The Thousand Brains Theory of Intelligence (2019)
numenta.com
numenta.com
I had this epiphany recently while meditating. While trying to make my brain be quiet, I realized that there were different parts that were being quite vociferous and just had a lot to say at that particular moment, a la the "inner voice" phenomena. In this one case I just let them chatter instead of whacking them into submission. And it was then that I realized that consciousness was probably a fairly large collective, like a jellyfish, rather than a single thread.
My innate pattern matching is probably above average. Functionally, something always speaks up and says, "I think I know what that is," with the probable answer, in its estimation at least. I may decide to double check the answer and send it back, soliciting more feedback. But the review process is mostly the same.
I postulate that one of my "brains", "columns", or threads is speaking over the tumult with the possible answer. And the larger your collective of better-trained "brains" in this allegory, the better one will perform at these kinds of exercises.
In the example given, trying to identify a cup based on limited tactile and visual information, there absolutely would be crosstalk while a consensus is formed within probabilistic certitudes.
This is all very much related to the discussion in the past few years about "inner voice", how some have it and some don't, etc. The larger your particular collective, the more chatty your inner voice probably is.
These are just my thoughts after empirically observing my brain after developing fairly good control over it over the past few decades.
People often experience something like this while driving, when you just get an instinct that a kid is going to jump out, or that other car isn't going to stop, without having consciously paid attention to it. You slam on the brakes and are left with that feeling that you just KNEW that was going to happen. A part of you did know it, and foregrounded that prediction when enough variables lined up, overriding whatever you were previously thinking about.
My current role is that of a principal engineer or quasi-architect and I do exactly this when I need to make a decision regarding which direction we're going to go with something.
An environment that is conducive to your subconscious performing well really helps.
In my younger days, I'd run for 30mins outside often times without any music. During those runs, I'd let my thoughts bubble up. It was a welcome chance to let my brain tell me what was bothering it. Sometimes these would be technical problems du jour, but after handling those, nearly always there would be a social problem, usually interpersonal, that would bubble up. It was a very good technique to identify those issues organically.
Something about the kinematic state also made it easier to accept, process, and set aside the idea than it would be if I was sitting in a room silently.
Nowadays, any time I'm doing anything involving semi-repetitious kinematic activity, I find myself making a choice between listening to something, music or a podcast, or simply nothing and using the time to sort through what ails me. There is a certain mental healing that comes with certain physical activities.
Yep, I'm a lap swimmer. It's amazing what your mind does in that state.
Can you describe what you've found are key traits of that type of environment?
I have a very peaceful home. It's quiet, safe, a little remote, about an hour or two northwest of Austin in the Hill Country. I have a library in my house where I do most of my work, and sometimes I get on my motorcycle and ride for a bit into the middle of nowhere, which becomes meditative, much like swimming, allowing my mind to work through problems on its own in a very beautiful environment.
I can work (whether it's actual coding or sitting quietly in thought) for 12-14 hours a day, max. Then I must stop and get 7-8 hours of sleep. That balance of good sleep and hard work is key to maximizing productivity and the ability to solve problems during available work hours.
A healthy diet free from high-glycemic foods and foods with a high inflammation index is also very important.
I wonder if our "conscious reasoning" is really just attending to a column that does that? Certainly, formal reasoning is a skill that is learnt and practiced - perhaps not so different from a manual skill. We can also suppose pattern-matching on this, to get intuition and hunches. All old ideas, but perhaps neuroanatomy is advancing sufficiently to test them?
Also: Minsky's Society of Mind
"I" am a story I'm telling myself about about a conversation between different drives in my head, and I am free to dis-invite those drives from the conversation, or create new ones.
The framing of such drives, and the voices given to each, appears to have been a causal element in many players' own personal growth and development, particularly around addiction, but certainly not exclusively.
Given that it's chemical and electrical signals, probably in some semi-hierarchical pattern, isn't that how it "must" be?
Your brains are simulating yourself, accepting the majority vote and presenting it to the outside.
Imagine a captain who becomes unhinged. What could the crew do?
I imagine this could be a metaphor for mental illness – although lacking in nuance.
Sounds useful as a feedback cycle, so that we can imagine/play ourselves (and other persons) in many imaginary situations.
clearly, in your examples some entity is thinking (and even makes decisions) about your internal chatter and must therefore be more than just pure consciousness. Or is it just an additional layer of chatter?
Doesn't have to be either. A part can watch the whole (even itself perhaps). Like our head+eyes+brain can examine our whole body in a mirror (or just looking at our hands, legs, torso, and so on). If we had more movable eyes (like some animals) an eye could turn around and directly examine the other eye too...
This shit is weird.
https://www.theguardian.com/science/blog/2014/aug/21/science...
That's not an accurate interpretation. In my case, I think "fully formed thoughts in rapid succession, sometimes colliding and in competition with one another" is more accurate. However, sometimes it does resemble a conversation as you start to reason through scenarios using game theory.
Seeing your other comment, it isn't just seeing imagery in my mind's eye, although that's possible, it's the ability to quickly construct fully formed models as well as receive a rapid series of fully formed thoughts almost faster than I can do anything with them, which is where practice in managing that firehouse of information becomes useful.
So you don't experience any of this? The speed at which my brain works is probably faster than yours. The synapses supposedly fire faster. To me it's kind of a superpower that allows me to do amazing things for people, so I obviously have fun with it. But I have learned how to manage it and maximize the potential.
I think you're right. I've often found that I'm way slower than others when it comes to coming up with answers on the spot (except for mental math, where lack of visualizing seems to be a major advantage for some reason - I think I just don't get confused as easily). I'm really bad at conversation and not witty at all, but I often find that I get to better answers in less than time overall, but my time to first answer is a lot longer.
In some ways what I have is a disability. My nervous system can easily get overstimulated with too much information, which precipitates an adrenaline dump that shuts down my body. Speaking in front of crowds, looking at all those faces with all that information bombarding me, will do it. So it definitely has downsides which I know that I have to avoid.
I know what I sound like to myself. I know what other people sound like to me. I've heard probably millions of voices in my lifetime by now.
So I can basically imagine a conversation between myself and others or an entirely fictitious entity. And it's like a memory that never happened. (I mean, as much as any memory actually happened.)
Ghosts as portrayed in movies are actually a highly realistic view of the common man's imagination from the first person. That's what it looks like to them, so you know. They're usually people because that's what matters most is people spesh the face that's like who they are in a 2D projection, they're imagined because the real person is not there anymore, they're not colored in first because that's work for their limited imagination, plus it interferes with the "imaginary" marker, plus they generally don't realize they're not seeing things colored in, that's something the movie producer only realizes because he's trying to sell tickets. Also because in white societies the last image the living have of the dead are their paleness--in white societies primarily, so Europe and Asia which dominate cultural exports--so they are imagined as they were last seen. That will always be the latest version of their face.
.
Then there's what I call "macro-phantasia" which is like, like a video game with perfect graphics at this point. It is a gift and a curse. As though there were a difference between gifts and curses. Overboard.
.
Because in fact aphantasia prevents you from certain forms of harm, like imagining yourself in terrible places as people are wont to threaten you with. Threats of like "place worse than you can imagine" that's so you turn your imagination on to the max, and harm yourself in your own imagination to coerce yourself to their bidding with no proof it ever happened. If you can't imagine that, guess what, police interrogation is a waste of time even for the policeman on time-and-a-half, you didn't cooperate and that's it, or the gang-land interrogation equivalent, what side you're on is in this case irrelevant. It helps against coercion.
.
Aphantasia. Curse and a gift.
Of course that doesn't mean there aren't novel concepts in their book and research, but just a bit weird to pose the base "thousand brains" idea as novel.
However, this concept dates back at least to the 1980s with Marvin Minsky's "The Society of the Mind":
https://www.amazon.com/Society-Mind-Marvin-Minsky/dp/0671657...
The (again, many) models in the layer above receive this evolving consensus as the output from subsets of nodes in the layer below — potentially with overlap in which upper-layer models receive outputs of which lower-layer models; and each of these models themselves evolve to a consensus.
By analogy: DCNN is like representative democracy, where each layer is a representative with the layer below as its “constituents”, forming a belief based on the multitude of lower-layer beliefs it hears, and then acting on its own formed belief when reporting to the layer above. This “thousand brains” model is more like grouping constituents into juries and running them through the Delphi-model iterative consensus; and then having representatives who act on continuous-feed statistical aggregations of the Delphi-model iteration-step outputs of multiple juries.
Buddhism and Modern Psychology on Coursera covered some mind boggling studies that give strong evidence of this, and subsequently support for core secular Buddhist practices.
Happiness Hypothesis by Haidt discusses at a higher level the connection of modern psychology to stoicism, Buddhism, and the many brains theories.
And Waking Up by Sam Harris takes a slightly more incendiary take, but is interesting for its ultimate conclusion that there is no self, just a cacophony of selves running on autopilot.
I personally digested these and came to the conclusion that we are on the cusp of human like AI, but we just haven't lowered our expectations for human like consciousness enough to realize it.
I'd be very interested in your take on these, too.
Ah, that's my take as well! The one thing AI is missing, IMO, is the ability to train over its own output and model itself as a single simplified self, which would then act as a guide signal to get all these diverse micro-selves "on the same page."
Ie. self-consciousness.
As I understand the training approach, LaMDA isn't aware of its own output in a way that would allow it to introspect like that.
I know a lot of my ability to organize the world and characterize objects comes from the organizational structures refined and propagated by these institutions