I was surfing around for find cool stuff for my new T610 Sony/Ericsson phone and came across this news on the official Bluetooth web site:

Sony today announced the world’s first Bluetooth monitor. Such a monitor was long dismissed as impossible because of the limited bandwidth of Bluetooth, but the Sony SDM-B101S makes it work…

The obvious WTF comes to mind, given that my bluetooth experience is limited at 3kbs. So, never for a second I think this is a joke and I don’t look at the date of the news. I do follow the link to the original news which says:

April 1, 2004 – Sony today announced the world’s first Bluetooth monitor. Such a monitor was long dismissed as impossible because of the limited bandwidth of Bluetooth, but the Sony SDM-B101S makes it work. The actual display panel is a modified version of a high-end Sony 17-icnh LCD. The main challenge engineers faced was power management. The monitor plugs into the wall for power — hey, it’s a baby step — but video data is transmitted via Bluetooth.

The monitor ships with a USB 2.0 module that communicates only with the monitor. The locked pairing ensures that other Bluetooth devices won’t steal the bandwidth.

The monitor will ship in Japan Fall 2004 and will retail for roughly $4,000.00. An American model will ship April 1, 2009.

So, at this point is obviously a joke, but it’s such a plausible one (for non-techies) that those marketing heads over at bluetooth.com didn’t get it. Now, besides the humor and the cluelessness (or boredom) of the curators of that web site, this shows something that I’ve seen emerging in my studies on sub-symbolic artificial intelligence: give sufficient lack of interest and attention, the brain seems to perform the same sort of analysis that statistical unstructured text algorithms do.

My personal experience in fact shows that my continuously-retrained bayesian spam filter is far better than I am in filtering spam after a few hundreds messages.

It seems to me that what we call “attention” could be a connection between some first-pass sub-symbolic prefiltering and some higher semantic processing areas of the brain. And those tend to disconnect when the amount of signal/noise ratio drops lower than a certain point, leaving the first part of the brain performing the analytical part by itself and performing worse than today’s machines on similar tasks.

Computer scientists spent decades and billions of dollars on AI research and produced very little that could be used.

Subsymbolic spam filtering (LSA, Bayesian, etc.) is, IMO, the first artificial intelligence that beats a human under similar conditions.

But we still have a looooooong way to go if we want to have a spam filter that will reject all spam unless it’s funny.