Thursday, December 3, 2015

Secret Science Club Post Lecture Recap: Collective Intelligence

Last night, I headed down to the beautiful Bell House, in the Gowanus section of Brooklyn, for this month's Secret Science Club lecture, featuring Simon Garnier of the New Jersey Institute of Technology's Swarm Lab. Dr Garnier's topic was collective intelligence, how millions become one.

Dr Garnier began his lecture with a brief discussion of American football- football is a complex problem of coordination involving actions with partners being performed against the actions of opponents. Dr Garnier joked that it is such a difficult problem that four tries are necessary to move the ball forward. He then noted that traffic is a problem of coordination as well. That being said, fish are able to move in coordinated schools, even though they have very small brains. He then noted that starlings are the champions of coordinated movements, showing videos of a "bird ballet":

Dr Garnier then showed videos of leafcutting ants, genus Atta, in action. These ants form nests up to twenty meters wide and eight meters deep, which house generations of workers- one single queen can give birth to five million workers. These ants are masters of coordination, even though their brains contain fewer neurons than a human pinkie does. The human brain is also a 'master' of coordinated activity- life is dependent on one's body being coordinated.

Dr Garnier posed the question, how do one million act as one? The short answer is 'self-organization'. He then showed an adorable video of Scotties feeding, noting that, with time, they coordinated their movements:

Coordinated interactions are local, repeated behaviors... if these behaviors are not constant, dispersal occurs. Coordination involves coupling of action and reaction, with various actors synchronizing like metronomes on a moving platform:

This synchronization takes place in a few steps, which Dr Garnier demonstrated in an audience-participation exercise. Audiences can synchronize clapping. He had us clap our hands, then exhorted us to synchronize our clapping with our neighbors, then to listen for more distant audience members, and to synchronize with them. Within moments, the whole audience was clapping in unison. Well played, Dr Garnier, well played.

Coordinated movements are necessary for information transfer, construction, decision making, and traffic organization. Dr Garnier posed the question, how do we make decisions? The fasted way to make decisions is to make many random trials. A better way to make decisions is to search for information and narrow down possibilities to optimize- choose the best course of action. There is an exploration/exploitation trade-off... one must spend resources to search and select the best option based on current knowledge. Dr Garnier compared the decision making process to a multi-armed bandit, a series of slot machines with different programming, with some machines paying better than others. In order to seek optimal winnings, one must try multiple machines in order to determine which provide better outcomes. Dr Garnier joked that some birds are better at making these choices than some humans.

After noting that most decision-making experiments are performed with animals that have a lot of brainpower, that organization is possible without a brain, whereupon he showed a slide of the U.S. Congress. He noted that some plants and some bacteria engage in coordinated behavior, which he termed the Homer Simpson Paradox- how does an organism thrive without brains? He then launched into a long digression about slime molds. A moving slime mold is a single-celled organism, but that single cell can have billions of cell nuclei. The yellow Physarum polycephalum slime mold and the 'dog vomit' slime mold, Fuligo septica are two of the better known slime molds. While in their mobile stage, slime molds start oscillating by pumping cytoplasm and then move in the direction of food sources, effectively making a decision in their search for sustenance. When the food runs out, the slime molds stop and develop into a sporulating form in order to reproduce. Dr Garnier informed us that slime molds are used a lot in research because they are cheap and fun to work with. He treated us to several time-lapse videos of slime molds moving, similar to this BBC video:

This zero-neuron organism is able to beat the 'multi-armed bandit' in its movements- in environments with consistent rewards, the slime mold tends to move in one direction, mainly toward the last reward. In environments with irregular rewards, slime molds will change directions, with the general movement being in the area with the highest mean of relative successes- they move in proportion to the number of reward sites. Dr Garnier paused and gave us the Twitter version: slime molds ignore failures and focus on successes.

Dr Garnier then showed an image of Berlin, taken from the International Space Station:

He noted that there was a city center with radiating arteries, allowing for ease of defense and the control of a large territory. He then showed a slide of the foraging paths used by Argentine ants (Linepithema humile), noting the similarities to the roadways of Berlin. The trails of the ants are marked with pheromones, so that other members of the colony can follow them. He then returned to the subject of genus Atta, which forms well-defined paths through its forest habitats, removing debris from these pathways. The leaf-cutter ants cut vegetation into pieces and use these scraps of vegetable matter to cultivate edible fungus in their nests. They have not only figured out traffic control, they also engage in agriculture.

The talk then shifted to army ants, specifically the genus Eciton of Central and South America. These ants form colonies of up to two million individuals. They move for a period of about two weeks and then form stationary colonies for about three weeks, during which the queen lays eggs. Due to their cycle of movement and lack of a permanent colony, they cannot form well-defined paths like the leaf-cutter ants do. They have to move quickly while carrying the queens brood. When their movement is restricted by obstacles, these ants form bridges with their own bodies, using hooked feet to lock together:

Swarming ants can also form rafts, having hydrophobic bodies:

The ants self-organize through a basic rule, "Walk all over me!" If there is a lot of traffic, an ant stays in place, if the movement behind stops, the ant proceeds. Dr Garnier had a hilarious aside about the the bites of army ants... they really hurt, which is why grad students are made to do the studies. While moving, ants seek to move the shortest distance, so their will form shifting bridges to bypass sharply angled paths:

Ecologists are the economists of the natural world, they perform cost/benefit analyses- the ants balance the building costs with the benefits, the less distance they have to move, the more efficiently they move. It's possible that swarming robots could be developed which operate in a fashion similar to the ants' behavior to bypass obstacles. Dr Garnier noted that the army ants are blind, they follow pheremone trails and are basically automatons designed to kill things and to bring them back to the nest as food. The individual ants have little memory, but they lay down little 'fridge notes' which can be followed, a process known as stigmergy.

Dr Garnier then shifted to the subject of traffic organization among humans, noting that, in 2011, Americans wasted 5.5 billion additional travel hours and 2.9 billion gallons of fuel due to traffic jams, at a cost of $121 billion. We know why it exists... density, and how to solve the problem. The number of cars that pass along a given stretch of roadway per hour is known as flux- when there are few cars on the road, passing is possible. As density increases, flux is diminished until a critical density is reached and a 'crystallization' process occurs, a traffic jam. Dr Garnier then ran a neat traffic simulator to demonstrate how density affects flux. He joked about wishing to set 'politeness' parameters, ranging from 'Swiss' to 'New Jersey', then noted that any perturbation of flux creates a traffic jam. He then showed images of the 'Snowmageddon' which crippled Atlanta in 2014.

Throughout the lecture, Dr Garnier would make an aside to display photos of his colleagues and to give short biographical notes, stressing that science is a collective adventure. Personally, I think that is a wonderful statement, and kudos to Dr Garnier for being so good to his grad students. He then described some an experiment in which subjects had to avoid a stationary individual while walking through a corridor- subjects usually showed no preference as to which side they passed a stationary subject, fifty percent passed to the left, fifty to the right. When passing individuals moving in the opposite direction, there was a social convention among individuals to pass on the right in countries in which motor vehicle traffic travels on the right. In conditions of high density, such as the Hajj in Mecca, pedestrian movement occurs in stop-and-go waves, with extremely high density causing turbulence, in which people lose control of their movement, a very dangerous situation. In areas of high population density, traffic problems increase. Dr Garnier characterized traffic jams as 'human self-organization gone wrong', but noted that we know the solution. He proposed a density-reducing solution based on a combination of public transportation, bicycle use, and smaller transportation footprints. He exhorted us to 'go back to the ants', to take turns. He noted that a flexible traffic system was needed, and mentioned that smart traffic lights could reduce congestion, noting that the city of Zurich implemented the use of traffic lights with no fixed timers, using lights that optimize traffic flow. He also mentioned the redesign of traffic patterns, such as substituting roundabouts for intersections, and ensuring that drivers don't have to cross in front of traffic moving in the opposite direction. He ended the lecture proper by joking, "I welcome our new ant overlords, may they fix the FDR and the Tappan Zee."

In the Q&A, some bastard in the audience, noting that eusocial species have evolved in distantly related clades, if swarming behavior can be induced in species which don't have well-defined social groups. Dr Garnier noted that some spiders can flip between sociality and non-sociality, probably due to hormone levels in their eggs. During optimal conditions, they tend towards a solitary existence, but under marginal conditions, they exhibit sociality.

All told, Dr Garnier delivered a fantastic lecture, one with applications to the crowed metropolitan area in which his audience resides. Let's hope that the findings of the Swarm Lab can be applied to the human swarms that take to the roads every day. Thanks again to Dorian and Margaret, and the staff of the beautiful Bell House, and to Dr Garnier. Here's a video of Dr Garnier briefly covering the subjects he mentioned in his talk:

Pour yourself a nice beverage and soak in that Secret Science ambiance.


mikey said...

What's interesting to me is a LOT of applied big data and machine learning research is based on a lot of that kind of coordinated behavior. When you add in multi-spectral sensors, lots of fast processors and many-to-many high speed network connectivity you can produce an enhanced version where systems update their own knowledge base based on outcomes and probabilities and you get very fast learning systems whose knowledge can then be transferred to new systems.

Also, the autonomous, self driving car will apply these kinds of coordination processes to reduce or even eliminate traffic jams. When these cars can communicate with other cars and the road itself they can know the conditions all along the route and optimize or selectively re-route to keep all the cars moving at the best possible speed...

Big Bad Bald Bastard said...

I think the biggest shift in AI/Robotics was realizing that, at least at the early stages of development, robots should mimic insect behavior rather than human behavior.

Personally, I'm ready for the self-driving car. Driving isn't any fun anymore. I'd also prefer a car with a giant rubberized bumper all around its exterior, kinda like a motor launch has tires around it to prevent docking mishaps.

mikey said...

That was a key conceptual breakthrough, but the functional breakthrough that facilitated rapid robotic development was the release of ROS in 2007. With a standardized, open source OS with off the shelf libraries for routine purposes like datatyping, perception, mapping and communication, developers for the first time could start with a functional system and simply add customized capabilities.

And a properly designed autonomous vehicle system won't need bumpers - it will transmit changes in velocity directly to the vehicles around it along with the system management platform, which will in turn instruct all the affected vehicles to speed up or slow down in order to maintain their intervals...

Big Bad Bald Bastard said...

That was a key conceptual breakthrough, but the functional breakthrough that facilitated rapid robotic development was the release of ROS in 2007. With a standardized, open source OS with off the shelf libraries for routine purposes like datatyping, perception, mapping and communication, developers for the first time could start with a functional system and simply add customized capabilities.

The open source movement fascinates me- the best software that's out there is the free stuff, which gives me hope that the Apple/Microsoft oligopoly can be broken.

ifthethunderdontgetya™³²®© said...

Congress is S.M.R.T. See?

mikey said...

I'm in late stage negotiations with an early stage open source Hadoop infrastructure software outfit with a bunch of VC money. I'd honestly love to get back into the open source software business...

zombie rotten mcdonald said...

When I saw the "Mould..." titled video, I was hoping for some Sugar, or maybe some Husker Du....

did you ever see when the Mythbusters tried the metronome sync experiment on a table full of metronomes? Turns out there's a little too much entropy in the system at that scale...