BLOG

 

Some further considerations regarding:

One future direction of blockchain technology.

Heterogeneous tokens algebras with history dependent free behavior.

 

      Word count: 1.350 ~ 6 PAGES   |   Revised: 2019.3.26

 

 

— 〈  1  〉—

LET'S MAKE IT

 

A general problem solver is a system that can try different operations on inputs, compare with goals, and backtrack. Or simply do something randomly or not selected on a given, pass the transform onward, this being the given to another general problem solver, that more or less randomly decides to do something and does it, and so on, and eventually the transform is compared to some goal, and if it fails, the system backtracks from some point and tries again. Noting the failure, and such notes can be passed around like or rings of genetic material between bacteria and books between human individuals.

I'm thinking again about a system of automated services, which can appear in a single virtual world.

---

Was thinking about one use case that can be aimed at with these kinds of systems. Which has both social media and game and phone applications uses.

Intelligent "picking things up" and "moving" them in a scene. Or even doing something more. Not just classifying an image by what appears in that image.

Suppose an end user drag and drops an image with an relatively contrastive object somewhere that inside says cat in white. (Image may be without the tag, end user could spawn an agent to tag parts of an image themselves, say where to print what word.) Next that end user types "move cat left some" ... and system displays another image where the box is now some number of pixels to the left.

Potential for making new memes conveniently on a phone or without leaving the system may be interesting.

If available on a particular network, end users have reason to switch. To have more flexibility when mobile or lazy when making memes. (As opposed to opening up an image editor or just typing text over an existing image in one of a few places as some sites offer ...)

The "picking things up" in a "scene" which traditionally is not "interactive" at all, is "old media", if we use the terminology of McLuhan [MCLU64, MCLU70], or certainly not "interactive" without another large traditional program that requires human control and intelligence, and doing something with these object "in" that "scene", is something that could be popular or could have wide demand.

Most importantly, in the new language, a concise, impressive algorithm in terms of actor agents should be possible. Or more correctly, feasible. What is more concise is often more feasible. As opposed to not. Given some limit on development resources. Higher probability that such a network with the same number of developers could add more such features quickly. Possibly less risk in the sense that for making any feature less sunk cost would occur.

(For example, stack create( ..., [operation1, operation2, ..., create(... create(...), create(... create(...)), cre...], [...) One reason why we like this capability, two to three sentence programs with surprising power. )

Basically a significant fraction of "old media" becomes an interactive scene. Which makes it more convenient to make viral content --- popular content.

Underneath what actually happens is feature extraction by a small horde of general problem solver agents, which can backtrack and communicate while pruning transforms until they find labeled features. Making parts of the image editable by the user with a few words. Interactive memes may also be a theme, if a script is allowed to be transfered along with an image with tags. So image with text, which the internet is good at sharing, becomes viewed on the network, a small "world" in which end user can collaborate. Could be interesting.

The key point could be discussed as the fact that the current internet and social media is full of existing content that can suddenly in such a system become interactive if minimally labeled by end users during the process of making memes, giving a higher probability for users to prefer to switch to the new network.

The internet makes it easy to share images but hard to share programs, just the way it is currently. So allowing images to be interactive, treated as scenes, may be appealing. Could mesh well with existing systems end user would like to continue using.

---

In general, had the following idea regarding simplest to deliver capabilities assuming one has a network of networks of general problem solver type actor agents. One of the things that is useful and impressive but for a program can be developed very rapidly and concisely using a language that concisely controls such agents.

One type of use case demo that can be shown, which has various game application uses is intelligent "picking things up" and "moving" them in a scene. Or doing something more with them in a scene. Not just classifying an image by what appears in that image.

Suppose an end user drag and drops an image with a box filled blue somewhere that inside says cat in white.

Next that end user types "move cat left some" ... and system displays another image where the box is now some number of pixels to the left.

The "picking things up" in a "scene" which traditionally is not "interactive" at all, is "old media", if we use the terminology of McLuhan, or certainly not "interactive" without another large traditional program that requires human control and intelligence, and doing something with these object "in" that "scene", is something that could be popular or could have wide demand that allows a concise, impressive algorithm in terms of networks of this kind. The test operate test exist neuronal model building and operate learn backtrack operate random patterns with sufficiently many given operations plus operations constructors that add new operations admits very concise algorithms for extracting the proper subset of the image marked cat where sufficiently obvious and putting it elsewhere and filling in previous spot with background, in sufficiently many cases to be interesting. Games with plain images, if any image can be interacted with just a few phrases, may be viral on a social media network. The significant question is, as says [DIR39] regarding different languages, not so much what can or cannot be said or done in a language but what can is easier or far less easy, automatic or far less automatic, more or less error prone, more or less concise in that language. Something that is difficult to quickly write in a language, but still quite possible, is easier to automate and even automate the automation of, and far more concisely written in another language.

 

— 〈  2  〉—

POINT

 

It seems to me that one significant bonus provided by having such tools "live" on and "breathe" the data of a blockchain or any other distributed graph having sufficiently persistent internet paper-like structure (as compared with anything and everything that is significantly easier to change or revise after being displayed or distributed and seen by those who will see it) is the ability to build [MIN56], Simon, Hewitt, ..., [PEA84] types systems that backtrack quite a bit over a space of operations.

They simply can backtrack more likely than not: the data is not going anywhere. Or even if the images or some parts of the data are less persistent that the rest, more like the typical internet to change, there is far more that can be relied upon to backtrack over if, for example, the end user gives instructions, and returns a few hours later, and find that they could have been automatically carried out better and notifies their program. Especially if the end user notifies the program in a persistent manner. For example, such that relatively distant past notes of this type can also be viewed and the program can guess what the user wants even if very sloppy or strictly vague language is used by that particular end user.

[Will be updating this post later ... A formal way to approach such things via considering a set of actor agents as a sheaf over a test space of operations on data having various types, where the significant thing is that each operation, not necessarily each given, each datum, has an unambiguous type. This is something that can be considered unsupervised learning, but seems to me to be supervised in some genetic sense as suggested as happening in nature by VAL13 and other before him, in that the type of each operation is known, it is known what range of motion an arm can do, and if some things can be manipulated by the arm and other things cannot, even vaguely, that classifies things into two classes based on information that one inherits or begins with, even if nobody is marking behavior as correct or not. This is much like [SELF59] with his model of the Pandemonium System.]

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License Text and images: ©tibra. Disclaimer: This text is a popular, speculative discussion of basic science literature for the sake of discussion and regarding it no warranties of any kind exist. No promise to do anything or that anything is done is involved. Treat it all as conjecture about the past and the future. As subject to change. Like the open future itself is subject to change. Except if this text happens to be explicitly fantasy or science fiction, in which case it is just that. Then exists another Disclaimer: This is a work of fiction: events, names, places, characters are either imagined or used fictitiously. Any resemblance to real events or persons or places is coincidental.