Slide 1

<body text=#000000 bgcolor=#FFFFFF link=#000080 vlink=#800000 alink=#000080> Dangerous Ideas is lynx-friendly, at least partially.<BR> In general I only include the teaser for the text version, but in this case I have full notes for all the slides. Please let me know if there anything still making this page difficult for anyone to access. Just mail gremio at ai dot mit dot edu. <HR> <CENTER> <H3>Radhika Nagpal</H3> <H3>from: The Amorphous Computing Group</H3> <H3> <a href="http://www.ai.mit.edu/lab/dangerous-ideas/">The Seminar on Dangerous Ideas</a> </H3> <H3>1 p.m. Wednesday, April 10, 2002</H3> </CENTER> <h1> Teaser - Emergence is not Mysterious </h1> It is fascinating how global phenomena can emerge from the local interactions of millions of simpler individuals. Colonies of ants and termites cooperate to acheive complex global tasks. Simple cellular automata rules produce complex patterns. Cells with identical DNA cooperate to form incredibly complex structures, such as ourselves. As new technologies make it possible to embed millions of tiny computing and sensing devices onto bridges and airplane wings and into robots, we are forced to move away from a centralized mindset and these emergent systems look incredibly attractive. But how can we use these mechanisms when they appear so mysterious and when we are not sure what "emergent" really means. <H1> Slide 1 - Motivation </H1> Smart Materials, Smart Dust, Self-Reconfigurable Robots <BR><BR> Technology like MEMs and biocomouting making it possible to embed all sortf of novel and radical applications are becoming possible <BR><BR> programmable material that can change it shape - like an airplane wing or sheet that can programmabled to be a whatever, selfrepairing bridges<BR> reconfigurable robots that change shape, space structures that fold compactly<BR> Many of these apps are already being built<BR><BR> Programming such systems is a challenge:<BR><BR> For the most part focussed on centralized applications of traditional control theory, heirarchical control, or heuritisc based searches. Not only are these not scalable, but they are fragile, and they ut pressure to build complex reliable pieces rather than cheap ones. Although many attempts to understand global phenoma, hard to use as engineering tools. cellular automata, artificial life, eveolutionary computation no framework for poducing local rules that solve particular goals Evooltionary approaches produce rules for particular goals but without any understanding why they work, makes difficult to erify that and crafting fitness functios.<BR><BR> Example of programmed self-assembly that combines ideas inspired by studies of embryogenesis with engineering <H1>Slide 2 - Centralized Approach</H1> <UL> <LI>Centralized or Hierarchical Control</LI> <LI>Centralized / Heuristic Searches</LI> <LI>Issues: <UL><LI> Vast numbers of almost identical elements </LI> <LI> Irregular, time-varying interconnects </LI> <LI> Local interactions and local information </LI> <LI> Limited resources; Limited global information </LI> </UL></UL> <H1>Slide 3 - Emergent Systems</H1> <UL> <LI>Pattern Formation</LI> <pre>[mathematical patterns, contoured beehives, etc. -Ed]</pre> <LI>Ants, Flocks</LI> <LI>Physical Models</LI> <LI>Game of life</LI> </UL> <H1>Slide 4 - Inverter Patterns</H1> <pre> as an example of a non-symmetric pattern which needs to be designed from a top level goal to emerge from bottom-up systems, we use a multicolored image of an inverter gate (red fork from left interacting with positive and negative yellow and green doped lines to a center, from where another red lead goes out to the right) -Ed. </pre> <H1>Slide 5 - Inverter Patterns</H1> <pre> more of them shown in a series. The emergent space has segmented itself to form several of these in a chain. -Ed.</pre> <H1>Slide 6 - Inverter Patterns</H1> <pre> The segmentation process is detailed: first a "local neighborhood size" is determined, based on the distance between the parallel edges, top and bottom which are previously started off blue. -Ed. </pre> <H1>Slide 7 - Shapes and Patterns</H1> <pre> What kinds of things can we generate emergently, which we have just thought up this way? </pre> <UL> <LI>Shapes: flat folded shapes</LI> <LI>Patterns: all plane-Euclidean constructions</LI> <pre> plane-Euclidian: anything you can construct with a ruler and compass. </pre> <H1>Slide 8 - Biologically-inspired Primitives</H1> [From slide notes: -Ed.] <UL> <LI> count up (estimate of distance)</LI> <LI> local average to improve integral distance estimate</LI> <LI> gradient has a name </LI> <LI> multiple sources may emit gradient same name (shortest distance to any) (almost always case)</LI> </UL> ?? time (away from), redundancy, asynchornicity issues ?? [My notes from the presentation: -Ed.] Gradients: point sources (compass) and lines of cells (ruler) can send out gradients, messages that weaken with distance (or hops depending on the message style). These can interact: a line gradient can establish an impassable "crease" that a point-source gradient might not be able to pass. The "biologically inspired" part is the bee in the top right corner. It shows both radial patterns and bilateral symmetry. <H1>Slide 9 - Shape Transformations</H1> [ Images of mollusks or similar shapes, but in different distortions. ] D'Arcy W. Thompson, "On Growth and Form" [ Parallel images of the inverter distorted based only on starting positions of the two boundary lines. ] <H1>Slide 10 - Dangerous Ideas</H1> <UL> <LI>Emergence != Mysterious </LI> <LI>"Global" rules are as important as local rules</LI> <LI>Evolution is not the only source of "general" ideas in biology</LI> <LI>Artificial Systems can contribute to understanding biology (or, system-level theories of biology are useful)</LI> </UL> <H1>Slide 11 - Neat Examples in Development</H1> [ Cockroach legs: If you splice them, they regrow. Interestingly, a piece reattached backwards will redifferentiate the right way around!!! ] [ Examples from biology of things with lateral or radial development ] <H1>Slide 12 - Related Shapes in Biology</H1> [ Fly heads that look a lot alike except for distortion, and line drawings of them in the appropriate distortions, again from only the starting conditions. The flies in question are antedisper, prodispar, dimidiata, dispar, dissimulata, and exuberans. ] P. A. Lawrence, "The Making of a Fly" <H1>Slide 13 - Dangerous Ideas</H1> <UL> <LI> Origami is not an art, it's an OBSESSION </LI></UL> <H1>Slide 14 - Origami</H1> [ Origami figures: butterfly, Pegasus, seashell, Escher shapes / drawings, dragon, and an ornate immense wall clock! ] Models by Robert Lang, Angela Baldo, and others. <H1>Slide 15 - Practical Applications of Origami</H1> EYEGLASS Space Telescope [ Folding Frenell lens, or solar cell array for space applications ] <HR> <BR><BR> email radhi at ai dot mit dot edu with questions about the presentation.<BR> email gremio at ai dot mit dot edu regarding Dangerous Ideas or the D.I. page.