- Related Stories
-
This is your brain on a microchip
May 11, 2006 -
Blueprinting the human brain
May 10, 2006 -
MIT group develops 'mind-reading' device
April 4, 2006
An "emotionally aware" computer being developed by Britons and Americans will be able to read an individual's thoughts.
The story "Coming soon: Mind-reading computers" published June 25, 2006 at 5:15 AM is no longer available on CNET News.
Content from Reuters expires after 30 days.






- Ohhhhhh, Puh-leeeeeeze!!! ...
- by Joe Blow June 27, 2006 2:52 PM PDT
- I was at the MIT Media Lab about 12 years ago, and saw a demo of what was purported to be a highly-accurate face recognition system that was based on models of facial musculature and bone structure, derived from a dual-camera system that used both visual and infrared data to detect these attributes that underlie the skin (thereby preventing use of makeup, latex prostheses, masks, etc., to disguise one's identity or attempt to look like someone else). Like most facial recognition software, it built up models of faces based on the ratios of the distances between facial features, such as between the centers of the pupils of the eyes, the corners of the mouth, the tip of the nose, etc. Using a database of several hundred volunteer MIT students, faculty and staff, the system was shown being able to perform 100% correct recognition on the subjects in the database. However, as soon as images of arbitrary faces were added, where the images weren't quite straight-on photos of faces under ideal conditions, especially using cameras/sensors not used to build the original database, the recognition rate fell off to well below 20 percent immediately, and got much worse as the number of samples increased. This is known as the "plateau effect", where a system performs well within a very limited domain of inputs, but as soon as the data moves off the plateau, performance and accuracy literally fall off a cliff. This is similar to what happens with generic voice recognition, handwriting recognition, and other such technologies.<br /><br />I wonder why anyone is continuing to fund this kind of malarkey, since any first-year graduate computer science student should be able to explain the difference between context-sensitive (which is what these kinds of problems involve) and context-free systems (e.g., computer programming language parsing - it's completely structured, with no ambiguities). These kinds of projects are going to keep failing to deliver unless and until high-resolution/wide-dynamic-range sensors, distributed processing power, and memory technology approach that of the human brain, not to mention the very-difficult-to-develop parallel-processing software to tie all the hardware together. Even then, it's going to take a long time to debug such systems, and there's a whole lot more important R&D that needs to be done in other, more pressing areas, like how to eliminate spam, once and for all.<br /><br />Anyone who funds technology research ought to at least be familiar with the first principles of the technology on which they're spending (usually taxpayer) dollars. However, the MBA crowd will tell you that management is an art, not a science, and that anything can be managed by a properly-schooled manager. Yeah, right, and maybe that's why things like the Space Shuttle keep turning into lots of little pieces, along with the multitude of other colossal management debacles like Enron, the Dot.bomb bubble, etc.<br /><br />Oh, but don't pull the purse strings too tight just yet - I have this great idea for turning lead into gold - it's gonna make us all bazillionaires ...<br /><br />All the Best,<br />Joe Blow
- Like this Reply to this comment
-
(7 Comments)