Neuroscience cannot read your mind
April 3, 2010
I’ve written about Newtonian science and the simple cause and effect interpretation of the physical universe that it embodies, and how the mathematics of complexity and statistical interpretations of the physical universe, such as quantum mechanics, have superseded that mechanistic view. What I would like to suggest is that neuroscience is treading the same path in its interpretation of the mind as a mechanistic system demonstrable by a physical understanding of brain function. This is not a new idea as far as I know, but I need to ‘say it out loud’ to show myself that I know where the limits lie.
I’ve been thinking about whether I believe that the mind could be replaced by the internet, and I think ‘no, but there could be functions that could be farmed out such as memory‘. Here I’m going to explore that idea with specific reference to the blossoming fields of neuroscience, neuromarketing, neuroethics and neuro-anything-else-they-can-think-of.
The mathematics of complexity and uncertainty, chaos theory, complex systems, or however else we wish to express the concept, all share the same fundamental tenet; that simple mathematical relationships can give in unpredictable results. This is shown by the Lorentz’s Butterfly Effect, but it it is also embodied in Godel’s incompleteness theorem and the Schrodinger’s Cat thought experiment. To me these are all similar aspects of the same idea; that we can never measure all variables sufficiently well to be able to have a 100% reliable model.
If we transpose this notion to the pursuit of neuroscience, where claims are being made almost to the point of being able to read the minds of experimental subjects, we should at least consider the nature of the systems involved before we accept the validity or even applicability these kind of claims.
Brain structure is primarily a function of the expression of DNA of the individual and DNA as a replicator a very mechanistic and ultimately predictable system. I say this because genes are quite simple. Their complexity is in their size, not their building blocks. That the Human Genome Project was able to sequence our genes using automated techniques suggests that a mechanistic approach to reading that material is appropriate. However once the brain has started to develop neuronal connectivity in response to stimuli (memories start to be formed in response to experience), that mechanistic interpretation is no longer applicable without having a set of meta-data that shows the context under which those connections are made.
What neuroscientists are doing using fMRI is establishing a work-book of that contextual meta-data under experimental conditions, and its very impressive that they are managing to exclude enough of the outside world to be able to see human responses such as lying and trust and jealousy in the data that they collect. I’m sure that, on average, they are seeing some functions of mind being expressed physically. BUT, what cannot, and indeed must not, be inferred from this work is that the responses from one individual’s brain can be directly equated to the responses of another individual’s brain.
We could go though a significant proportion of the human race, taking subjects from all walks of life and every corner of the globe and find average response curves for each chunk of the brain, but we would never be able to replicate the contextual meta-data to a fine enough resolution to be able to counter Godel’s incompleteness theorem as it applies to basic information or the individual’s brain development in response to experiences from its own unique viewpoint. The mechanistic interpretation of mind, that equates brain activity to mind function, breaks down under existing mathematical interpretations of the physical universe. We would need a whole new mathematics to be able to do what is currently being claimed for neuroscience. To be fair to the neuroscientists, many of them shy away from the grand claims, but enough are not that we see fMRI being cited in legal cases. Far from free will being dead and neuroscience proving a deterministic worldview, it is showing just how poor our quantitative understanding of mind really is.
This is not a new experience. Psychoanalysis promised an understanding of mind and motivation at the beginning and middle of the 20th century and arguably was the basis for the construction of the consumerist global economy. I wonder how far neuroscience will be pushed outside the lab.
Proponents see recent fMRI science as analogous to genetic fingerprinting; as a quantitative diagnostic tool. I would argue that it is more analogous to a form of psychoanalysis where interpretation is automated. In many of the new institutes and companies working with fMRI we see the objections to the wider application of fMRI-centered neuroscience being characterised as philosophical and relating to ideas of free will and determinism. I don’t see that as a valid or even relevant conflation. My counter-claim is that what is being claimed for neuroscience is not mathematically possible and that in ignoring the role of mathematical complexity scientists, lawmakers, economists and others are acting unethically. What is being seen is the brain and not the mind. That the brains responses are linked to the mind shouldn’t be a surprise but the simple Newtonian idea of cause and effect is not applicable where 100 billion neurons each have around 7,000 synapses many of which have been influenced by memory formation or physical conditions since, or even before, birth. Simply put, just because a specific cubic centimeter of grey matter demands extra blood flow in response to the same stimuli, it doesn’t mean its for the same reason.
If it is possible to mathematically model the mind, then it should be considered as a complex system inhabiting another complex system (the brain) and informed by a set of contextual meta-data (memories and experiences) as well as environmental stimuli. Divining motivation from brain activity is a step too far mathematically, but an approximation could be possible with a sufficiently large database to populate response curves with experimental data. Whether those response curves could provide useful predictive data can’t be known at this point, but what we can say with a good degree of certainty is that you’d need a large n-value to compensate for the free variables in two complex systems and the contextual meta-data.
Veracity Values Redux
September 5, 2009
The redoubtable Wikipedia has started the ball rolling on visualising veracity. Its not in the wild yet, but they have started testing.
Reported in Wired, NowPublic, and with tongue in cheek here.
Here is a search result from Wikipedia all about just how complicated it actually is to do this and a discussion on the different approaches that you might make at this kind of thing.
I’m glad that Wikipedia has made a move on this. Looking at the way that they are approaching it the algorithm looks relatively simple, at least nothing like as complex as it could be, but I suppose the issue for a non-profit is external cost more than anything. The reason that I say its relatively simple is because it only looks at internal data to come to its assessment of how much you can trust the information on a page. I don’t like that by the way. Trust is not what you want in data. Verifiability is what you want in data. Everything else is faith.
Reading the Wired article I think that I agree with the researcher from Palo Alto who said that normal readership would probably find it distracting. My idea would see veracity as much more like a security certificate that you could investigate if you wanted, but would otherwise be just an icon somewhere on the browser (or whatever is being used at the time).
So taking it to the next level, where external data sources are checked rather than just counted as this implementation seems to do, and you need to run it under a different business model. Until all data is open access, the scientific publishing hegemony is broken and no bad people exist checking that data is true, or even verifiable, will cost money. Last time I looked a single scientific paper costs about $30. At that price its no wonder a professional class exists to exclude mass access to information. When I do research I might look at 100 papers, properly read 20 or so and need to read more than the abstract of 50 to see if they are relevant. That’s a hell of a commercial barrier to entry. $3000 to properly research a note that no-one may read. That’s the sort of return that most speculators would balk at, hence academia. Or at least old model academia, where research topics were chosen by the researcher just because they want to know more about something.
So the moral to the story is – if you want to really kick off the knowledge economy using the new economics of data proliferation break up the scientific publishing houses, or at least force them to open their vaults a chink.
PhD Place
May 28, 2009
I’ve just heard that I have a place to do my doctorate, to start this September.
My research question is about how information flows between individuals (such as you and I) interacting online (as we are doing now) are translated into actions in face space.
I won’t bore you with the whole text, but I’ll be using the debate around energy, how its ‘won’, distributed and used, to explore how directly what is said here online relates to the actuality.
Around that are issues of subjective and objective truth, leadership, social network formation and evolution and the formation of social norms within those social networks. In a wider context there are applications within future political debates and engagement, cross-boundary politics, establishment of new social norms, assessment of online veracity and even interpretation of citizen journalism.
I’m looking forward to it. I just need to find a few extra quid to make sure that I don’t go hungry doing it 
Veracity Values
March 2, 2009
How do you know what is true ?
As a third party, how do you know that something is objectively true as opposed to subjectively true ?
Can you have such as thing as a truth ‘Richter scale’ ?
There are technological tools out there that should be able to help us with verification, even before the semantic web breaks through. Text mining should be able to pull out the key words and phrases of a document. You need to find a set of rules to feed the miner, but apart from that the issues with text mining seem to me to be logistical these days. Feed those results into the relevant search engines and databases and you should be able to automate the evidence gathering process.
Rate the relevance and ‘authority’ of the source and you could derive a ‘veracity value’ for the document.
Place some security around that veracity value and you have a ‘veracity certificate’ that has links to the evidence and a sliding scale of verifiable-ness.
The problem comes in identifying objective vs subjective. For example; take two academic papers, one on the physical qualities of a new metal alloy the other on Mayan cultural artifacts and their relevance to modern day Mexico. They have equal numbers of citations, equally authoritative reviewers, equal external coverage in conference proceedings and equal numbers and quality of references. How does a reader know which is objective and which subjective in nature ?
That’s something that humans do all the time, and frequently get wrong, but that AI has yet to approach.
Perhaps we leave that to the human for the moment and leave the automation at veracity.
The business model writes itself, so if anyone wants to try it get in touch
Catching up with the Digital World
February 28, 2009
This is a keynote speech delivered by Richard Sambrook of the BBC at the Media Re:Public Forum held at USC Annenberg in March 2008.
Interesting summary of the state of the art (a year ago), for me that’s pretty up to date. I’m on a bit of a catch-up right now.
I think that Richard’s concentration by omission on the conflict between objective and subjective realities is actually, though probably old news, still very relevant. The questioners (after 36:00) kind of prove that with their takes on his presentation.
I think that (subjectively of course) there is a set of tools out there allowing easy subjective reporting. The objectivity side of things is something that must be cultural.
For those interested in the objective truth, objectivity can be a rasion d’etre to the detriment of real politique. Whether driven by a moral purpose that information should be correct and hang the consequences, or whether they are fundamentally positive about the human condition. First hand reporting and hard data will always be a superior method of moving towards a better AKA a truer world.
For those whose focus is influencing change, whether it be getting the litter bins emptied more often or a major national political shift (some would argue that amounts to the same thing in the UK), maybe the weight of conviction is the lever that must be used to get closer to a subjectively ideal world.
Do we want Spock or Bones as our information go-to guy ? Spock may be hard work and not much fun at a party, but you know that you’re going to get the facts. Bones could be telling you want he wants your to hear or what he thinks you want to hear. Maybe that doesn’t matter in most circumstances. Maybe in some situations you want the passion. Maybe you need both. In either case you probably need to know which you are getting.
