3 ms·

Is artificial intelligence going off the rails?

• AndrewKemendo 9y ago
  >Everybody knows that my brain does not operate by having trillions of examples. This is where I disagree fundamentally with others in the field - though Trillions is also an exaggeration. In my opinion this needs a lot more study, but if you look at just the amount of visual data that we process per hour it's staggering. There is no coherent comparison between our human visual systems and digital system in terms of image processing, but we can use some rough equivalents as a starting point. MIT researchers have shown that humans can process images that have an exposure time of about 13ms [1]. Now, that is "recognition" of something that has already been seen, but it does show that it is the minimum time to discriminate visual features for the optical/visual system to "close the loop." If you assume that the 13ms per process "sampling rate" at 60Hz (where most people start to see screen flicker) then you could very roughly translate that into image processing at around 276,000 "images" per hour. Roughly extrapolated you are looking at 1.4B "samples" per year. I would imagine that a system with millions of years of evolution can do a lot to make that visual pipeline REALLY efficient, even if you keep the same basic learning structure. That's especially in comparison to our crude tools. I mean some of our best DL training sets only have 60-80,000 images. Again, these comparisons are not rigorous, but illustrate my point that it's much more data than I think Hinton et al give credit to. A lot more needs to be taken into consideration for this to be really studied. Like attention in the visual field, how much cognitive power is put into peripheral vs focal point etc... [1]http://news.mit.edu/2014/in-the-blink-of-an-eye-0116 http://news.mit.edu/2014/in-the-blink-of-an-eye-0116