Numenta A Thousand Brains: A New Theory of Intelligence by Jeff Hawkins, 11 Aug 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: A Conversation on A Thousand Brains with Congressman Ro Khanna (Highlights)

Description

Innovations in AI are transforming economics and societies globally, leading to political debates on the ethics, fairness, and accountability of AI. In this episode highlight, Jeff and Donna talked to Congressman Ro Khanna from California’s 17th district about the key takeaways from A Thousand Brains and the political challenges in developing AGI.

To learn more about Congressman Ro Khanna’s work, visit https://khanna.house.gov/
- - -
The 'Conversations on A Thousand Brains' video series aims to extend the meaningful dialogues that the book, A Thousand Brains by Jeff Hawkins, has started. We invite leading intellectuals and experts in various fields to discuss the impacts and implications of Jeff's novel framework of intelligence - the Thousand Brains Theory.

Visit http://AThousandBrains.com for more information.

A

Ro thanks for being here with us, um I I know you read the book and I don't know if you have any just initial impressions or thoughts or things on your mind that that you're interested in um talking about or we have some topics we can engage with you on.

B

Oh well, it was a fascinating book, the uh biggest takeaway I had, which you could correct me if I'm I'm wrong, but is that perception of uh experiences is, is based on a model of predicting uh what we're going to uh experience as opposed to uh constructed directly from uh perception, and uh it's an interesting idea that intuitively makes sense that if you turn around and see my table, which is messy that uh I'm I'm constructing a model of that as opposed to inputting all the uh all the data points.

B

One, I guess is that a an accurate uh summary of the one of the key theses uh thesis of the book. uh I guess we could start there.

C

Yeah, I think so I mean uh that idea has been known by a few people for a long time. So that's not a completely new idea in the book, um but the idea that we have this uh constructed reality. The everything we see in the world is is really built on this model we have in the head, which leads to all kinds of interesting problems for humanity at times it's a useful thing too, but that is one of the key ideas uh um in the book.

C

Of course, what what the discoveries we made were, how that model was built and how it actually comes about and how it's structured in the brain, but from a from a point of view of a lay person, I think that is right. That's a great takeaway!

C

You know we have this constructed world model in our head and that's what we perceive, whether it's right or wrong.

A

Well, that that's one of the things that interested me is the question of what it's wrong in this question of false beliefs, which is something you know in the political world, you know whether it's uh who won an election or vaccine efficacy or whatever it is, and you know, did that resonate with you at all row on this question of false beliefs. It is, does it explain and does it help in any way in understanding how people form these uh strongly felt false beliefs.

B

Yeah and I guess the question there is: um where does philosophy come in and where does science come in right I mean uh so you know I I mean emmanuel khan's famous theory. Is that we're we're constrained just by the uh the perceptions of our of the human mind and can't know reality itself like an animal, could have a different perception. So I guess the question is when you're talking about false balsity and reality.

B

Where is that based on the human mind's perception and and how are you defining that, because, obviously, that's uh such a deep question of what is.

C

True yeah, I think I think it's, although I talk a lot about false beliefs in the book. I think I think what you're talking about rosa what's false is can vary over time or what we perceive as false converting time. What's reality can vary over time. um I make a point in the book talking about like flat earth right flat. Earth is a really good model of the earth. It works really well.

C

We could all use that there's nothing wrong with it. We now think it's incorrect right. um So there's what there's a situation where you have a belief structure, which is quite useful, and we now know it's incorrect and there's lots of things. We know about the world which um we think we know about the world which.

A

C

To be limited by our experiences, as you said right, we can't we can't know the full reality of the world. I think the uh the thing that I'm hoping is that people can just accept, or at least understand uh the science behind this like well. Why do we have different beliefs? Even if you and I don't agree on what's right- um we can at least understand how we came about to have different beliefs. You know, how is it that two people can look at the same facts and come up with different belief structures about them.

C

That would be a first. You know thing that if everyone understood that that would be useful, but I think this is a fundamental problem, especially when it comes to complex knowledge of things. We can't all directly experience. This is where this is where we have the most problems. If you and I experience something directly, we usually end up with the same model of it like you could touch it. I can touch it, you can see it. I can see it, you hear it.

C

I hear it, but uh the problem we run into is that in society we don't all hear and see the same, the same thing, and uh so that can really be the different types of bullets.

A

It strikes me that it happens all the time in things like economics, where you take a theory, and you say: okay, a tax cut is that you know help the economy or does that hurt the economy and that people end up with these very different frames of reference to interpret that that policy decision right, yeah and.

C

They may even look at the same results and they look at the results of that. We do a tax cut and 10 years later, which I didn't work and some of those people say no, it didn't work see why and some people say, oh, it did work. You see why. I think I think the answer wrote. We don't want to be saying. We know the right answer these things. I think what the question is. Is it's educating people about this process and it could be quite detailed.

C

The knowledge of how this works um is that would that lead to a more just, equitable or at least um friendly societies. So.

B

What would be the the policy recommendation if there are any in terms of the the the sort of interesting insights that you draw about, how we perceive things and why we may be perceiving things differently?.

C

Well, I don't you know, gosh, you know you're the policy person not me, but um I think I think uh you know mr mentioned in the call to action I mean I have this belief. I have this desire that everyone should at least understand how they believe.

C

You know what is the what is going on their head and why it is easy to have false beliefs or the contrary, beliefs, and so one policy thing is to make you know- and I'm not even sure we're ready for this yet, but it's to to uh to make edu just like we teach kids about dna right right, we teach kids about the solar system, we teach kids about uh evolution at least most kids. um You know we ought to be teaching kids about brain theory and about how beliefs are formed.

C

You could wrap it and pair it with the sort of philosophical ideas that you've been bringing up, but there's a real hard science to it too. Right right, there's a real, hard science. You know we can talk about the implications of evolution, but you know there's a science to dna and how it replicates and all that kind of stuff. So that's uh I mean that's, that's the my ultimate goal. People I've often asked me. Why do I do all this work? You know how's, it gonna impact society.

C

In addition to the work on ai, which we talked about a lot, I think probably the biggest thing I can say is that if everyone understood how they believe how they understand the world, if it wasn't just this.

B

What would be the few takeaways? You would want people to understand.

C

Well, the first one says: is he brought up? Is this this model that you perceive this model in the world right, that's the first one um that you actually don't directly pursue the world and, um and your model is a is, by definition, going to be a subset of the world. It's going to be limited to what you can perceive.

C

um It is not going to be a full reconstruction of the world um and it's the best of an approximation, but just understanding that when we even when we just look out the window and see a car, we are actually perceiving our model. There may.

B

A

C

But but it's your perception that another person on the part of the world might see something completely different. So um so you you this. This is the first thing it's like we under have this model and then and how the model can be false, and then you know, if I would, then you can drill into the details of how the model is learned, how it's structured in your brain.

C

um You know how how your your neurons actually do this stuff, um but uh and then this whole idea of how knowledge is stored in reference names is a big big part of the model. It's it's basically explains how we take information and structure it, so we can act upon it and two people can take the same information and put in different references and get two different belief. Structures out a those are very high level concepts. You know we're talking about something that no one's ever done this before.

C

I don't think spring theory is part of everyday curriculum. These things.

A

So I propose to turn the conversation a little bit to ai since we're gonna. um You know be limited on time and I'm just curious bro. One of the things that uh jeff addresses in the book is, you know, is there isn't there an existential threat for ai and.

B

A

Know you've been involved in thinking through regulatory issues and so on, and you know how did the book uh help you in any way and think about that? Or did it give you other questions on this whole uh topic of the threat of ai and did you believe my arguments.

B

You know yeah yeah. Well, I guess the question is um one of the things that we've been thinking about in in congress is that which, which seems to be your model theory would support, is that there are more uh efficient ways to uh construct ai than just massive data uh processing that the human mind is more complex and that uh we're at the very basic stage of understanding uh how ai works, because it shouldn't require massive uh data input. um Is that accurate, or would you say.

C

Well, that's that's only accurate, that's one of the uh differences between true intelligence and the today's ai. That's certainly uh the question is why you know then, so we try to address that in the book. Why does the brain does not require this massive data to learn things? But yes, it's a big big criteria.

B

The thesis of that the brain produce perceives things through modeling that our perceptions are imperfect, that you need embodiment in motion to to understand how the human mind comprehends. Those ideas are relatively uh consensus and then what you're saying that the unique insights on the physiology of the brain that.

C

We figured out what's what's unique, is is how this model in the written brain is distributed in these cortical columns, 150 000 of them that's novel and how information is stored in those models using reference frames. That's new um and those.

B

Those but the the the so there's consensus, though over there.

C

I wouldn't say: there's consensus, I if I took the world of neuroscience and ai and I said.

A

Remember the part I.

B

C

The part you said I said, would you agree with that? I would say maybe 20 of those people would agree with that.

B

Oh, even the modeling part is not consensus. Yeah.

C

There's a lot of people have no idea about this. There's a lot of neuroscientists that don't know this there's a lot.

A

Of ai people, who.

C

Don't know this, it's a very, very diverse field. I think 20 percent is a lot sure yeah. It's we're in this sort of pre-paradigm state where there are so many people have so many ideas.

C

um But if you look at like the bulk of neuroscience work and the bulk of ai work has nothing to do with the modern, the vast majority of them.

A

And and now you say.

C

Well, people say: does it make sense? A lot of people say yeah, that's right, I think that's right. Other people say I don't think it's that important. um So there's no consensus about this today. I think there will be consensus shortly. I mean over the coming few years, um but it's growing, um but it I wouldn't say it's consensus today, not everyone would come up with jess. We need to have embodiment. I'm.

A

Very conscious of the fact that roe has another call he has to get to so um I I want to let you go. If, if you need to go row, I mean we could go on on this stuff. First,.

B

Well, this hasn't been fascinating. I I appreciated this because I you know it's a hard. It's a it's a very well written book, but it's obviously hard concepts to to to grasp right away, and uh I the more that uh people learn about it. I think it's it's helpful and I to me one of the takeaways is, I think, just a humility about how uh imperfect our perceptions are. It was one of the things that I took away from things and how much we still have to learn about ai, how far away we are.