Social insects can do amazing things collectively that no insect could do on its own. Ants, Wasps, Bees. - [Collaboritive aspects of social software and prosthesis - INTERESTING SIDELINE - ARE WE TALKING ABOUT developing the social code of laws between people [social rhetoric? maybe - rules of engagement? shared discourse? -codifying our natural social responses to democratise them] so that prosthesis becomes behaviour?] Algorithms and systems that can operate in the same way so as to solve complex problems - that's the aim. Shaped by millions of years of evolution - flexible a colony can respond to internal pertubations and external challenges robust taskes are completed even if some individuals fail decentralised there is no central controller in the colony self-organised paths to solutions are emergent. WHAT'S REALLY IMPORTANT - the decentralised bottom-up mindset. Bio-inspired computing (swarm intelligence mindset) is about looking at the world from the bottom up. [true?] SO - this is the question... HBow do we shape emergence? How do we define individual behaviour and interactions to produce desired emergent patterns. STARVE THE ANTS FOR A FEW DAYS IS THE ONLY WAY TO GET THEM TO DO ANYTHING. In the double-bridge experiment there are two equal paths that could be taken. The first ant randomly chooses. If the first ant has returned to the nest then it will have marked the path with pheromone and will be therefore more likely to follow the same path. Third ant ("french ant!") decides to follow a different path (rogue). Fourth ant still more likely to follow first one. Interesting - kind of like preferential linkage in network theory, although perhaps this is a forced distinction. QUESTION - how do they optimise? If you change the path so one is longer than the other, HOW DO THEY CHOOSE THE SHORTER PATH? Mechanism -> the first ant that returns to the nest has gone away from the nest and then returned. The RETURN is more likely to be direct (? have I got this right?) and therefore on average it weights the correct path. Not very robust... But if the long-path is the one that's shown first and it gets established, then the ants won't switch to the shorter one. Seven species of ants use evaporation of pheromones. The pheromone pathways can dissipate. Longer threads are harder to maintain, and so are gradually abandoned (particularly if there are better ones). INSPIRATION FROM NATURE TO DESIGN COMPUTING SYSTEMS - is it about computing systems or is it about the way people interact. Two things going on here - could be a pure tech thing - how do algorithms work, the other might be how do you get the dumb aspects of human beings to operate in an intelligent way reinforcing things. The pheromone thing reminds me a lot of the front page of the BBC site where your own trails are more heavily established over time. Doesn't help anyone else though, which is a shame... [THINK ON THIS MORE] [Consider the fact that google pagerank is a sort of pheremone trail to web sites...] [Good point] The Travelling salesman problem (with ants) Funny -> there are three things you don't do in public and one of them is mathematics. MATHS INTERLUDE I DON'T UNDERSTAND. [anyone able to annotate the maths here?] JASON/Tom: "This seems like stuff from last year so far. I wonder if there's going to be some new stuff? // Maybe the context is interesting enough? That this year it's being thought of by a lot of people interested in the social behaviour of people rather than the problem solving aspects of social algorithms?" [the context *is* different, you're right. last year it was new models for computing, now we can consider this in parallel to the swerve (? - emergence thing. people swerve in cities because of the architecture, planning etc. and that gets encoded back into the environment in the form of shops, etc)] [not the new model... moving beyond biomimicary/emulation to application: i will try and cf. some of my notes from doors of perception on biomimicry] [Ah, so it's a refresher course for folks...so that we have it in our short term memory for our conversations for the rest of the conf?] using ant-inspired models in JIT manufacturing/supplychain type stuff: copes with glitches and perturbation in the system very well [what would happen if a glitch was a false +ve... would it be amplified out of control??? there's a michael chrichton novel here!!!] The problems is that people don't work together by nature, they have to be prodded. This is I think there's something really interesting though in plugging into the darker, more instinctual or functional aspects of people and how they interact with one another socially (dominant, submissive, maybe? stuff of that level - geographical sensation and navigation) that could be enhanced by developing something that responds to how people respond to each other without them realising it. IGNORE THIS LINE - I THINK IT"S SHIT. x-ref eric bonabeau with robin dunbar... humans as social animals "gossip, grooming" [transactional analysis? strokes?] etc.... do ants gossip? only when they are voiced by stallone and woody allen? he reffered to them within the class of "social insects" [ants "gossip" (or share environmental information) via chemical markers that are left in the environment, it is geoblogging taken to the extreme] This reminds me of the problem of economics - people don't act intelligently. We're hard wired to ignor lot of instinct.mics always dothe the# Yes - people don't necessarily operate intelligently, we operate in part by instinct. That's interesting though, because if we operate by instinct in part maybe we can 'fake' individual instinct in the way that social software inserts the code of laws, the codified behaviour that we use between each of us. Maybe we can BUILD IN INSTINCTS (like fear or size or sensing dominance or whatever - but the NEXT STAGE, you know? new stuff?) to compensate for our less helpful (outdated territorial?) instincts and replace them with more collaborative ones. Instinctual prostheses... [see all of the recent research and publications on "evolutionary psychology" and group behavior and psychology work (tipping point, meme work, etc)] 2 Simple rules rule [nice better than K.I.S.S. :-)] Bucket brigades in harvester ants - Messor barbarus in southern spain retrieve seeds from a source in a bucjet brigade of up to six workers. The first and smallest ant collects a seed from a source and starts to carry it along a trail towards the nest until it meets a larger worker. Theis larger worker takes the seed from the ant and continues to transport the seed towards the nest while the smaller ant turns and walks back towards the seed source and so on... georgia tech modelled this with humans - human bucket brigade at taco bell: http://www.isye.gatech.edu/faculty/John_Bartholdi/bucket-brigades.html (there is a nice little java app illustration on that page) [there's some interesting queue theory docs about this, to do with removing bottlenecks and encouraging progress through processes in the NHS] this slide is all greek to me I CAN READ THIS IF HE STICKS WITH IT! Dammit! Put it up properly, and I'll translate it for everyone! [ooh, look who knows Greek...] RULE ONE: pick a protector and an aggressor, then move so that your protector is always located between you and your aggressor. caption: you | protector | aggressor [this puts you in the role of protected] RULES TWO: pick a protected and an aggressor, then move so as to be always located between your protected and his / her / aggressor. caption: protected | you | aggresor [this puts you in role of protector] Bad news! Difficult to predict collective behaviour from individual rules. Ask one of the participants it won't tell you anything about the function of the group. Small changes in rules lead to different group level behaviour. Good news! Possible to predict group- level outcome using Bottom up simulation. Possible to efficiently control (grrr) Simple rules and bottom up modeling at SWA Problem Optimize cargo routing Use simple rules Results 71% improvement At least $10m/year Internal reinforcement mechanism can go haywire- the lemming metaphor, circular ant mills. Rule #3 - no one needs to be in control [humans always want to be in control] [but they don't NEED to be. If you can structure a site to build in a way that people's desire to be in control are controlled and channeled then it'll work - cf. parlialmental democracy - no ONE person is in control. And that's when people CHOOSE to obey the rules. You could easily develop a site which instituted a perfect Anarchist state because the rules would be UNBREAKABLE as in "an-archos" : "no leaders" [what is the human driver for control/status? to tom's point... do we NEED to be? what is the evolutionary explanation?] ] wrms of robots {cf. natalie jeremienko's packs of feral robot sensing dogs] apwheromonic false positive + reinforcement mechs causes "circular mill is this different to the Wolfram stuff? small rules cause large unexpected results. same as the wolfram stuff, mathematically, at least. It is not "Wolfram stuff" as it was all original years earlier. (though Wolfram is a nice package of the ideas, ignoring the "I invented a whole new science" constants) correct -- "cellular automata research" might be a more appropriate term. where's jeff goldblum to lecture the hubris of the scientist who thinks he can control nature! "Nature will out!" Optimal work-sharing emerges spontaneously? [didn't all the time and motion studies of the 60s show that it doesn't? - but it's not neccessarily good for workers] I think this is in very specific circumstances. We lost a lot of this in the transliteration process... hasn't all this been done before - crowd theory, movement of people through spaces etc. etc? interesting to x-ref with the work of CASA: http://www.casa.ucl.ac.uk/research/index.htm] [wicked: it's wizball!: http://marticus.midnightrealm.org/wiz/][paint the mountains!] DON'T LEAVE WITHOUT LEAVING YOUR NAME FOR THE KEY (people using Hydra will know who wrote what [doesn't saving the file get rid of the colours? I don't know. Maybe we just can't save it! There are a variety of ways you could improve the program, but I think it's still useful in its own way.]): true.. the list of authors below is good. Maybe we should take screenshots? Could be. someone tb ping http://www.oreillynet.com/cgi-bin/tb/tb.cgi?__mode=list&tb_id=e_sess_3548 when posted please. It would be cool to dump this into a wiki for subsequent collaborative editing and addition. dump it here.. (/me finds url).. here's the "official" wiki for the conf: http://www.socialtext.net/etech/ for the record this is an incredible thing. esp. being send the app in a heartbeat to join in over rendezvous!] This green is Tom Coates This is Jason Kottke jason kottke.org http://www.kottke.org This is Jim McCoy This is Matt Webb matt interconnected.org http://interconnected.org/home/ This is Chris Heathcote ( c deaddodo.com http://www.undergroundlondon.com/antimega/ ) This is Matt Jones This is Trevor F Smith (tfsmith parc.com | http://trevor.smith.name/) This is Jesse James Garrett This is Allen Murray (allen murray.net) Kibbitzing by Rael Dornfest [and the notetaking goes on...] [Was this rule #4? I missed the slide that preceeded this] Specialization and Polyvalence Automatically switches as the colony reaches a particular size System efficiency is tied to colony size, starts at growth phase, goes through an inflection point and reaches saturation. This is "swarming" (or at least the trigger for it) Breaking off a chunk of the swarm returns it to the point where increased size was delivering diminishing returns. the science of prey ("are you lunch or are you planning lunch" - robin dunbar: http://www.liv.ac.uk/evolpsyc/rimd.htm) Rule #5: Lets hammer on Michael Crichton :) "emergent behaviours by design" [ Shaped emergence? [Gardening! yeah!]] I think so, you shape the emergence by tweaking the rules slightly (a la CA) - or by genetic algorithm-like selections of rules? "interactive evolution" So, emergence (as used in this conversation) is orthogonal to intention? Yes. Emergence is the interaction of intentions into group action. The "group" in this case includes both the agents and the agent system designer... The result ("group intention") are not necessarily easily predicted and often depends on environment, etc. Environment and wierd bugs in your simulation often end up having a big effect as well :) [insert various stories of GA/GP research finding the "cracks" in the simulation and doing very strange stuff...] "life is what happens while you're making other plans" - lennon [?] "How do you make God laugh? Tell him your plans." -- ? intention is environment driver - for adaption/adaptation (which one is it?) This reminds me of the "persuasive engineering" book I have but have not read. chinese streamer :) [is it streamer or screamer? can't tell.] more conway's life stuff. local answers are different to the global situation COMING SOON: More distributed optimization / control applications (scheduling, resource allocation, routing Self-healing, self organising comm networks swarm robotics will move beyond mapping self-aggregating devices self-organised satellite deployment and maintenance (NSF/NASA/EPRI) Controlling swarms of UAVs, UGVs, UUVs Swarm-based sensor networks, smart dust [a line from a presentation by some experimental architects i saw "see the city from the point of view of a swarm: by attaching mpeg cams to pigeons][mobile phones for cats] Social engineering of collective phenomena [smart-mobs? i think that the "engineering" part of this is still to be proven, but observations seem to be starting... [isn't that like marketing, product placement with the mavens, etc?] I can buy the case for marketing and some political stuff being a concrete example of engineering, but I think that we are still working in the dark on some of the theory. Lots of cool speculation, but nothing is really proven. We know what works, but not necessarily why. Gibson's "pattern recognition"] blimey... what a bombshell to drop as the last bullet [do we engineer for the corporations or for the users?] [Do engineered emergent systems differ from non engineered system? detectably? from within the system (with only local knowledge)?] - I'm not sure I know about this - IS this smart mobs? Is this what we;ve been dalking about? yeah totally... MORE! Questions: How do you deal with diversity and what about dishonest/malfunctioning agents? Agent diversity is good and adds to robustnes. The latter will probably end up being social network reinforcement... (herds, crowds, etc.) question: is this a blueprint for a "non-human readable world"???[cory i think] [yes it is. after spending too much time trying to decipher evolved lisp code I can tell you that the only thing your can really figure out is that it works, the more complex the environement the more opaque the logic of the optimized "solution"] e.g. you flight is going via all sorts of non-obvious but efficient waypoints noise is good. it can peturb a system out of a local maxima that is not globally optimized. [this is such aDarwlobne oof concbastealintncally said thmaximing.is thais holding back the system !!!1! Yuou're not wrong... Anyone able to strip it out? Undo That might work. Ok. I'm going to go into the hall now. Hopefully I wont' lose the share, but it's possible. Articles (just in case we can't get a copy of this afterwards): Swarm intelligence, Scientific American, March 2000 Swarm intelligence, Harvard Business Review, May 2001 Predicting the Unpredictable, Hardvard Business review, March 2002 Real-world applications of agent-based models of human systems, Proc. Nat. Acad. Sci. USA, 2002 Scale-free networks, the key to robust networks, Scientici American, May 2003 ******************************************************************************* Would be interesting to have a post mortem on this experiment after the lecture Everyone should add email, http to their name callout above... ******************************************************************************* (Allen Murray: as part of the experiment, I am watching the dialogue from the hallway) well, that was lots of fun. that was fantastic. who wants to play The Game at some point IRL? Shit.e as a destination. "If you are going to