Yes, shape, properties, temporal patterns of higher composition, AKA purpose. I don’t see how either, recursion is not easy to visualize. And there is no end to it, that cup may be an element in any number of higher-level patterns. All interesting properties have one thing in common: they are predictive or invariant. So, the only way to design a general learning algorithm (general for any type of inputs or ultimate use) it to define and then select for quantitative measure of invariance, in my terms match. And it must be scale-invariant: defined in the same way from the pixels up. Sorry to repeat myself, but I am answering repetitive questions.
1 Like
The three outcome I listed were all based on experience and needs and not from any intrinsic property of the cup.
The cortex is a learning machine, directed by the lizard brain. This need based direction chooses meaning from an almost infinite palate of possibilities based on the selfish needs of the lizard.
One of the concerns regarding AI is that it won’t share in human motivations - that it will be incomprehensibly alien in what it chooses to do. Possibly dangerously so. Need based choices are understandable and expected.
I suspect that you will need to add memory to your kernel and some sort of motivation to prevent it from obsessing on something useless like the best name for every blade of grass on earth. (Bill? George? … that one looks like a John!)
3 Likes
Memory is already there, the inputs are buffered in patterns, on all levels. Extrinsic motivation: supervision and reinforcement, will be added depending on use. I am working on core unsupervised alg, these add-ons (supervision is an analog to instincts and reinforcement is to conditioning) is a totally separate matter. Anyway, appreciate your interest.
3 Likes
Bitking, please explain why is the sensor fusion still a problem? (My idea about sensor fusion is like merging of vision and motor, or vision and audio. Am I right?)
1 Like
What part of “me” is the unitary and seamless combination of all your perceptions, internal and external, into your here and now?
Sure, there is merging but each sense codes in different formats and with radically different properties- how exactly do the come together in the temporal lobe?
My best guess is that sight and sound get processed by the WHAT and WHERE stream into a common format. The spatial processing relates that to the posture system to place the body sensors, all in relation to the vestibular system.
2 Likes
Thanks 
1 Like
There was only the old brain first, then a lot of its functions, for the most part, were transferred to the cortex (with the great improvements!).
I don’t see why this process can’t be completed. Imagine a cortex-like intelligence with a very thin “body” - just an API to any external tools. The rest of the feelings have moved to their abstract (and painless) version of themselves in the cortical-like (so, much richer) representation to support action-driving at all levels.
Sounds not that bad, as for me 
3 Likes
I look forward to seeing your working code examples.
1 Like
I’m in love <3
Oh wait. I may get obsessed like that.
2 Likes