Category Archives: AI

As we try and link ourselves ever closer to technology, the industry of human-computer interfacing is developing rapidly.   I thought I’d check out the state-of-the art brain-PC interfaces, on the basis that if any of them seem good enough, they may change the way we use the internet forever.

1. OCZ Technology – Neural Impulse Actuator (£73)

Pitched as a gaming device, this headband combines a electromyograph (which detects the electrical potential generated by muscle cells), an electroencephalogram (which records electrical activity across the scalp), and electrooculograph (which detects electrical activity associated with eye activity).  Phew.  That’s the pitch anyway, although I think it may be simpler – it essentially detects electrical activity at 3-4 spots across the forehead, and uses software to try and discriminate between the different sources of that electrical activity: your forehead muscles, your eyes, or indeed your brain.

Various reports including this one from OC3D suggest that after a LOT of training, the headband can allow you to shoot enemies in games by slightly raising an eyebrow, or clenching your teeth.  Response times (promised as at the speed of thought) haven’t been shown to be much faster than mouse, even after quite some training.

Verdict: Unproven, probably too simplistic, but an interesting start

2. Emotiv Systems – ($299)

Emotiv have been developing their headset for a couple of years now, and it’s finally being released on New Year’s Eve.  Again, their target is console games, and Emotiv use both scalp electrical activity, and sensing of facial expressions to deliver gaming control.  Their 2008 launch was delayed because “The public demo didn’t go as planned; the device simply didn’t work in front of the media who attended the press conference”.  Hm.

As it hasn’t been launched, it seems that few independent review have been done, so the jury is still out.  However, just on the basis that it detects facial expressions, it probably has greater potential than OCZ’s

Verdict: Unproven, but facial expression recognition could be interesting

3. Neurosky Headset – ($199)

Neurosky appear to have built the biggest ecosystem around their headset – games, integration with music visualization, bluetooth cellphone integration.  However, their headset only has 1 (yes, 1) sensor, which supposedly shows your “emotional state and how you react to the world”.   This really is the most remote from believable “brain control”.

Verdict: Interesting (and smart commercially)

Conclusion
I started looking at this area thinking I’d find a bunch of companies trying to commercialise “thought control”.  I only found three key players, and none of them represent what I can see as being a realistic attempt at getting mass-market mind control of computers.

Part of the problem is that it’s hopelessly simplistic to try and read the brain by measuring electrical activity across the scalp.    Of course, it can result in some simple communication or control of a cursor on the screen – which is a life-saver in the caseof severe disability or paralysis, where it’s enabled a patient to communicate.  However, the ability to do complex tasks just isn’t possible yet (or ever?) without invasive brain surgery – or a far greater number of more sensitive and calibrated sensors across the scalp.

However, there is hope in sight.  DARPA, for instance, has $4m in 09/10 for “Silent Talk” – a project that aims to “allow user-to-user communication on the battlefield without the use of vocalized speech through analysis of neural signals.”  And DARPA generally does interesting things with its research budgets – cars that can autonomously navigate complex environments being probably the highest profile example.

I’m watching this space – the day we can control computers using a mix of keyboard, mouse AND thought (I don’t believe that thought will, for a long while, replace manual control) is that day that the internet becomes truly fast and immersive.  I can’t wait.

UPDATE: The crew at Neurosky were switched on enough to catch this post, and send me a whitepaper with some stats in it.  Seems their single electrode compares favorably with a Biopac, a well-used medical wet-electrode EEG system.  I don’t have time to analyse the data, nor can I verify it – but if it’s true, then good job Neurosky (I’m actually more impressed with the fact they caught this post).  I’m still unsure about how useful the technology is today, though…

In somewhat of an overstatement, that is how a scientist described progress on robot AI over at Willow Garage.  Head on over to Willow to see “PR 2 Alpha” navigate into another room, and there plug himself into a  wall socket to recharge. 

It’s not quite jumping into the car to fetch Pizza, and really, it’s a very long way from our persistent dreams of useful, humanoid robots.  That seems to have been the status quo on Robot AI for quite some time now.  Behind the scenes though, they promise us, greater things are happening to the code that underlies this sort of behaviour.  Certainly these guys, Microsoft, and probably half a dozen others are trying to come up with a Robot OS – a common platform for developing Robot AI on.  This should help, but I expect that the leaps/bounds towards humanoid robots will take something more significant.

I spent the morning with the bright students at Hult University here in London who are working on a marketing project for my company right now, and we spent a bit of time discussing how A.I. and those sorts of technologies could start to become part of next wave of technologies.  At a stretch, what’s happening with Robots like PR 2 Alpha is part of this (if only he can find a way out of the building…).  But I believe it will be this kind of work, combined with a LOT of different, diverse and currently unconnected technologies to make the leap to something that can “fend for itself”.