“Self-Organization in Football Teams” – Mark Evans

Download mp3 HTML Transcript

“Self-Organization in Football Teams” – Mark Evans

Mark Evans:
Good afternoon. My name is Mark Evans with Liverpool John Moores University. I am a cybernetician. Very much in the tradition of Stafford Beer, who is a hero of mine, who I’ve been studying for about 10 years.

Here it goes. I should warn you, this presentation, the football angle is an effective low test of the Viable System Model. Stafford Beer’s Viable System Model, here and after referred to as the VSM. It’s not about football. Football is a convenient vehicle for a low test. I come from the very famous city of Liverpool and we have a fantastic football club, but I am not a footballing exponent. We have an overview.

Why? Inside the Viable System Model, system 5. It’s notoriously difficult to extract to the signature for the policy function of a viable system. That’s the purpose of the research. Viability signatures, unknown.

Who? We have two real, opposing football teams. What are we studying? We’re studying the policy function. The transfer function. The control function, if you’d like. Of how they interact, and control themselves with respect to each other. System 5.

Where? Of all places the Scandinavian football league. I’ve developed, or have been very fortunate to obtain an industrial partnership with a technology company that specializes in this field. The dates that I’ve been lucky enough to obtain has been obtained directly from the Scandinavian Football League. The date has been fully anonymized, of course, it is fully verifiable and it is real, and in high volume.

When? It was legacy data. I don’t know when it was taken. That’s not my concern. My concern is the dynamic profiles that can be elicited from the volumes of data, and how we use laser scanners. If you imagine that white slide as a football pitch, you would have a series of laser scanners mounted on tripods about the perimeter. Effectively creating a laser net over the pitch. Technology, I know it’s a superlative, but it’s a very impressive piece of technology. It’s capable of simultaneously tracking all nominative targets on the football pitch, at the rate of 1 reading every 40 milliseconds, which is 1 every 25th of a second, to an accuracy of 1 centimeter. The file that I analyzed was 3.4 million lines long.

Okay, viable systems. I’m mindful of the time so ill accelerate this.

Speaker 2:
Actually, you’re [inaudible 00:02:59] of the speakers.

Felipe Pait:
Is that okay? Very well, thank you. This, and I’m not being disrespectful of Beer, wherever he may be, but this is a stylized version of the Viable System Model. I don’t really have time to go through the theory that underpins that. There’s copious amounts of literature on it. The purpose of this exercise is that viable system is a system that survives within it’s environment. It adapts to its conditions and overcomes them. To essentially sure its identity is maintained in the system. It’s a master organizational template. Any system that requires organizing; business system, computer software system, whatever it might be. Beer himself most famously used it for Allende government scheme in 1973, and was abruptly stopped when the coupe took place, but never mind.

That is the topology of functions, 6 functions. That’s all it takes for any system to be organized effectively. We, as human beings, are viable systems. The model was developed along neurocybernetic lines, and we are by definition, neurocybernetic beings. This has inspiration in human neurophysiology.

When football plays come together, to do what they do, they are viable embedments. They are embedments to system 1. They cause this wider system to be operationalized in its environment. They are what the system does. Yeah? We have a viable systems coming together to create a viable system. Viable systems as people, coming together, synergizing to create an artificial systemic structure. A team that has to be viable against the viable opposition.

Laser tracking. Transmit. Receive. Time 5 seconds; there are milliseconds. Position x, y. Coordinate data two dimensional environment. Another laser ports, 40 milliseconds later. Different reading, same player. For all 22 players, on both teams, simultaneously.

These slides really don’t transition quickly enough. We could say, well here we have a player confronted with opposition on another team. He strategists, that is my predicted path around the objective. That which would seek to thwart my viability on an individualistic basis. I’m moving around the opposition, the obstruction, and I’m taking readings as I go. Player positions as policy, that’s an important consideration which I’d like you to bear in mind as this presentation continues. From that point, we’re using a simple…simple, heavens above…learning loop. Daniel Kim came up with the single loop learning model, which is in [inaudible 00:06:22] book.

Observation of the situation and assessment of that situation, of that observation are design in response to that assessment and an implementation of that design response to that opposition. That’s what I have in mind, however, last slide, that is my tactic. That’s what I want to do. This is what I’m doing, as I’m circumnavigating the obstruction. [inaudible 00:06:53]more hast-less speed.

In terms of the meta-system of a Viable System Model, which is the executive of the system; system 4, system 5, and system 3. System 4 is the intelligent function, the outside and then. Looking out towards the system. Eliciting what’s going on in the environment and making informed judgments based upon it. System 3 is the management, resource allocation, to system 1. What we’re doing, what we’re getting. System 5 is policy, that’s my area of research. It’s that which is really important. In all of Beer’s diagrams, system 5 is usually depicted at the top; system 4 and system 3 underneath. Function of system 5 is to balance varietally or homeostatically. System 4 and System 5 and the policy is what the system does. I am talking to you now, it is my policy. There is a transfer function going on in my brain. I’m talking to you because that is my policy, that is what I am doing. The system is what system does, Stafford Beer, pure and system.

With reference to this, what we’re doing, is we have motor instruction, this is what we’re doing. We’re moving about. The informations coming on. At the same time, we’re making observations; observing, assessing, designing. That’s what we’re wanting. What we’re getting and what we’re wanting create a debate inside the system. That transitions through into policy. Policy is what we do. Policy doing through to doing. Doing sends a signal to the system resources. In this particular case, I’m doing it now, and I’m walking across the room. That’s what I’m doing. It’s my policy. It’s my motor response to a visual assessment design and implementation, again, with respect to a situation.

Next. Here we go. Signal comes in through this path wanting, getting in the case of a footballer, as the person is moving along the pitch. They’re getting feedback from their motor control systems. That’s informing what this overall system is getting, but they’re also receiving a visual signal in here. That’s what comes into here. Creates the synergy. Issues the policy. Sends a signal out to the motors, and the player moves. It’s a simple feedback loop but it’s in continual motion. In some respect, it’s concession to past interaction of actors, but I’ll get to that in a moment.

Here we can see the laser scanners. This is a tactic. This individual is here. At he next time stamp, they’re here; they’re here, they’re here. Purely and simply, I know it’s a little bit small to see, but that’s the mechanism in operation, at every stage of their trajectory through the environment. We take a laser reading at every point.

Here we have, there’s only a few more slides, sorry to appear to be rushed. My theory is that, if they make a decision at one point, that with a view to the next time step. So in other words, the difference between deciding what to do and then doing it, in this system is 40 milliseconds. Their position is their policy. Their position is what they have chosen to do. I am not going to be across that room in 2 seconds. That’s my policy but I formed it over there, but I completed it at this point. That’s what that is trying to illustrate. The policy and the implementation is lagged by 40 milliseconds in this system. The player could be idling. The player could be running, whatever… I think I jumped on there, here we go.

Policy formation lags implementation by 40 milliseconds. Player has Cartesian spatial coordinates. In other words, an x y position on the pitch. Policies formed at each coordinate, both the current position reading is the policy implemented that was formulated 40 milliseconds ago. We can’t get a real time reading. I think 40 milliseconds is a concession to Beer’s point of pseudo real-time. Which is great. Where they are now, at t=5, is because they chose to be there then, 40 milliseconds ago. Yeah? The system is what the systems does. That is actually inscribed on Beer’s tombstone. The purpose of a system is what it does. It’s a small sentence but it is extremely profound. The purpose is what it does. If it doesn’t do what it’s designed to do, it’s epidemically deficient. It’s not achieving it’s objective. As Beer observed, what’s the point? The system does what the system does. A person, equals it’s policy, equals it’s position, in this scenario.

We have a data capture rate, 1 reading every 40 milliseconds. To a spacial accuracy of 1 centimeter. For every player on two opposing teams, in a 90-minute match, 3.4 million lines of data, real data. Not contrived, not pseudorandomly generated; real. Final score was nil-nil. I wasn’t intended that as a joke, to point in fact, because there’s an important cybernetic concept there. I wonder if anyone can get it. The situation is, that the outcome of the match was varietally balanced. They were as viable as each other. They held against each other to the conclusion of the match. If you consider these two opposing federations of homeostats, the whole situation was held, imbalanced by those homeostats continually. Certainly to the point where the 19 minutes concluded, which is the important point.

I’ve got all the data coming through. Fire FTP. Which will allow me to discern any of the profiles for victorious matches, or whatever, by certain levels. For example, a 3-3, would be akin to this. We’d expect the profile to be the same. You could also have marginal victory, or emphatic victory. An emphatic victory would be, say, a 3-nil. Where a marginal victory might be a 1-nil. So it goes and there are other permutations. I’m not interested in the numerical outcome, I’m interested in the dynamics, and what it tells us about viable systems.

As I said, if outcome is variably balanced; we’re nearly there folks, we’re nearly there. Each pay is largely autonomic. Function of a manager inside a football match has no direct control over what actually happens in real time. So how can he control the situation? Answer: he can’t. He can theorize and strategize, but he can’t control the situation. Therefore, in accordance with Cone Attachment Theorem, 1970, doesn’t have a really an adequate model to control the situation. He’s in direct contravention of that rule, or that law.

Dynamic interaction and policy formation, for each individual and each team. Now, this is important, an internalized control emerges. If we were all playing football, we would all be synergizing with respect to each other. I see my colleague over there with the ball. I’m signaling by my actions, please pass it to me. As we’re progressing down the pitch, we’re actually self organizing to achieve our common objective. With respect to proactive opposition that would seek to thwart out rings, and they’re thinking the same thing as the match progresses. The players self organize. The role of the manager not here nor there. Team is a complex, adaptive, viable system. Which is a system, off-switch systems, players as enviable embedments to a viable system. Importantly in this context, policy is the position vector of the player. What I’ve done is I’ve analyzed that using Zipf’s method.

I suggest you have a look at the paper. This is only one side of the results. I’ve analyzed the policy vectors of all the players and as you can see this line is very discernible here. It actually corresponds to the goal keeper who predominantly only moves side to side anyway. Notice the similarity, I’ll get to this side if I may. The linearity across a significant majority of the data, also as well.

A log 2 plot against the log 2 plot, or a log-log plot. If that comes out as linear, that’s indicative of the power law. Power laws are ubiquitous in nature. They’re also a signifier. Take knowledge in the complex adaptive systems community that emergence of power law, presence of a power law, is indicative of self organizing systems. That’s a common signature. Thesis is, I suppose, that power laws characteristics, presence in policy, power laws are a hallmark of self-organization. Policy belongs to viable systems within a viable system as a collective, and if a team is viable is self organizes.

The viable system model, of that performance if that performance of team maps 2, it must also behave the same way. It must display the power law. Policy in system 5, the VSN, is difficult to isolate, but we consider that this provides insight. The viable signature, one concluding comment, if I may, the viable system model rarely hasn’t been adequately low tested since the Chilean experiment in 1973. I think at best, it’s only been tested in financial terms, which is an entire contradiction to Beer’s aims for it. I don’t know anyone who’s actually using Beer’s performance [inaudible 00:17:49].

My focus here is to provide an effective low test in real time of two viable systems that are real, and in opposition to each other. I believe that we’ve managed to go someways to achieving that. There’s other work to do. I have other data sets to analyze and so forth, but the research is looking promising. And that’s it.

Thank you for your attention.

I think we have time for some questions.

Felipe Pait:
No we don’t. (Laughter)

Question 1:
[crosstalk 00:18:26][inaudible 00:18:27]

Felipe Pait:
It worked. The conclusion was it worked.

Question 1:
[inaudible 00:18:38]

Felipe Pait:
No. The in operation, to give you a for instance, when it took them two years what they managed to achieve, and they did a lot in two years; and they regulated the entire socioeconomic infrastructure of Chile, all 3000 miles long of it. They had socioeconomic data coming into Santiago, and into the presidents office, with no more than a 24 hour delay, and this was in 1973, before the popularization of the internet, and moreover, they did it all with one computer.

Now Beer received criticism for centralizing power for Dr. Andy, but that an nonsense because they only had one computer. As Beer would say, they were centralized because they only had one resource. If they had resources they would have used them. They had to resort to using telexes.

A huge country like the United States, or Great Britain, the economic data at the time, certainly, I don’t know about now, it was out of date. In fact, Harold MacMillan back in the 1960’s, that’s a right trying to run the economy in a Great Britain at the time was like trying to catch a train on last years bus time table, or last years train time, rather. They were trying to control a situation on last years data. In Chile in 1973, where this was load tested to a great extend, the last ton, as far as I’m aware. As far as the literature tells me. They did it within 24 hour. There was a report into the presidents office within 24 hours of it being generated, and they were using elementary radio links, microwave, that kind of thing.

It was proven, but not in a dynamical systems context like this. This is real down to the 40 millisecond tolerance. Not a 24 hour period. There’s a huge difference in temperal terms. It’s near real time control. Near. Pseudo real time, as I said.

Anybody else?

Yes? Anybody.

Question 2 :
[inaudible 00:21:04] – [inaudible 00:23:15]

Felipe Pait:
Yes, but the difference between a physical system and this, is that this is a teological system and its purpose is implicit in its actions. They have literally, and figuratively, a direction, a purpose. Through varietal space. Through state space.

I take your point, but that’s not the purpose of this.

Question 2 :
[inaudible 00:23:40]

Felipe Pait:
Please don’t misunderstand me. I’m not dismissing you by any means. Indeed, I looked at entropy, because I’ve been studying complex systems of gender as well. Yes, there’s some very powerful ideas there, but it’s not part of my work. I’m looking at the practicalities of it, and just seeing what I can elicit from the data, and trying to relate that to the viable system as a test case study.

Thank you very much.