On Mon, Oct 28, 2019 at 11:11 AM <[email protected]> wrote:
> Do you mean, instead of feeding the net data and learning, to instead > request new output data/solutions? > You could put it like that. Without seeing an exact formalization it is hard to say. You make the example of zebra, horse, dog, mouse, cat. You group them heterarchically based on sets of shared contexts. (Your innovation seems to be a more efficient representation for the shared contexts??) That's OK. But perhaps I can distinguish myself by saying what I do is not limited to groupings. I don't only group words heterarchically based on sets of shared contexts. I use the shared contexts to chain words in different ways. Saying to look at the way these things chain, might capture what is different in what I'm suggesting. Because the groupings are a heterarchy, the patterns possible when you chain them expand much faster. Perhaps something like the way the number of representations possible with qbits expands, because each element can be multiple, so their combination can be exponentially more multiple etc. Traditionally language has been looked upon as a hierarchy, so we miss this complexity. That has been the historical failure of linguistics. Deep learning also looks for hierarchy. There is the potential for heterarchy in their representations, but as soon as they try to "learn" structure, that crystallizes just one of them, and the heterarchy is gone or at least reduced. Such crystallization of a single hierarchy gives us the "deep" part of deep learning, and it is also the failure of deep learning. So, I guess I'm saying, yeah, heterarchy, but chain as a heterarchy too. Which means abandoning the singular, "learned", hierarchies of deep learning. -Rob ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Ta664aad057469d5c-Mfc729479fb75cb9e16f09856 Delivery options: https://agi.topicbox.com/groups/agi/subscription
