Both are right.
The ostensible contradiction only comes from different angles of view ... and from exactly what is the term 'overexposure' supposed to mean. Bruce was looking at the topic from a technical point of view, and 'overexposure' in his context means overexposure of the JPEG image that comes from the capture parallel to the raw file (or from the raw file via standard settings in the raw converter). When the JPEG is (slightly) overexposed then the raw file often still will have some headroom to work with. Michael is looking at the topic from a practical, hands-on point of view. In real life, underexposed images can be saved through proper processing much easier and more often than overexposed ones.
The critical point is, 'as close to blowing out without actually doing so.' If you go just a quarter of an f-stop above the blow-out limit then the image is damaged beyond salvation. If you stay two or even three full f-stops below the limit then the image usually will still be fine.
[{POST_SNAPBACK}][/a]
Shooting fully to the right requires base ISO so that the electron wells of the sensor are filled, collecting the maximum number of photons and giving the best signal to noise ratios at all levels. If you are shooting at an ISO higher than base, you are essentially underexposing and boosting the amplification of the signal. For example, if the base ISO of your camera is 100 and you expose at ISO 1600 you are basically underexposing by 4 f/stops and compensating for this by increasing the amplification of the signal. This is usually done by setting the camera to a higher ISO, but alternatively, one can do this in the raw converter. Because read noise is higher at low camera ISOs it is better to use the ISO route, at least up to [a href=\"http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/]Unity Gain[/url]. There is little point in increasing ISO above unity gain, since you lose headroom protection from overexposure but collect little if any more information.
Digital cameras have no shoulder on the characteristic curve, and clip abruptly at sensor saturation of overflow of the ADC. You can recover 0.4 to 1 stop of blown highlights with highlight recovery, but beyond that there is a complete loss of highlight detail. Furthermore, highlight recovery is less than perfect and may involve color shifts.
On the other hand, a camera with large pixels and low read noise (e.g. Nikon D3 or Canon 1DMIII) can produce quite acceptable pictures at 4 stops underexposure as defined above. Overexpose by 4 stops and you will lose 3 or more stops of highlight detail and the results will most likely not be acceptable. However, you will get good shadow detail.
As Andrew stated, we should be striving for proper exposure. In an imperfect world, slight overexposure can be tolerated, but gross overexposure will cause major data loss. If you have a good camera, gross underexposure will give more noise and less dynamic range, but you will at least have an image. Since signal:noise is proportional to the square root of the number of electrons, doubling of exposure will improve the S:N by a factor of 1.4, not 2, and this improvement may not be worth the risk of data loss if ETTR is carried to far.
Some argue that a main rationale of ETTR is to make use of the 2048 levels in the brightest f/stop of a 12 but linear file, which contains a total of 4096 levels. However, the eye can distinguish 70 levels per f/stop at maximum, so one can afford to lose a few of these levels without incurring posterization. Posterization in the shadows could occur, but usually noise is the limiting factor there.