called Vxyzt and the relation is called Schrodinger's equation. It works at all sizes.

Leibniz had the genius to see that experience or perception must be the same thing - the relation of a dynamic unit to the universe. There are lots of complications along the way but I think he was right and so we do not need any dualism or transcendence. We just have two descriptions of the same relation from different points of view. One is a chain of physical relations ending with an experience and the other is an experience determined by incoming physical relations - and in fact entirely necessarily and determined by the last, or immediate relation, as physics says too, which is important for when I reply to Woolie!

What people tend to worry about with this sort of general explanation of experience is that it has to allow everything in the physical world to be 'conscious' - it is what is known as panexperientialist. But this does not matter because we can never test for the presence of experience anywhere other the one here and now, we can only test for the operational relations, which we call physics. Moreover, we have every reason to think that the sorts of experience we have will only occur if the incoming relation to whatever is here and now has been pre-prepared in a highly sophisticated way by a collating machine called a brain. What is a mystery is what the quantum mode that feels pain or senses the taste of basil is. It does not need to be a traditional subatomic particle or anything like that but it has to have the right relations.

I could go on for pages on this, as you can imagine, but that is enough for tonight. Interestingly the man who I think puts the case for this best is the father of psychology, William James. But James hit a problem in making it work and gave up despite having given his own answer a few pages earlier.

 

[alex3619]

What would happen if the neural networks of a human brain were mapped 1:1 to an artificial substrate? Yes it's science-fiction, but philosophically speaking...

Currently that is completely impossible. We cannot even map a single neuron to a different substrate. Its a complex three dimensional network of biochemistry, involving chemical conformations, reactions, electrical and other activity ... we do not know enough to say what is important, only that some things are very important.

However I do think that if we could map a brain to an artificial substrate we will have created something that blurs the line between AI and human intelligence.

I do think consciousness will be completely explicable by normal brain function. However I would be delighted if we could prove the mystical. It would be a very much more interesting world.

For the record, I have also explored the mystical, and been heavily into meditation. Its modern versions of empirical rationalism, and the scientific method, that interest me the most though. Show me the evidence, and show me sound reasoning, and show me how it was tested, then standby for questions.

 

[Jonathan Edwards]

What would happen if the neural networks of a human brain were mapped 1:1 to an artificial substrate? Yes it's science-fiction, but philosophically speaking...

David Chalmers addressed this problem at length. He concluded that you ought to get the same experience but with the weasel caveat that the simulation would have to be at 'sufficiently fine grain'. In my view the sufficiently fine grain is the grain of fundamental physics, although the size involved might be microns (dynamic grain and size are not the same). In effect you need cell membranes, cytoskeleton, synapses...

 

[alex3619]

Quantum phenomenon, based on my own training, are counter-intuitive and very difficult to grasp. Quantum mechanics violates many presumptions we have about the world. Even the science of this is still being figured out. I strongly suspect that at most only a handful of people really understand quantum science as it is today, and much of our advance in understanding involves esoteric mathematics. Anything could lurk there, but I do not want to presume I know what that is.

There is indeed mystery in the universe, from the big bang to quantum physics. Is string theory right? Or is it something else that explains how things are put together? Mystery is not gone, and science will in time delve deeper and deeper into that mystery.

If nonphysical elements we might call mystical are involved in the mind or consciousness I think science will eventually reveal that, even if we understand it about as well as quantum mechanics. What I don't know is when, or how important it will be.

Ultimately, my position is that reality is what it is, but our understanding of it is flawed. Science is about fine tuning that understanding over time, and is always a work in progress. Its more process than destination.

 

[user9876]

David Chalmers addressed this problem at length. He concluded that you ought to get the same experience but with the weasel caveat that the simulation would have to be at 'sufficiently fine grain'. In my view the sufficiently fine grain is the grain of fundamental physics, although the size involved might be microns (dynamic grain and size are not the same). In effect you need cell membranes, cytoskeleton, synapses...

Are you suggesting you need quantum effects which I think Roger Penrose suggested a few years ago or are you just referring to the scale and detail you feel is required?