This piece on AI research and the MIT Intelligence Quest reminded me of some of the deliberations in Douglas Hofstadter slightly problematic but eminently thought provoking I Am A Strange Loop. It’s been a few years since I read the work, but it expounds on the self-referentiality inherent in consciousness. One of the more notable quotes:
In the end, we are self-perceiving, self-inventing, locked-in mirages that are little miracles of self-reference.
The question that the Intelligence Quest is partially examining is whether the current approach of Deep Learning can be made sophisticated enough to approach human thought (more generally the project is concerned the engineering mechanisms of human intelligence). As pointed out, the current Deep Learning methodology requires far more data than comparable human learning, just to yield results that are often opaque and extremely narrow. One has to look no further than Janelle Shane’s often humorous applications of deep learning to candy hearts, music genres, or paint color names.
This isn’t to say deep learning research can’t yield significant results. Rather it begs the question of what other mechanism sit between pattern recognition and intelligence that make humans able to learn with a much smaller data set.
Read more at: AI reaches a crossroads