Researchers show for the first time that part of the efficiency of the human brain is based on the omission of predictable details
The fact that neither software speech recognition nor artificial senses for robots have yet worked satisfactorily has a lot to do with understanding. The system, which is supposed to pick out the decisive fragments in the right order from the multitude of stimuli flowing into it – whether they are snippets of speech from different speakers or visual impressions of moving and stationary objects – is faced with an almost impossible task. Unless it understands what it is doing, knows what kind of data it is dealing with and knows the laws behind it.
Automatic translations were easier when the translators were knowledgeable in the subject matter; a domestic robot could distinguish commands more confidently if it knew that humans give certain commands in a strongly gender-dependent manner, for example. The mp3 format shows that the knowledge of the basic laws makes more efficiency possible: music files encoded with it are several times smaller than the original, and nevertheless as a rule no difference is to be heard. This is due to the fact that the algorithm takes into account the perceptual abilities of the listeners – which on the one hand concerns the audible frequencies, but on the other hand also takes into account masking effects: in the silence of a library, even a leaf floating to the floor can be perceived as loud, which in the noise of a school class nobody would pay attention to.
Human perception apparently also uses similar tricks. Researchers had already suspected this, but so far no conclusive proof had been found. Scientists from the university of wisconsin have now succeeded in doing so; they report on it in the publications of the us academy of sciences (pnas). Although it is amed that such mechanisms apply to all areas of sensory perception, the researchers have now demonstrated the effect for horsemanship.
To do this, they first constructed special sounds in which the two parameters of rise/decline of the loudness hull curve and the spectral curve were strictly correlated with each other. The subjects were exposed to these sounds for seven minutes.
Afterwards, the study participants had to distinguish between three short sound snippets – an easy task without the prior sprinkling. But now the subjects only succeeded, if the snippets of the correlation were enough, which they had registered unconsciously beforehand. If one piece did not respect this artificially introduced lawlessness, the listeners were at a loss.
Obviously, their brains had already in a very short time become absorbed in the internal rules of the background music and then, trusting in the regularity of the rules, ignored all that did not fit into them. If the subjects were then exposed to non-correlated sounds for some time, they regained their former discriminatory ability. The experiment is somewhat reminiscent of the baffling experiment in which viewers do not notice gorillas running through the picture when they are engaged in an observation task – but this is about so-called inattentional blindness, a different phenomenon of perception.
Is the effect a loss? More likely an advantage, says keith kluender, one of the researchers:
If you press your ear against the wall, you understand every word of a lively discussion next door, even though the wall filters two-thirds of the acoustic information – your brain reconstructs the rest.
Is that why robot researchers and programmers should be ashamed of their work?? Hardly – in the course of his evolution man had almost infinite time to get used to the laws of nature. Compared to this, machine vision and hearing are already very advanced today.