4 ms·

The part that always looked strange to me of comparing AI to humans/children is that humans have had years of time to learn while e.g. neural networks a few day

The part that always looked strange to me of comparing AI to humans/children is that humans have had years of time to learn while e.g. neural networks a few days at most?

pfista 11y ago
Exactly. Our brains process information constantly, even while we sleep. And they also process multiple senses - not just pixels in an image, but sound, sight, touch, etc. I think it's a little early to assume that consciousness can't arise from complexity because we are far from representing that complexity with any technology today.
pmelendez 11y ago
True, but probably humans are more power-efficient although the math to compare this is not trivial.
doctorpangloss 11y ago
Yeah, but it's not obvious if every single frame of the whole movie children have experienced is necessarily important. To put it one way, what is the _rate_ of learning? How many _concepts_ do children learn per unit time? Concepts: Our computer systems take whole images and train on them, but before a pixel-based image gets to the learning part of a child's brain, it has probably been conceptualized in some way already. Rates: And then, from this feature vector—maybe 1 feature vector per 10 seconds, for a child—how often is it incorporated in a learned concept? Maybe one concept per hour? Per day? Think about how challenging it is for children to learn vocabulary. A young child doesn't know that much. The important thing is that a child is a self-learning machine. It chooses what's important and has a way to shift its attention, a critical enhancement of new (2015) AI approaches, deep learning and otherwise. It can explore its own world and somehow choose the "training instances" that matter. By comparison, a neural network training for days looks at millions of images. A very high rate. But the most effective approaches (the massively-multilayer networks, the deep in deep learning) are making concrete what a child's brain must do already: separate perceptual components from learning components, in a conclusively more structured and hierarchical way.
- spdionis 11y ago
  What if you can achieve such high learning rate only because (in ML) you process a really narrow view of the data which cannot lead even close to how a human perceives things? I mean in terms of conceptualizing and connecting things/thoughts.

The part that always looked strange to me of comparing AI to humans/children is that humans have had years of time to learn while e.g. neural networks a few days at most?

pfista 11y ago

Exactly. Our brains process information constantly, even while we sleep. And they also process multiple senses - not just pixels in an image, but sound, sight, touch, etc. I think it's a little early to assume that consciousness can't arise from complexity because we are far from representing that complexity with any technology today.

pmelendez 11y ago

True, but probably humans are more power-efficient although the math to compare this is not trivial.

doctorpangloss 11y ago

Yeah, but it's not obvious if every single frame of the whole movie children have experienced is necessarily important. To put it one way, what is the _rate_ of learning? How many _concepts_ do children learn per unit time? Concepts: Our computer systems take whole images and train on them, but before a pixel-based image gets to the learning part of a child's brain, it has probably been conceptualized in some way already. Rates: And then, from this feature vector—maybe 1 feature vector per 10 seconds, for a child—how often is it incorporated in a learned concept? Maybe one concept per hour? Per day? Think about how challenging it is for children to learn vocabulary. A young child doesn't know that much. The important thing is that a child is a self-learning machine. It chooses what's important and has a way to shift its attention, a critical enhancement of new (2015) AI approaches, deep learning and otherwise. It can explore its own world and somehow choose the "training instances" that matter. By comparison, a neural network training for days looks at millions of images. A very high rate. But the most effective approaches (the massively-multilayer networks, the deep in deep learning) are making concrete what a child's brain must do already: separate perceptual components from learning components, in a conclusively more structured and hierarchical way.

spdionis 11y ago
What if you can achieve such high learning rate only because (in ML) you process a really narrow view of the data which cannot lead even close to how a human perceives things? I mean in terms of conceptualizing and connecting things/thoughts.