What is numentas biggest limitation?

What’s the biggest obstacle that is slowing down numentas progress?

Is it funding? Lack of scientists who need to be trained? Lack of neuroscience research that is slow/hard?

Hope this is the right place to post

1 Like
  1. Progress … to what goal?
  2. Scientific breakthroughs don’t come on a set schedule.
3 Likes
  1. Sorry I could been more clear. I’m assuming the ultimate goal of numenta is to create AGI? That’s the goal I had in mind.
  2. I get that breakthroughs don’t come on a certain schedule, but is there any way to speed up the process?

Numenta existing at all has greatly increased the chances of the creation of neuroscience inspired AGI being developed. Is there some obvious bottleneck to this process? If someone gave numenta $1B dollars would that help a lot? Do we just need more scientists working on these problems? Is it an issue of marketing? Do we need to wait for technological breakthroughs in brain imaging? Do we need faster computers to be able to run bigger/better simulations?

If I had to pick any of those three, I’d imagine the last one plays a big part.

If there was more sophisticated equipment to instrument a live brain with, it would surely make it easier to understand what’s going on.

Hi all. Here is my answer to your question. I have a close view on this having been part of Numenta from the start.

  1. I know it can appear progress is slow. At times it is. This is not for lack of funding or computing but because we have to solve many difficult scientific problems.We have relied on neuroscience to guide our research and neuroscience is a complicated field with partial, and often conflicting, empirical data. Although it has taken a long time I am excited by the progress that we have made.
  2. Our understanding took a big leap forward with the elements of the Thousand Brains Theory. This gave us a broad roadmap to AGI. My book is probably the best summary of this progress. It covers what we knew up to about 18 months ago.
  3. We have made significant progress in the last six months filling in missing pieces. We have not published, or in any other way made public, most of this progress. I apologize for this. One reason is that we have been too busy and focused on our research. The other reason is that Numenta has transitioned over the last couple of years. Our team has grown and is now almost exclusively focused on ML and AGI efforts. We haven’t yet decided on how to best communicate our new work. We hope to resolve all of this in 2022. This year, you can expect to hear much more information from us about the technology we are building.

Again, the biggest obstacles to our progress are hard scientific and engineering problems. I believe we can, for the first time, see the end goal, and our progress will hopefully accelerate going forward. I hope you will be excited by what we are working on.

15 Likes

A"G"I has a G for General. Which I haven’t heard one bit in any Numenta talk.
Numenta wants to copy the cortex and subcorticals to accelerate machine learning. They explicitly say they will leave out IMPORTANT topics like qualia, emotions etc (which i don’t agree).
The thing that holds them back is the thing that holds anyone back: a theory for the brain. How do you get the best features from the brain without going full cargo-cult science? Put a symbol somewhere and abstract away the rest. But abstract away WHAT?

I personally appreciate the great progress made both by the field of neuroscience as well as by Numenta. I would dare to say (as my subjective opinion) that progress does not come in a linear continual process but rather as a series of quantum jumps. Some big ones and a few smaller jumps in between.

With each jump, we narrow the search space of possibilities that lead to AGI but also increase the focus. Within the new focal points, we can better identify some missing pieces, but can never predict when and how these pieces will be discovered. We know what we know. We also know some of what we don’t know. But we don’t know if there is more to what we don’t know. (The eternal dilemma of basic research).

I would be very interested in a high-level roadmap of what we expect to find on the path to AGI. (The expected milestones on our path to AGI.). Many of these milestones lie in the past, behind us. Numenta has been instrumental in many. But what do we expect to find ahead of us, on this path?

Finally, I would like to add, that in my opinion, this roadmap should be defined in terms of given functional capabilities and not in terms of areas of application. (Often times, goals and milestones are described as areas of application. That is not scientific in my opinion). Capabilities at a neurological, computing or cognitive set of dimensions would make more sense.

1 Like

28 posts were split to a new topic: AGI - what part does HTM/TBT play?