Read on Twitter
I’m curious why RL people don’t spend more time thinking about collective behaviors (/swarm dynamics).
We can drive a system with only a few simple, local rules – and watch as wonderfully complex behaviors emerge!
Here are some collective behaviors basics
1/n
We can drive a system with only a few simple, local rules – and watch as wonderfully complex behaviors emerge!
Here are some collective behaviors basics
1/n
First, take a second to appreciate the beauty of collective systems in nature!
Even into the 20th century, scientists were so perplexed by bird flocking that some asserted ‘telepathy’ must be the answer.
2/n
Even into the 20th century, scientists were so perplexed by bird flocking that some asserted ‘telepathy’ must be the answer.
2/n
But fear not, gentle scholars, because once laptops became powerful enough, some Italians strapped cameras to the rooftops in Rome and tracked those birds in 3D… 3/n
…and we learned that the rules of the game were much simpler. No telepathy needed!
For example, each bird in a flock only seems to pay attention to its ~7 closest neighbors.
https://www.pnas.org/content/107/26/11865#ref-22 4/n
For example, each bird in a flock only seems to pay attention to its ~7 closest neighbors.
https://www.pnas.org/content/107/26/11865#ref-22 4/n
Even before 3D bird tracking, we were building simulations to use simple rules + produce global order.
One classic is the Boids model by Craig Reynolds, with just 3 rules (separation, alignment, and cohesion):
5/n
One classic is the Boids model by Craig Reynolds, with just 3 rules (separation, alignment, and cohesion):
5/n
And there’s so much more to understand beyond birds and boids! Fish schooling and migrating herds, sure, but also:
-People? If you want to figure out how to best design buildings and metros for emergency evacuation, you gotta study how people crowd:
6/n
-People? If you want to figure out how to best design buildings and metros for emergency evacuation, you gotta study how people crowd:
6/n
(Metalheads can just mess around with mosh pit simulators instead:) http://mattbierbaum.github.io/moshpits.js/ 7/n
Or,
-Cancer? Isn’t that a divergence from normal coordinated behavior?
-Skin cells (working together as a tissue)? If only we could speed up wound healing…
https://twitter.com/i/status/1344811050600960003 8/n
-Cancer? Isn’t that a divergence from normal coordinated behavior?
-Skin cells (working together as a tissue)? If only we could speed up wound healing…
https://twitter.com/i/status/1344811050600960003 8/n
-Robots? Imagine capturing the rules which govern biological systems and using them to coordinate groups of bots.
Like Prof. Radhika Nagpal’s termite-inspired construction teams: 9/n
Like Prof. Radhika Nagpal’s termite-inspired construction teams: 9/n
And so much more!
Anyways, it’s a neat field that I hope the broader AI community can engage with more fully. Anyone interested in discussing modern models + papers should feel free to message me! 10/n
Anyways, it’s a neat field that I hope the broader AI community can engage with more fully. Anyone interested in discussing modern models + papers should feel free to message me! 10/n
