In a study at UCLA, subjects were asked to do one of 8 different tasks – from saying whether words rhymed, to counting the number of tones they heard.  So – very different, varied tasks. 

Investigators using fMRI (functional magnetic resonance imaging) – the very cutting edge of brain scanning – and then had to guess from the results of the scans which tasks the subjects were doing.

Chance alone would get 1/8 or 12.5% of guesses correct.  They managed 80%.  Pretty good – but let’s just remember that they were looking at tasks that were very (very) distinct.  And they still got it wrong 20% of the time…

How did they do it?  With machine learning math (the same kind of math that we use at Cognitive Match).   And yes, it’s a bit of a good result for this field of technology.  But far from perfect, and very far from mind reading.

Of course, all this means is that fMRI and the way we interpret the results from that technology could be better.  Much better. 

But more importantly – maybe it’s just the wrong thing to do (which is my take on it).  It’s a bit like working out whether the stock market is up, based on satellite weather images.  The brain just doesn’t work based on large areas of electrical activity on a scan.  It’s complex, enmeshed, multi-layered, abstracted and part of a body of integrated biological mechanisms.   And I hope that’s clear.

Firstly, I noticed ma.gnolia.com was down. Aside from a frustrating domain name, they had a reasonably successful social bookmarking service. Sadly, due to lack of backup (!), they’ve lost the majority of the bookmarks/favorites that they stored on behalf of their users…

Bang, useful personal data gone.

Secondly, I tuned into “More or Less” a great statistics-focussed radio show on the BBC, on a recommendation from my Dad. Aside from a really great interview with the author of “Sustainable Energy: Without the Hot Air” which I’ll write about another time, the presenter mentioned Daytum. Setup by the Nicholas Feltron, the guy who exposes his personal stats meticulously collated and designed up at feltron.com each year, the site enables you to have your own “Personal Dashboard”.

Thirdly, I spotted an ad which had a “YouData” logo on it. Smelling a 2.0 startup, I checked out the site - and yes, it’s a (US based) service that lets you sell your attention – the old “pay me to advertise at me” model, but brought up to date.

So how do these strands tie together? Well, they are all about people realising that their own data is:

1. Valuable and useful to them
2. Valuable and useful to others
3. Therefore, has a monetary value

Problem is, losing bookmarks at Magnolia is a greater value by some margin than what someone like YouData would pay for that data. And so that’s the opportunity – finding a way to bridge the gap between how much I value my data and time, and how much others (typically advertisers) value it. The answer may be that in most cases, that gap can’t be bridged?lady gaga poker face

- how happy and productive you are
- your social status
- your social group

Fundamentally this is just an extension of any form of data mining – take a large amount of data, and try and make some determinations from it.  The examples based on social group and status can be fairly easily explained – by where you spend your time (the types of shop, street, district), and other mobile phones that yours tends to hang out with.  Happiness and productivity was a correlation they discovered when they combined location and call data with questionnaires.

The same group are doing some interesting work with other areas that use mobile data – such as “social serendipity” – trying to match users that happen to be in similar locations, and that have similar profiles or interests.  People have tried to release products into that space for as long as I can remember, but no-one’s yet cracked it, so it will be interesting to see if this research helps.

A lot of reality mining to date has been to do with mobiles (like BlueTooth MyBlogLog), but obviously anything that can sense us and feed data about us will add to this: cars, PCs, toasters… The more, to my mind, the merrier.

Well, actually Phorm’s ideas are, if not that exciting in terms of what they’re doing with data, a valid attempt to try and personalise advertising content for users.  By observing what individual users do (in “buckets” of interest category), and then replacing ads within the OIX with ads that match that category, they are at least attempting to deliver more interesting and relevant content. 

They are heavily restricted by their domain of influence (ad spaces within OIX sites), and by the fact that the hyperbolically negative PR (or justified concern, depending on your viewpoint) has restricted what they actually store about a user’s behaviour. 

So where is the benefit as a user in this area of behavioural targeting?  Well, ideally the targeter will work as a “trusted friend”, providing input and ideas for that user based on everything they know about that person.  Trusted friends have access to a fair amount of information – the more trusted, the more information.  They (ideally) won’t share that information, abuse or expose it.  And ultimately their suggestions will motivated by being to the benefit to the USER.  Advertisers, shops, other people – all come second.

That, anyway, is where I think this industry is heading, slowly (and often unwillingly).  And so the winners in this market space will have found a way to take the position of that trusted friend, and grow the trust, the profile and the benefit over time.

No suprise that I’m working on just such an idea…