I am sick today so I had time to watch two talks on why the interpretation of what humans (want to) thrive for is outdated and needs to change. Both are about 50 Minutes long, but have speakers of different age and background. Both can help in understanding this new idea of how mankind is starting to interoperate, but both are also not a final answer  - they are descriptions of what is in the beginnings and may turn out to happen in this or a similar way. They are also not the only people who think about this theme, of course.

Hans-Peter Dürr (interview is in German) is an accomplished quantum physicist and explains how from that point of view, "Wirklichkeit" is a much better term than "reality". He then talks about how he fought in World War II, worked with Edward Teller and Werner Heisenberg and met Hannah Arendt, leading him to later engagements for peace, more sustainable communities and economies.

Jeremy Rifkin (talk is in english, with convenient subtitles), one generation yonger, is someone who runs think tanks, advises the EU on energy planning and sells books. His latest theme is the "empathic civilisation", and here he is invited at Google to explain it and the challenge it faces during the coming energy crisis. He uses the occasion to tell Google what role he thinks they should be playing. He introduces the empathy-entropy paradox we need to break: As mankind has widened the circle of empathy, the energy footprint became worse and worse.

When we use a common good, we cause the most expense and anger when we all use it at the same time. An idea that is now being tested in several places is to convince a small percentage of people to deviate their usage of the good from this popular time points. It is generally agreed that in many cases,  a change in the behaviour of very few people can already relieve a system of much pressure.

The current consensus seems to be that the system operator should pay people for that. If someone actually has the flexibility to deviate, the opportunity to earn or save money might make him actually identify and use it.

I see this in electricity markets, where demand peaks can maybe avoided by convincing a couple of devices to stop operating for a while and also in traffic management: The city of Utrecht plans to pay a commuter €4 per workday if he doesn't use the highway A2. They will have to devise some solutions where they film the numberplates of passing cars, but building new streets might be much more expensive (not to mention how much trafic jams hurt the economy and the mental wellbeing of commuters).

This can be a very effective way towards more efficient usage of common goods - but there are hurdles like the increasing hunger for usage data that comes along with it. Not many people will like that and we will need to talk about this. The good news is that if the mathematicians are correct, not everyones flexibility is needed.

Update: The Netherlands now plan to tax highway usage for everyone via satellite by 2012 (like the Germans already do for trucks). This system is more flexible, but also harder to accept.

I have been making the optimisation of systems, based on local information, one of my specialities. Such systems are optimised in a decentralised manner (which says that control happens on local levels and decisions about that are made also locally, according to local information). As such they are resilient against the shortcomings and the failure of a central planning node. That's good for users of the system. It also sounds nice. But is it really nice, in daily perception? Let's look at two examples:

On dutch motorways, the control system sometimes imposes tempolimits (say, 70 km/h or 50 km/h) on parts of the roads in order to prevent traffic jams further down the road. So the system knows about congestion problems in the vicinity of you, and might slow you down so you do not make a traffic jam out of it.

Recently some scientists modelled the problematic public transit system in Mexico city. Though buses run according to a plan, they sometimes stay in the stations longer (e.g. when a lot of people have to get in) and so they built up irregularities over a day, which ends up in a lot of buses being in one place and none in most others (this problem even has a name: "platooning"). They concluded that buses should just leave after a short time and not let all attending passengers board. Those passengers left out should take the next bus and overall, in the long run, everyone is better off because the whole system will run smoothly.

All this is true, but from the local viewpoint of the user of such a system, it feels wrong. I cursed when I rode on the dutch motorways and had to slow down for no reason that I could immediately apprehend. And it will definitely suck when a bus driver just leaves with you still standing right in front of the bus stop. We would hate those bus drivers.

I can tell from my own experience in understanding such systems (while researching them) that it would be hard to make everyone understand the situation. On the one hand, it's based on a lot of information and that information is decentralised - it is not all in the same place as a local observer. On the other hand, our brains are also bad at processing such abstract stuff like the platooning problem.

What to do? Would humans be happier with problematic systems, simply because they would feel more in control? Or can we evolve a system thinking in our culture that appreciates such complex, decentralised solutions? Maybe it would help to put out as much information as possible about how these systems work so that everyone can look it up herself. Maybe we'll need to go the extra mile and also visualise them really well. We should have graphically appealing real-time overviews of traffic situations (already in the making), public transports, elevators, electricity grid congestion and so on - for everyone to see and to discuss.

But I also think that in the end, not every optimisation procedure can actually be accepted - it needs at least to be understandable to stand a chance.

Yesterday I was lucky enough to take part in the Seminar of Collective Decision-Making at the Hanse-Wissenschaftskolleg Delmenhorst. Interesting talks about current research, all of them involved socio-economic experiments with human subjects who got real money based on their behaviour. I'll try to briefly summarize all three talks here:


Manfred Milinski

A look at cooperation towards a common good and a stab at representational democracies. Subjects played the "climate game": Everyone has to invest in a common good, or all are at risk to lose all they have. Players start with 40 Euro, and can for 10 rounds invest 0, 2 or 4 Euros. If after 10 rounds less than 120 Euros have been invested in the common good, they all are at risk (here they tried 10%, 50% and 90%) to lose all they have left. I was late to this talk, but I heard that often the players didn't make it to invest 120 Euros, even in the 90% risk case (all of the rules were obvious to everyone).
In another scenario, 18 players formed 6 "countries" a 3 players. Each country elected a representative, based on the strategy (s)he proposed (via chat). This setting was played three times, so that countries held several elections. It turned out that this worked even worse. Representatives who worked towards the common good left their countries with less Euros left, so they were not re-elected. In the next round, representatives all paid too little towards the common good. Sad.
Afterwards, there was a lively discussion about the transferability of this experiment to the real world (someone called the discussion a "snake pit").


Bernhard Kittel
"Coordination and communication in multiparty elections with costly voting"

This research started with modeling a problem in voting theory, but ended with a look under the hood of collective decision making in human groups. In some elections with three candidates, the winner is not preferred by most voters, but the ones who oppose him just split their votes on the two other candidates (think Bush vs Gore and Nader in 2000). Those voters should have been more strategic. In addition, lots of voters don't show up.
In the experiment, voters (the subjects) were assigned payoffs for each candidate, modeling that they all had different preferences. All voters wanted either candidate A or B while they all despised candidate C. In addition, there were costs assigned to actually voting, so that some voters would decide to abstain.
Voters knew what the preference distribution was. The researchers added communication among the voters via a chat before they actually voted. This enabled the voters to coordinate their behaviours in a way that candidate C got elected less often and the voters maximised their individual payoffs.
What then is really interesting is that the chat logs contain the "black box of group decisions". The researchers plan to analyse them in order to find out more about the coordination process of humans. Results are to be expected in a couple of months, but they can already say that
  * the voters who organised most of the coordination got less personal payoff in the end
  * ca. 50% of the people simply stick to their candidate in all rounds, no matter what is being discussed
  * communication among all voters is needed for an optimal outcome
Very interesting. I think that if these chat logs are really a "black box of group decisions", then this data set should be opened for other researchers.


Judith Avrahami / Yaakov Kareev
"Do the weak stand a chance?"

These experiments concern situations were one player clearly has less resources than his opponent. What happens in terms of strategies?
In the "pebbles game", each player has 8 boxes. Player A and player B have 12 and 24 pebbles, respectively, which they can distribute among their 8 boxes. Then, one box of each player gets randomly chosen and who had more pebbles in his box wins. Player A, clearly being weaker, should leave some boxes empty so that he at least has some boxes with a good chance of winning. Expecting this, player B should also make his distribution of pebbles uneven (rather than putting 3 in each).
In essence, inequality in player strengths introduces variability into the strategy space on both sides.
The researchers than got further into the role of the evaluator: what difference does it make that only one box gets evaluated, rather than all? The hypothesis would be that adaptive agents are pushed into trying harder when they know that only some of their work will be evaluated but don't know which in advance. They ran an experiment in which subjects solved addition problems on 6 pages. When the knew that they would only be evaluated according to one page, performance rose.
This managing style got invented to save lazy evaluators time, but it might actually raise performance. I wonder then, if subjects who knew that they were weak in adding numbers also left some pages out to concentrate on the others.

When I run computer simulations, I have to have a solution for the same set of technical tasks each time:

• How to combine variable settings in experiments
• How to store log files nicely
• How to plot nice graphs from those log files
• How to run these expensive computations on remote servers (e.g. university servers)

Now I have one solution for all of this and I am very happy with it:


The above bundle of tasks is universal to a lot of scientists that need to to run computational experiments. I myself will run into this over and over again.

It is a natural reflex of a software developer to build a good tool for this over time and this is what I did. While working on several projects during my Masters, the latest of which is my thesis, I developed scripts for all of these tasks and bundled them together so that you could now call it an application: Combex (Combinatorial Experimentor).

It has turned out to be very useful to me lately and I would like to share this tool with anyone who is interested (this is its home). There are a lot of things that could be even nicer (I already maintain a ticket list), so I welcome contributions.

Note that Combex lets you program whatever you want in whatever language you want, all it wants is that you write log files.

I will let the (sparse) documentation speak for itself and just throw in another screenshot which shows how it all nicely comes together: A (dummy) experiment gets chopped into several tasks and those are shipped to remote servers. I can nicely check if they are done. If so, I have Combex get the data and generate nice plots.

P.S. To talk about our process and find general, reproducible solutions seems to be a general trend in science, see e.g. myexperiment.org.

P.P.S. I am unaware of any other software that offers this task bundle, even if its commercial. Anyone knows software like that? I only found quite specialised approaches, but what I like about Combex is that it doesn't care what the hell your code is doing as long as you configure variables and log data. It is more of a very simple workflow with nice tools along the way. In principle, a lot of other software (say, statistical processing) could be hooked into this as needed.

P.P.S. I am still open to be convinced by a better name for this.

I recently accepted a new job. From October 2009 on, I'll be a junior researcher at the CWI (Centrum Wiskunde en Informatica), a national research center here in Amsterdam. I'm very happy to be working with Prof. Han La Poutre.

I will actually be a Phd student but do research in a predefined context - nowadays, a lot of Phd positions in the exact sciences are a mixture of research work and research exploration. An exciting part about this project is that it is about a very real challenge and there are many stakeholders involved whose views on the problem are important. I will do fundamental research on dynamic and adaptive multi-agent systems, but I will also be busy wearing the hats of different stakeholders and try to bring their view into the model.

[image via http://www.usgbc-centraltexas.org]

The project I'll be working on is called IDeaNed (Intelligent en Decentraal Management van Netwerken en Data, only dutch description so far). It deals with one of the big challenges we face in the next years - a redesign of our energy distribution networks.

As we begin to exit the age of fossil energy, energy will soon become a problem for us, not a cheap and abundant catalysator for progress. Now, as an AI researcher, I can't help out where it is most important - say, actually invent an effective renewable energy source or useful, longlasting batteries. But what we need for sure is a new energy infrastructure. There is already a term for this - the "Smart Grid"*.

Our existing energy networks are of an old, centralised design which is not suitable for the way we need to look at interconnected things. As the report "Grid 2.0" (PDF) puts it:

"It is hard to make the link between flicking a switch and the distant power station that made it possible to turn the light on.
(...)
Partly as a result of this lack of a feedback mechanism, and partly because of technological constraints, Grid 1.0 is surprisingly inefficient. Only around 40 per cent of primary energy input (coal or gas) used in power stations is converted into usable electricity, the rest is wasted heat. A further nine per cent is lost as the power moves through the transmission and distribution system. Then a further third is lost in our homes and offices because they are poorly insulated, not designed with energy in mind, and inhabited by people who do not see themselves as players in the energy game."

We need a network that is decentralised (can work with input from several local sources, not only big central plants, can communicate locally about sharing loads) and avoids efficieny problems in peak times. Such a net needs real-time price mechanisms and needs to accomodate a lot of players: customers want to use effectively, producers want to sell efficiently, governments want to distribute evenly **. Meanwhile, there is a real, non-abstract component: the network. Voltage capacities at different parts of different sub-networks dictate what is possible and what is expensive to do. A lot of these things change all the time. This can get very complex very fast.

What we'll be doing is modeling Multi-Agent systems in order to learn what good price mechanisms are and what good automisationable strategies of the local players could be. We will get input from another Phd student from Eindhoven supplying technical findings about the infrastructure and we'll work with companies in the dutch energy market (big players and consultancy agencies that  develop energy network simulation software). This highly integrated approach of designing scientific projects is a unique dutch approach and I am quite excited about it.

This topic is not only being picked up in the Netherlands. Big companies like GE or IBM are already promoting their competence in this topic which isn't even entirely understood yet.

They smell money in creating a market better targeted at our actual energy usage and in savings due to efficiency. This might be one of the rare cases where what they want and what society needs have a significant overlap. For instance, I recently learned that energy efficiency on the last mile is multiple times more effective than energy efficiency at creation time (due to all the energy losses by conversion or distance that already accumulate until the last mile). And it is easier to do.


* There might even emerge a bubble with all the money being thrown at this topic right now by governments, but I hope that in 4.5 years, when I finish this project, things will have come to terms, the crooks took their money and left or were thrown out and we know where to go.

** Especially when there is not enough. In a post-fossil world there may be "bad weeks" with few energy available. Or, in more grim scenarios, we'll have few energy at all times and almost none in "bad weeks". Making sure distribution can be fair is crucial and not as easy as it first sounds.

I am reading reddit.com for the interesting links, but also for interesting conversations in the comments (I still have to learn to skim over the trolls, though). A special kind of conversations came up recently. People would claim that they are interesting or different in some way and invite peoples questions.

I remember that it started with a guy saying he is hetero, but works at a gay sex phone line to support his studies. If anyone wanted to know anything about it? People wanted. Some girls said she has epilepsy. Any questions? Sure.

Then, someone had the idea to make an own subreddit for these kinds of conversations (a reddit-style forum). It's called I am a .... Currently, it's pretty lively. Some people can be asked about being french or from Afghanistan, others have strange jobs like Lobbyists or are illegal Immigrants.

Every scientist needs to have a collection of papers (s)he reads and might use for reference. For the last research projects, I used a desktop program called JabRef that was decent. Now, I switched to something really promising.

Mendeley is a startup that wants to do with research papers what last.fm does with music: Make it easy and fun to find some that you don't know yet but would probably like. This goal is really, really mouth-watering, since the amount of papers out there seems to grow exponentially. Today, there is no other way of knowing if someone else already did what you want to do by following references from papers and maybe keyword search on Google Scholar. It's a growing concern for researchers.

They want scientists to have a profile on their site and fill a giant pool of data by citing their research interests, uploading and tagging a lot of papers and maybe build social connections. Then, recommendation becomes possible.

Of course, they'd have the hen-and-egg problem every social community project has. Value can only be created when people are there in big numbers, and people only come when value is already there. This is why the Mendeley people decided to start with solving some other problems scientists have, so that users have a reason to come:

• They built a really slick desktop client to organize papers. It runs on every OS, does full-text search and extracts meta data from PDF automatically if you want. I think it's already better than JabRef.
• The client also syncs automatically with your web portfolio. You can have your library with you wherever you want.
• You can open a research project and share papers only within that research group.
• You can always import and export your library (for instance to Bibtex).

They said recently that users have uploaded one million papers already. That's a lot of potential for data mining right there. And did I mention that they have a Last.fm investor on their board?

I just watched the first part of "The Trap: What Happened to Our Dream of Freedom?" by Adam Curtis. He claims that our basic notion of human behaviour has been crooked ever since the cold war, leading to inhumane psychology and selfish politics.

His main starting point is that in the cold war, the two big military forces faced the double contingency problem of how the other party might think of what they were thinking of them themselves (and so on) - all in the realm of nuclear destruction. The USA threw a lot of scientists at the problem and they invented Game Theory. This concept, in which people are regarded as only selfish and numbers can be put on behaviour, got picked up by psychologists, economists and politicians, slowly taking over the way society looks at human nature.

I don't disagree fundamentally with the basic historical notion here: In the 50s and 60s, Game Theory still had a rather simplistic view on human behaviour, portraying humans as only self-interested. Later, western societies went through a lot of change but left the discussion what constitutes freedom in the realm of rationalism. I hope we're making some progress to develop this idea of what makes us happy and free, but this discussion has been stuck for some years now. In the meantime, societies changed a lot, sometimes built on a much too strong view of humans as being rational. It makes sense to discuss what went wrong because of this.
I agree with that, but I think, Curtis overshoots with this movie. Here are some points of criticism:

No History context
For starters, Curtis bends history so that it fits his story. Game Theory wasn't invented during the cold war, but already in the 1930s. The view that humans are rational beings is also much older.
More on context: Curtis recognizes that western societies needed to change in the 50s, 60s and 70s. Psychology until then was really, really inhumane, Families were often little dictatorships and government bureaucracies (in his movie the British) desperately needed reform. Curtis never addresses the view on human behaviour that was in place until then (I think it was close to the view that you can make people do whatever you tell them to).

But any new concept should be explained in the light of the concepts it was made to replace. What was the general idea that the rational model replaced? Were there other good solutions on the table at the time than to try to view people as rational? Maybe it wasn't only scientists who placed this concept in the mind of society. Maybe it was time to look at people that way and maybe this was already much better than the view before.

Crazy Scientists
Curtis portaits proclaimers of Game Theory ideas (in which people are only selfish) as crazy and old. That is necessary, because today, everyone younger than 70 who works Game Theory wouldn't look at things in this one-headed way. For instance, when using the Prisoners Dilemma, the only modern models I have seen ask the question: In what settings does cooperation emerge? (Axelrod first made this new notion public 29 years ago). It's always some crazy scientists misleading everyone else. For example: Buchanan, an econmist, is guilty for Thatchers politics because she invited him to talk once.

I understand that you can't show all context when you shoot a documentary, but Curtis is destined to blame a dozen people for all that went wrong regardless of what was there before, and that smells of conspiracy nut.

Causality
He claims causal links that I can't follow. For instance, here is the gist of 10 minutes of the movie: The psychologist Laing critized old-school Psychology in America  (somehow, hew was of course inspired by Game Theory in the beginning of his work). As a consequence, American Psychology turned to using automated, oversimplistic questionaires and tested a lot of people for mental problems. Then, it turned out that every second American has some history of mental problems. As a consequence, Americans were oversensitive for the idea of what "normal" is and since then ask their Analysts to turn them into this new strange notion of normality. That is one hell of a causal chain, proving that Americans running to Analysts is also a consequence of Game Theory.

Here is my theory: Society is a compicated organism. Sadly, things take time. Good ideas take decades to develop into mature concepts that everyone has understood.  And it seems to be a pattern that they  cause a lot of damage when people carry early versions of them into the real world. See Game Theory or also the influence on early physics ideas on economics. That doesn't mean you can't make the theories better, like it happened to Physics and I believe, also Game Theory. Of course, if certain people already are happy with the dangerously simple ideas and cause damage, we sometimes have to wait until they are somehow out of power. They say that old ideas are only gone when all of their proponents are dead.

But enough of that - back to Curtis. Sometimes, I got the feeling that he really agitates against using numbers for anything that could describe human behaviour. Fine, that is something to discuss, but then he made the wrong movie.

P.S. This critique makes some of the points I made more clearly.

I am currently reading The Origin Of Wealth - a book which tries to explain in what deep explanatory troubles classical equilibrium economics have gotten into during the last two decades.

One of the main messages is that economic systems are no closed systems that could theoretically reach an equilibrium - they are highly dynamic and interactions are so complex that they will only -suddenly- reach an equilibrium when all involved players are dead.

This simple drawing I made should resemble some of the most basic physical classification of systems and where economic systems belong. The author says that classical economics would want to place an economy in the right branch and therefore never have a model close to reality.

After economics borrowed the notion of an equilibrium from physics (roughly 100 years ago), physics moved on and discovered entropy and the second law of thermodynamics. Chaos and Complexity have now been discussed fundamentally by almost all basic sciences, but seldom in economics.

Later on, the book discusses that in an open system like an economy, the creation of (temporary) order is what we call "wealth creation" and that this happens by evolution-like processes (the best system/solution/product replaces others):

I am interested to see how the author defines a "complex adaptive system" (which is his own term) later on...

My last post concerning cooperation  was not about any research that I would personally do, but highlighted the best and most useful results concerning the old question why there is cooperation that I have come across. I am studying Artificial Intelligence (AI). Today, I want to put on the shoes of an AI researcher trying to tell other researchers why research in cooperation is a good idea.

Cooperation as an AI topic
A natural question is: Why should someone in AI spent precious research money on cooperation? After all, to explain why there is cooperation is generally a domain of the sociologists and biologists, maybe economists. AI is based on Computer Science and while it surely should try out inspirations from other disciplines, its main purpose is still to build things that work in a new way (initial goals of building something entirely new have been refined). More and more (due to a lot of fruitless approaches and now also due to the economy), I hear the question why any approach would be useful to pay for. What can the new thing do or show that lets people do things better (for instance more efficient) than before?

I think that cooperation is a fruitful theme in this context and want to explain why.

Autonomous agents
An important goal for the future are intelligent agents that do tasks on behalf of us. We don't want to tell them too exactly what to do (it's work and we might give bad instructions) and want them to interact with each other in the world. This is why they need to act autonomous. They also need to be reactive in an unpredictable environment which for them mostly consists of the actions of other agents.
My simple point is that the notion of cooperation is a good tool to model this. First, actions of other agents can mean a lot of things to me. But if I was pressed to put it into really simple terms, I could label those actions as being good (cooperative) or bad (defective) for me. It's a simplification of the world, sure. But we have to start somehow*. With a modeling tool like the Prisoners Dilemma I can already model that agents are autonomous and depend on what other agents are doing. That's already some modeling effort covered, even in a way that researchers agreed upon long ago to be a standard method. I can still play around with some settings though, like the utility values or the number of involved agents per interaction.


Efficient Systems
When I say said I model cooperation, I mostly get looks that imply I'm being labeled a "Hippie". But in reality, cooperation means efficiency. Cooperation makes sense if the outcome of it is superadditive. This means that the utility produced together is more than just the sum of the utilities the agents would have produced alone **. It is also safe to assume that when one agent defects the other, he takes home a lot more utility, but the overall utility is still less than if both had cooperated. So for the system performance, it would be great if agents decided to cooperate often. This should be sold as a hard fact more often: Cooperation makes systems of autonomous agents efficient. It makes sense to do research with this goal in mind.
One thing about complexity and superadditivity: Not all interactions are superadditive, of course. But I think that when the behaviours of agents become more advanced and complex, their interactions in multiagent systems will be superadditive more often (since superadditivy often arises when noone is an expert for everything).


Multiagent architecture - Cooperation built-in?
Researchers come up with design guidelines for multiagent systems quite often. That cooperation is an essential design guideline for decentral systems with autonomous agents is sometimes already part of this. Take for instance the AMAS architecture. There is a subheadline called "Self-organisation by co-operation". They assume that each agent always tries coopreation and thus the system becomes efficient. I think that is a little simple. Take humans - for us cooperation is natural and feels right. But we don't have to cooperate. In fact, we often don't. If there are agents in the system that defect, then let that happen. Work around them, if you can. Defect them back if you deal with them or maybe see if they change behaviour.
Multiagent systems can - like nature - be heterogeneous and then it doesn't make sense to assume some behaviour for all agents.

Outlook
Cooperation is one way to start modeling complex interactions. There are simple models to use that everybody understand and we can agree that cooperation makes systems efficient. However, it is important to note how we mostly talk about the question if an agent wants to cooperate with another. Much more complicated is the how later on.
How do two agents cooperate once they both try? Can we make any general models for this at all? There is so much context involved and so much circular dependencies, so much source for uncertainty. I hope we can find a simple model for this that can be as usable as the Prisoners Dilemma.

* And we have to talk about it. To communicate about science is really important and to use simple and accepted models helps big time in this.
** You can nicely model such a situation with the Prisoners Dilemma.

I got my new MacBook some days ago and am very happy. Especially having a built-in webcam feels very much like finally being in 2009. Visiting my friend Marcel in Osnabrück, I tried out Photobooth, a nice app to make pics with the webcam. It's great fun, not only for kids:

Here are our best shots :)