I wasn’t ready to record a spatial pooling episode of HTM School this week, so this is what I came up with instead.
My simple definition of intelligence: the ability to acquire knowledge in an attempt to achieve a wide range of goals.
I definitely agree with your extremely long-term vision on machine intelligence. As you say, such technology will be our legacy as well as have profound benefits for humanity within our lifetime.
Finally, I needed a new audiobook so thanks for introducing Kevin Kelly’s book. Looking forward to the spatial pooling video!
Prediction …
Thank you for tackling some of the philosophical considerations on the subject of intelligence. I wrote a blog post here upon watching your video.
Many philosophers–following the observation of their own conscious operations–have discussed the centrality of images for human intelligence. Not only do we express what we understand from sensible data in words, but we also convert words into images in order to concretely understand abstract concepts. The former seems to be intuitively grasped in the AI world (e.g. image recognition), but the latter is not even universally accepted by philosophers (some speak of a “mind’s eye” that looks for the nexus between concepts, for instance).
Has the brain science literature addressed this? Does the neocortical region that processes language first associate inputs with other related words, activate predictive contexts, then move to imaginative representations of the world?
Are NuPIC algorithms sufficiently capturing the third stage of this human process? If not, is the new sensorimotor theory (with its overlapping columns) a breakthrough in more ways than one? That is, are the ‘free images’ of the human mind made possible by combinations of partial representations of the world?
Certainly not in the state they are in. But remember that HTM is foundational, especially the spatial pooling and sequence memory algorithms. Mechanisms that implement things like attention and creativity could someday be built upon them.