Five – Dhairya

DigitalHands

Assignment 5 – Cassandra

Structuring Information for Peer-to-peer Learning

(aka. swapping out your brain for someone else’s)

Models

Google’s mission statement is to organize the world’s information, but rendering knowledge in searchable form is only part of that problem. The other part of the problem is to provide structure to information. As Ashby framed in the end of his 1956 book [1], intelligence amplification is mainly a “selection” problem. We are simply presented with much too information to utilize. Some of this content is wrong or low quality. In contrast, high quality content is both factually correct and reusable in the sense that one high quality idea can be used to beget many new secondary ideas. This type of high quality content with a large impact is a model. In fact, the objective of the entire scientific discipline attempts to take vast amounts of downstream observations and organize the observations in such a way as to be explained by or inferred from a minimal number of upstream models. The quality of a model is judged by both its ability to explain current observations as well as to make predictions for the future. The structure provided by these upstream higher order models (and not the downstream observations) comprise intelligence.

Some of Google’s information is indeed about these models, but it represents information about models equally and in the same fashion as other information. It may be of interest to instead treat models as a separate and more powerful class of knowledge when indexing world knowledge.

Life Models

There are many classes of models. Scientific models are the subject of academic research. There are also implicit life models that each individual assumes. These individual models can arise from (1) conscious deliberation, or they might simply be (2) implicit habits picked out through life. Indeed, a considerable part of variance in cognitive intelligence seems to arise from a person’s deliberate selection of a world model and their daily habit patterns. These two types of models, conscious models and habitual defaults, may serve as vessels for transferring intelligence models from person to person.

As a user scenario, imagine an application for a wearable like Glass that lets you “wear someone’s habits”. If you find a daily routine that works well for you, you can share it as an effective habit schedule for other people to try. Alternatively, you can try out the habit schedules of others by downloading them and having Glass prompt you with just-in-time situational alerts for executing a particular habit. This device would allow you to be more productive by emulating people you respect. Or it might let you “live the life of someone else” as an interesting costume change. Or it might serve as an therapeutic mechanism for breaking out of daily habit loops.

This type of device also has interesting implications for collective intelligence. By crowd-sourcing questions to such as “what are the most effective patterns for daily life?” and “which belief frameworks are powerful?”, we can curate a haul of answers from the world’s population. Each person on the planet has either explicitly or implicitly reached a personal answer to these questions. Our only task is to render their answers into forms easily adoptable by other persons.

The task can be phrased in a variety of ways. How can we share our individual brain structure with others? How can we share our world models with others? How can we share our patterns with others? Two different approaches are proposed in the next section.

Implementation

a) The Logical Approach

Perhaps the most obvious approach is to condense our individual philosophies into their purest most form and explicitly express them. In this case, the technical challenge is to design simple descriptive languages [__] for models of this type. Interested parties may wish to swap out their current model for a new one, consciously and deliberately reprogramming themselves.

Such a descriptive language for specifying world models will likely require both what and why components. The what-component can be modeled after rule-based logic systems [__] of the style “WHEN [EVENT A], DO [THING X]” The why-component that converts rules into reasons can be used for building hierarchical trees that can be further used converting reasons to higher-order value systems.

In a continuation of the habit-based example above, some may want to adopt the morning routine of Mr. Rogers, the beloved actor from the children’s show Mister Rogers’ Neighborhood. Mr. Rogers’ morning routine [4] can be summarized as “waking up at 5 a.m.; praying for a few hours for all of his friends and family; studying; writing, making calls and reaching out to every fan who took the time to write him; going for a morning swim; getting on a scale; then really starting his day”. The first few lines of a Mr. Rogers daily program could be written as:

(what) WHEN 5AM, DO wake up. % (why) “Because time is precious”

(what) WHEN DONE, DO pray for friends and family % (why) “Because I believe in God” AND “I love my friends and family”

Interchanging models for those of others will causes us to swap out our routinized defaults for a new set of defaults and will likely result in a cascade of effects. It may be particularly advantageous to replicate models of the people that you admire, for instance, an established researcher in your discipline. It may be fun to adopt the models of one of our ancestors, for instance, to live a day as your grandmother would and capture her behavioral and life lessons. It may be a literal way of experiencing “life in someone else’s shoes”, and feel more connected to either strangers or people that you know and care about.

b) The Chaos Approach

The alternative approach is to give up on rationalizing behavior and instead attempt to infer method from madness. Consider the following design of a file system designed for brainstorming. First note, that it is possible to infer the content of someone’s brain from her computer file system, particularly when the computer is used for storing and organizing ideas (as in the case of a researcher). It is possible to organize the file system in such a way that it can be used for brainstorming if specific structural rules are followed. My personal file system is organized in such a way to facilitate brainstorming by having two searchable folder ‘mentors’ and ‘me’. The ‘mentors’ folder contains plain text versions of the most inspiring research papers. The ‘me’ folder contains subfolders with plain text ideas of my own and plain text notes inspired by the content of presentations and meetings.

The logic of this organization is inspired by two research papers with powerful ideas: “What Would They Think” [2] and “Remembrance Agent” [3]. First, the idea of the Remembrance Agent to look at what you are reading and writing and propose past content based on related keywords. Second, the idea of What Would They Think is to show the affective reactions of respected mentors when reading new content. We can combine the two ideas to address Ashby’s “selection problem” identified in the first section of this paper– that a large component of intelligence involves the selection of the most useful information out of the large pool available. The proposed file system for brainstorming searches keywords against the ‘mentors’ and ‘me’ directories to bring up relevant material as well as near-miss material that are less relevant but serve to bring in tangential ideas.

The important idea here is not my file system itself but that if multiple users organize their file systems in the manner described, the file system becomes an external representation of an individual’s mind. File systems then become interchangeable in such a way that I can brainstorm with someone else’s personal ideas and mentor preferences. The file system brainstorm serves as an information dump of someone’s mind that has not been distilled into logical rules but instead persisted as the chaotic pile of quirks and nuances that embody ourselves.

Proposed Artifacts

To recap, the artifacts proposed within this paper were:

1. Wear someone else’s habits wearable for REPROGRAMMING SELF WITH SOMEONE ELSE’S HABITS
2. Remembrance Agent + What Would They Think file system for BRAINSTORMING WITH SOMEONE ELSE’S MIND

Conclusion

Both proposed systems propose the augmentation of intelligence facilitate the spread of powerful habits and ideas, giving us the framework to swap out our minds for the more preferred and intelligent minds of others. The proposal is only for systems to provide additional structure and transparency to knowledge that the world already has. Control remains firmly in the hands of each individual who has suddenly been endowed with the capability of self-intelligence augmentation.

References

1. Ashby, W. R. (1956). An introduction to cybernetics. Taylor & Francis.
2. Liu, H., & Maes, P. (2004, January). What would they think?: a computational model of attitudes. In Proceedings of the 9th international conference on Intelligent user interfaces (pp. 38-45).
3. Rhodes, B., & Starner, T. (1996, April). Remembrance Agent: A continuously running automated information retrieval system. In The Proceedings of The First International Conference on The Practical Application Of Intelligent Agents and Multi Agent Technology (pp. 487-495).
4. Hattikudur , Mangesh. 15 reasons Mr. Rogers was the best neighbor ever. http://edition.cnn.com/2008/LIVING/wayoflife/07/28/mf.mrrogers.neighbor/index.html

 [Slides from class]

Assignment 5 – Sophia

Good Eye

Teaching composition and design principles to artists and designers is difficult.  Some people have an inherent knack for composition, and, for others, it is learned.  In both cases, it develops only with much looking and practice, and it is not something which can be easily learned from a book. Traditionally, it is taught by showing slides of the work of other artists and designers and then by critiquing finished assignments. Creating a painting or making a design, however, is made up of a series of hundreds decisions where the artist/designer is working without guidance. Only after many hours of work, hundreds of steps later, might an artist receive any feedback on their work. Often the most paralyzing part of the process is staring at a blank or nearly blank page and not being able to determine what the next step should be. I propose to augment this process by building a tool which assists its user in developing a “good eye”.

 To do this, I suggest a system that works with the user during the process of developing a composition, aiding them in making all those small decisions as opposed to only receiving critique once a piece is finished.  I imagine that the user would place a shape on the canvas or screen, and the system would propose a series of possible next steps that the user could make on screen or projected directly onto the work.  Even just being able to see how a one choice might affect the rest of the composition without committing to it would very helpful because this is something that can be hard to do when using physical materials (cutting out shapes of paper, etc.). The possibilities would be varied in shape, value, texture, color, and line, which are the ways that an artist or design can influence the feeling of the final composition.  The user could cycle through the possibilities and choose what feels best.  The user could repeat this process for as long as it is helpful.  The way the system would make suggestions would be derived from an algorithmic analysis of a bank of images of the work of other artists and designers.

I see a few different use cases for such a system.  I imagine that this could be very useful for teachers of art and design.  Teachers commonly tell students to study the work of artists before or after giving an assignment, but this sort of system would help the student during the process of making a work.  A teacher could assign a set of images to feed to her students’ systems to hone certain skills.  In addition, a teacher could tailor the input to each student, diagnosing weaknesses in the student’s design sense and prescribe artists that would help to counteract those weaknesses.  I also see such a system helping artists and designers working alone.  The user could teach themselves by inputting the works of artists they admire, and using the system would help them understand what it is about those works that is so appealing and successful.  Furthermore, the images fed to the system need not even be works of art but any source of inspiration, any image, and the system could become a new way of making work in general.

A system like this could also be easily adapted to record how an artist arrives at a composition and then play back those steps to someone else.  This would be almost like a paint by number but with an emphasis on the process of arriving at a composition and not just reproducing the final result (in which it can be very difficult to understand how the artist/designer got there), and it could also be a very useful learning tool.

When working digitally, something like this could be built directly into the graphics software. For sketching and painting, a camera could record the work in progress and the suggestions could be projected onto the work itself or shown on a screen on top of an image of the work.  Ideally, the user would be able to see the suggestions directly on their work and not on a separate screen.  Using projection would be problematic because projected colors do not have the same quality as those made with real materials, which will affect the user’s ability to weigh out their options effectively.  Also, artists need to work in bright light, and projections would not work as well in those conditions.  Projecting onto already colored surfaces also would also be a problem because the colors would mix and shapes would overlay and not occlude each other.  Many of these problems might be avoided by using AR glasses.

Such a system would also be useful for learning to arrange 3d spaces, making sculptures, etc. but this use case would likely have a confusing UI and be hard to implement.  Starting out, the algorithm for making suggestions would likely be very primitive and might not always analyze the input images correctly.  Understanding that one shape occludes another and that they are not just two shapes side by side also seems like something that would be hard to implement in the algorithm.  In addition, often good compositions are nuanced ones, and it would be hard for a computer to understand these nuances.  However, despite these limitations, such a system could still be very useful.

That this system might train artists and designers to all work in a similar way is a valid concern, but I would argue that this is no different than using any tool or technology (paint, a pencil, Photoshop, etc.), which also imposes constraints on how to think and work. By being able to iterate through the design choices more easily and reversibly, users could develop a “good eye” more quickly.  This system would also help the user learn from someone else’s working process, which is hard to do now.  I imagine this could be used in class, but that it would be particularly useful as a way for people to teach themselves when they do not have access to expensive art and design schools.  Furthermore, I believe that such a tool will help its user find her unique voice as an artist much more quickly.  As the user trains the system, she will discover what she likes and define her own style. Finally, such a system might even grow to be more than a learning tool and become a new method for creating works in ways I cannot even predict.

Slides