ftlm

thought-dump

Table of Contents

An ongoing thought feed dump fest.

Childhood

Childhood should be seen as a biological adaption in its own right.

  1. Maybe childhood is not only a maximized march towards growing every increasing capability.
  2. There must be a lot of adaption in the Child <-> adult relationship, too.

There is obvious use for a phase of relative weak power of building a very powerful system. (See vehicle 17 of my vehicles, future). This way the system can try out random aspects of it's capabilities without hurting anyone.

It is not that a child is lacking some ability because its system is inadequate. The indaquacy might be useful to the system, too.

Consider how children don't have big goals but do what their adults say. Its useful for both adult and child if the frontal cortex of the adult is allowed to carry both.

There might be push and pull balances, too. Where some aspect of development is being held back for a while by some other part of the system, etc. The story of development might not only be a grandious forward march of increading capability, where each aspect is growing as fast as it can.

It is scary indeed to ponder: What if the most interesting kinds of intelligences need a childhood to grow? What if we make machine intelligences and categorically exclude some important aspect of real world embedded intelligences with souls?

This is the reason Ben Goertzel is talking about making a baby AI play ground or something, with the idea that some AIs might need time to grow their minds.

Substances, building materials, mass nouns

For a mikrobiologist, bacteria are entities with workings and mechanisms.

Consider soudough. For the baker, bacteria are a kind of substance, a building material. She knows how to grow it, how to shape it. Its a multidimensional dough, manipulated and coaxed into doing what the baker desires.

For the mikrobiologist, bacteria is a count noun. For the baker, bacteria is a mass noun.

So it is with neuronal nets and other substances that make mind and intelligence. To the AI engineer, the nets are a substance to be shaped. And I think so too for the mind, somewhere somehow sometimes I see a knowledge substance, or compute substance, or capability substance, being shaped, manipulated and coaxed into something.

It is only natural to consider this neuronal substance a mass noun entity, it is stuff.

The shapes of thought

The only shape of thought available to LLMs right now is left to right.

It strikes me much more natural to think thoughts inside out, not left to right. I wonder if some thought is represented well as trees, some thought as sequential step, some thought as heterogeneous association web, and so forth.

(step-2  <--------------- pretend I can think of step-2, as if step-1 is already figured out

 (step-1 ...))
          ^
          |
          |
          |
          |
          +------- here I am thinking about step-1 in a more local manner


That is thinking inside another thought.

Same thing, as a tree:



                 +--------------------+
                 |                    |
                 |                    |
                 |     thought        |
                 +--+--------------+--+
                    |              |
                    |              | child (association)
                    |              |
            +-------+              |-----+
            |                            |
+-----------+---+                 +------+-------+
|               |                 |              |
|sub thought 1  |                 | sub thought 2|
+---------------+                 +-+--------+---+
                                    |        |
                                    |        +------+
                                    |               |
                                            +-------+--+
                                            |          | <------------------ Here I am thinking of a more local leaf
                                            |   ....   |
                                            +----------+

This is good for structured reasoning, …

Association web:


                    concept
+-----------+          |               +---------------+
|           |----------+-----+- - - -- |               |
+-----------+                |         |               |
 concept                     |         +---------------+
                             |           concept (more loosely associated)
                             | association
                             +
                    +--------------+
                    |              |concept
                    +--------------+

This is good for representing analogies/concepts, creativity, and finding essences between seemingly distant things.

Sequential:

+-----------+  t1  +-----------+  t2 +-------------+
| s1        |----->| s2        |---->| s3          |
+-----------+      +-----------+     +-------------+


  source                            destination
+----------+  transformation     +-----------+
|          |-------------------->|           |               (this is a Minsky transframe)
+----------+                     +-----------+

This is good for storylines, representing causality, transformations, and state transitions.

Interfaces:

I don't know and I don't want to know - Rich Hickey

mind                 motor competencies    muscles
                        +
move commands           |---------- movement
     --------------->   |
                        |
                        |
                        |
                        |
                     interface

The particulars of `movement` are abstracted away in the motor competencies interface between mind muscles. Thankfully, I don't need to know about which neurons to fire to move. I only need to issue move commands on my user interface to the motor competencies interface of the mind.

Another name for a related concept is `abstraction barrier`. This is a useful thing, to have a part of the system not be concerned with the details of another part. It lends itself to elegant design if the abstraction barriers are chosen well.

This is why I draw an interface as a line. It divides the system into the user of the interface and the provider of the interface. An interface is defined by the operation it supports.

The same concept, with different emphasis, is called a `map`. This we find in the perception maps of neurophysiology, like the "body schema".

In computing, we have memory-mapped files, which are virtual memory, correlated byte-for-byte with some portion of a file or file-like resource.1

Now you can pretend you have a program memory, not a file. This is a lever over having effects and observing files because manipulating memory is easier than file streams.

Other shapes / other perspectives on the same underlying shapes:

  • Recursion
  • Iteration

Measure problem

https://en.wikipedia.org/wiki/Measure_problem_(cosmology)

Indeed, sounds like one of the things you run into when thinking about the multiverse. How it is more likely to be a Boltzman brain in every given "instant". Maybe there is nothing wrong with this, I mean how else is it supposed to work?

What is the opposite of opposite?

Is it just me or is this a question that comes up "a lot" as you are a child?

I feel like it's the kind of thing we would have asked, and for some reason, I feel a mental slap from the mental ghosts of some adult saying "Just do your homework". Painful. See What I was not told.

Stimulated by Hofstadters Essences and Surfaces (books). I am thinking of concept space and moves in concept space - analogies.

Words:

+-----+ analogy +--+
|     |-------->|  +
|     |         |  |
+-----+         +--+
 source        destination

I am pretending that analagy is a category that contains the concept of opposite.

So I go and ask my friends and my AIs:

  • The same, Synomym, but not everything has a good opposite, Benni.
  • The opposite of opposite is to not divide, but unity.
  • The opposite of opposite puts you back to where you started. It is identity.

We realized that opposite is only a valid move in analogy space when you actually have something more like a pair. Black and white, round and pointy, heat and cold…

My first answer was even more the same. Like a deeper move into the essence of the source.

But now I have this in elegance with the answer (same with identity):

           target (deeper essence)
          +------+
          |      |
   -->    +------+       +---+
+--+                  <--+   |
|  |     ^               +---+
+--+     +---+
         |   |
         +---+
  cluster of ideas
  <source>

This supposed inward move in idea space doesn't have a direction but the direction between source and destination of opposite is clear, even the defining aspect.

This is why I think now of exacerbation, find the direction in which your source is salient, and move further in that direction.

        opposite of opposite      opposite
+-----------+      +--------+         +-----+
|           |<-----+        |-------> |     |
+-----------+      +--------+         +-----+
  target1           source             target

Exacerbation as the opposite of opposite in idea space.

I mean you cannot deny there is now a certain symmetry to this.

The opposite of running is walking slowly, the opposite of the opposite of running is sprinting.

Unity

The opposite of heat is cold. The opposite of the opposite of heat is Boltzman thermodynamics or something. The kinetic energy of particles doesn't have an opposite, but it unifies the concept of heat and cold.

The same thing happens with black and white and the amount of light. Where black maps 1:1 on absolute zero.

Well, sounds deep at least.

Percy Bysshe Shelley

It has been the persuasion of an immense majority of human beings that sensibility and thought [as distinguished from matter] are, in their own nature, less susceptible of division and decay, and that, when the body is resolved into its elements, the principle which animated it will remain perpetual and unchanged. However, it is probable that what we call thought is not an actual being, but no more than the relation between certain parts of that infinitely varied mass, of which the rest of the universe is composed, and which ceases to exist as soon as those parts change their position with respect to each other.

Damn, this Shelley dude had Dennet's materialist perspective in the 18th century.

Wikipedia

-—

He was vegetarian and atheist, too?

This made me reflect on the phrase to be ahead of ones time. And I decided this is getting the concepts upside down.

It is not that some people are ahead of their time, it is that common sense people were always there and the problem is a memetic one, the bad drowns out the good.

This is why Socrates is still relevant because he simply made common sense about the world and humans. And the underlying world stays the same across time.

It is really strange when people get this wrong. And you talk to somebody from the general population about AI and computers and they think that only the last 2 months of development and thinking are relevant to building minds.

It is a strange kind of hybris, to think that nobody from earlier times in history could have gleaned the true shape of the world.

-—

Seriously, I cried a little learning about this person being sensible about the nature of thought and then being vegetarian. To be connected to somebody through time and space, to know the same spirit lives inside us.

Maybe Shelley once looked up at the stars and wondered, who else in all of human history is looking up and will look up at the stars with this conviction in their hearts? That common sense, love, harmony and ethical truths make us converge on the same ideas, binding us at the deepest level. The level of the soul of the universe.

See The Civ We Want To Be.

Multiverse Theory

There is a profound symmetry in nature and our explanations of it.

Give me everything possible.

I cannot think of a simpler algorithm than this. And from this, the whole of reality, all its content, comes (emerges you might say).

The fact that the simplest algorithm is making the most amount of stuff. It is like a perfect symmetry with simplicity and explanatory power and the amount of stuff explained.

The dial of complexity is turned all the way down and the dial of amount of content is turned all the way up.

The algorithm of the multiverse theory doesn't speak of time and laws of physics, the way the algorithm of natural selection doesn't speak of organisms.

I call algorithm and what it produces form and content elsewhere. This to me points to the same essence of what emergence is about a useful meaning of the word.

So too the form can be the computer program and the content it's dynamic - well content. When we build minds, the minds will be the content of our program.

See

Data-oriented design

Consider this:

               environment
                |      ^
                |      |
                |      |
+------------+  v      |
|            |---------+-
| machine    |
+------------+

The imperative way is to make a machine, which modifies and reads from its environment.

A data-oriented trick is to reify, to make explicit, the what to do in data. A recipe for the machine. Now my machine can be superbly simpler, for it only needs to be a small layer of carrying out instructions.

+-------------+            +-------------+                 +-------------+
|             | analyse    |             |  parameter      | (simpler)   |
| environment |----------->| transformation <--------------| machine     |--> environement
+-------------+            +-----+-------+                 +-------------+
                                 |
                                 v
                              freedom

Transformation is a recipe, a plan, or an intermediate result, expressed as data.

Data-oriented design makes it first class to speak of my plans, revise them, log them out or share them with others. Or to modify them, mirroring the dynamic power of Lisp. (Or IPL, really see History of Lisp).

Analyse is a pure function, a calculation. I can develop with some example data and know I am getting it right. I can write property-based tests, I can fabricate input and output etc.

This indirection can make code with a step function simpler to understand, there is the magic of abstraction itself in this I think. Because the one crazy magical superpower thing in computer science is how abstraction is power. So when I see something powerful, I wonder if it is abstraction again, seen from a different angle.

This is a big deal. Code, which would be utterly byzantine in imperative style, can be made to look self-evident with this.

The key to understanding Lisp is realizing that it's simply layers upon layers of abstraction. - Hal Abelson

See:

When data-orientedness is in your blood

I just realized this looking at the benchmark output in a Python project context of a teammate. He prints json which is my intuition, too. Because now I copy the output and read it.

But what is this?

``` "foo": "1.32(20x)" ```

Why the hell would you glom the seconds together with the call amount in a single string value?

This is when I noticed how it feels to live and breathe data-oriented programming. I simply cannot put my mind into a state where it is a good idea to glom together 2 pieces of data instead of keeping them readable as the numbers they are.

At Clojure's heart, there is a deep underlying assumption that data will be at your side forever to conquer all problems.

All the edges in need for layers of cakes

I had the idea, that you take user input and plugin an interpretation layer, using an LLM or similar, between the user and the application.

Now the program sort of guesses from the kinds of user effects the user wants to have.

Not only conversation but inputs like button presses, mouse movements, touches etc.

I feel like this is a doorway into a world of tremendously more intuitive interfaces…

Welcome to the age of generative AI programs

Once we start putting the generative AI as part of the functioning, not merely the content, of our programs;

Welcome to a strange new world, where the docstrings are more important than the code. Where code is written to be read by computers, and only incidentally by humans.

To know the power of computers, think of the internet

from my audio diaries

The general population is aghast at the idea that brains are universal computers. They say things like this is too reductionistic… For some reason they have Hollywood ideas of this, the cold scientist and computer numbers persons, thinking things like humans are predictable

Why does the general population not have the colorful beautiful infinitely malleable, sci-fy ethos perfused, open-minded, world-embracing mind expansion in mind when they think about computers?

The way that I do?

Consider the internet. The internet is so good it is a force of nature. Not the content, which is commercial bullshit. The architecture. The ideas, the implementation.

I think they just forget it.

Fistness has appeared

Simplicity is the ultimate sophistication - Leonardo da Vinci

let x = Consciosness, Awareness, Qualia, whatever

How does x arise from the neurons?.

Urm…,

How does acceleration arise from the car engine?

Bit of a strange way to put the question, isn't it? You are putting in mystery from the outset. What a strange limitation.

How does the car engine work? seems to me like the more straightforward question to ask. With good answers.

So we simply ask: How does the brain work? and are right to expect much fruit from this inquiry.

Programming is about how to think. One insight from programming is to keep things simple. The world is complicated enough, You will not succeed if your ideas are accidentally complex on top of it.

It is the thinker who can keep things on the grounds of common sense that is impressive.

See Marvin Minsky - What is the Mind-Body Problem?

Immutability, Calculations and multiverse theory

I am only a humble biologist and programmer, so take it with the context.

Is it only me or is there something profound about the physics of how time does not exist?

The notion of the block universe is to say look at reality as a 4-dimensional block. Time does not exist. It comes from the arrangement of the universe's snapshots.

Multiverse theory goes further - Reality is not one universe made from a 4-dimensional block. It is all possible universe snapshots, like a mirror made from ice exploding into so many snowflakes.

Why do I feel time going forward? Ask why I think* I feel time goes forward. There is a logical succession of universe snapshots. Like a red trail going through the multiverse.

To go a trail that is not random noise, but organized enough to have a machine that thinks, I need to take a trail that sort of stays true to certain regularities. So the trials on which you will have life forms, brains and computers will adhere to what we call physical laws.

From our perspective, it does not matter if things get slightly weird on the microscopic level. If some quantum go back and forth in time, or tunnel around etc. It does not affect us too much. For our functioning as organisms, we only need determinism and stuff like time going in 1 direction on a relatively macroscopic level. Our physics, our red band, is allowed to be slightly thicker than a narrow line.

I am not 100% sure if I interpret this right at this moment in time but I think this means that time and physical law exist on the same plane. - They sort of do not exist at all the first time you look.

Darwin's dangerous idea is that there is design without a designer. Design comes from the logic of what replicators are. There are more good replicators around than bad ones.

Everetts's dangerous idea is to have physical laws without a physics engine. It would simply come from the logic of all possible universes existing. Some universes are a logical continuation of each other.

Time and physics go the same way as the designer. They are replaced by a more fundamental logic. Is this not satisfying? Anything can only be made out of simpler things. Whatever is can only be explained in terms of something simpler. This is another reason why multiverse theory has the rhythm of being true to me.

Another thing coming of this is that there would potentially exist alternative timelines with alternative physics.

After all, you can draw arbitrary red lines through the multiverse. The vast amount of them will be random, static noise - brushing more substantial realities like ours in places - but still static noice.

I am not sure if that works in our multiverse but if I think about making a game of life multiverse, the interesting part would be to explore possible physics. The rules of a game of life board do not come from the board. They come from how you decide to draw a red band of time through the multiverse.

I don't feel the earth move but it does. And I don't feel the universe split into eternal, timeless snapshots.

Either way, I think maybe there is something profound about immutability in Clojure and functional programming. It allows us to model reality in a simpler way.

We are using the same model as the multiverse: eternal, timeless snapshots of something that cannot change. Time is ours to create and destroy. By arranging the snapshots as we please.

Sun Aug 20 12:56:02 PM CEST 2023

I could eat a bucket of amino acids and puke a better brain architecture than that.

from Less Wrong.

Believing falsely is knowing nothing

Apparently, Wittgenstein said "believing falsely" is incoherent. He believed that one has to accept a concept as true in order to believe it.

What a strange limitation.

For the scientist, I believe just means I believe falsely.

For we will always go closer to the truth but never know it. This is the meaning of Socrates I know nothing.

Programming and clouds of thought

To think the same thoughts as somebody else (or you from the past or future) is part of the art of programming. To make somebody else think the same kinds of thoughts as you do. Programming is the ultimate high art of communication. It includes having a literal and cultural feel for what somebody else might expect.

We chose at times the obvious, rather than the clever.

The effectiveness of languages

One crazy thing about this. I don't think it would have to be that way. Is that shared, public, interpersonal languages are better than private ones. Nobody comes up with their private programming language and programs better in it. My claim is that even if you make a private language, if it is a good programming language it will be good in ways that other minds will appreciate, too. You will need to read it later, as a different person. Making organization and regularity key aspects of success.

This is a kind of regularity that exists on the plane of what computer science is about. It is about how to structure and communicate ideas.

This is something that exists, maybe in a whole subfield that is yet to be discovered. Why is it useful to make a language to talk about a problem and then talk about the problem in that language?2 Maybe information theory, cybernetics, process analysis whatever, psychology and computer science are somewhere on the edges of this.

You might call it the surprising effectiveness of languages or something. And a twist to this is that private ones are not better than shared ones.

Parallel thought paths:

If multiple persons observe the same cloud, they might see similar patterns, they might say similar things about the cloud. Without ever knowing what the others have said or thought.

To see a cloud and to have a feel for what others might think, without knowing what they say - this is an advanced communication problem. A parallel thought paths communication problem.

One way of sharing the same thoughts is to simply be sufficiently similar.

This is one of the reasons why self-evident code is the most powerful code.

When a superintelligence writes code

It will solve communication problems we did not even know could be solved.

Superintelligence might come up with programming paradigms and ways of expressing themselves to the computer and other programmers that will make a grown person cry and explain the beauty of nature. It might produce code that looks like it was carved from nature. And I mean it would look like the problems are trivial and self-evident to solve. So it will make nature look self-evident or something.

One funny thing is one of the first things it programs are the engines of intelligence. So after we achieve machine AGI, we will get an AGI program from it, and it will be so simple, so self-evident, so obvious, that we will wonder why it took so long.

When it re-writes itself in Lisp, this is where I draw the line and say, Now we are starting to get serious.

Footnotes:

2

Eric Normand has a podcast episode on this.

Date: 2023-08-20 Sun 12:53

Email: Benjamin.Schwerdtner@gmail.com

About
Contact