Vilfredo vs Parkinson Law of Inefficiency.

[Pietro Speroni di Fenizio]

There is rule of thumb in social system that states that Committee should not be too big or they will not achieve much. Actually it was first stated by Parkinson. The same that said that: 

Work expands so as to fill the time available for its completion.

 

And Parkinson found somewhere around 20 the limit of how many people should a committee be, to still be efficient. The reason why this is of interest to us is because if Vilfredo could break this limit, and let a bigger committee work efficiently then it would become really useful. While Parkinson statement was only semi serious, there are more serious work done recently: Klimek, Hanel, Thurner: Parkinson’s Law Quantified: Three Investigations on Bureaucratic Inefficiency. And again the same size comes out. The paper is more complex, and I have not had the time to read it fully. It also considers the pension time, and speed of raising in the corporate ladder. All things we do not consider at all in Vilfredo.

But my question is this one: Vilfredo finds a limit somewhere around 20 people. With 30 being generally above the limit. The same limit that those other studies found. Could it be that the two limits are connected, and maybe expression of the same hard limit?

When Vilfredo started I expected it to be able to scale. It was only in the Milan event when we had some 17 people that I realised that it was not going to be as easy as I thought. If people were to vote at random, the Pareto Front would expand very rapidly, very soon. Instead, I reasoned, as people are more similar between them, the Pareto Front will only grow up to a point. Boy, was I wrong. Yes people are more similar. No, the Pareto Front does not hit any hard limit, but keep on growing. At least for what we could observe. Then I thought, maybe this is because people do not pay attention when they vote. So we introduced Key Voters, asking people in key positions to reconsider their vote. If they voted badly they would now change their vote. We also asked people to comment why they voted against a proposal. This DID made things a little bit better, and we managed to be able to handle bigger groups. But not as big as we hoped. The fact is that while it is true that people vote not always consciously. It is also true that they are also enough diverse between them to let the Pareto Front keep growing. We just slowed that growth.

So is it all lost? I don't think so.

First of all we might have pushed the limit a little bit. If we can handle bigger groups than traditional committee, still the work is valuable.

Second Vilfredo, by computing the Pareto Front is showing clearly where everybody stands. It might be that this diversity among the various position is ultimately what is letting large committee fail. But it is a lot of work to handle big groups in reality. If this can be done by a computer, we can have groups of 25-30 people easily. And this is helpful.

And finally, not all committee need to decide unanimously. If we cross the Pareto Front with the popularity information, we might be able to make a much more thorough investigation over what are the possible alternatives. We will still decide using majority, because we cannot find an unanimous alternative. But the whole process will be transparent. And it will be clear, proven, that the result is the best solution this committee could come up with.

Here are two pictures from the latest experiment. The first one is from a moment when 30 people participated:

You can download the graph here (for some time at least): http://vilfredo.org/map/607333790821f3393cf648b20d0a10a4.svg

This one is from the same question, but a moment when only 18 people participated:

this can be downloaded from: http://vilfredo.org/map/3d0c6247198129cb01ab99959b73a68f.svg

Still a complex graph, but a level of complexity we can cope with. The first one can only be handled by a computer or by a centralised human that imposes some top down simplifications. The kind of results we don't want.

[Victor Vorski]

Btw. I think even if the tool only works naturally for groups of up to 10 it can still be incredibly useful. I would venture to guess that most human decision making is done in groups of under 10 people.

[Pietro Speroni di Fenizio]

Indeed most decisions take less than 10 people, but I wonder if this is because so few people are necessary or because it is clumsy to decide in more, or for other reasons. If the group is very small it usually does not need Vilfredo to decide. Although Vilfredo would make sure that all point of views are heard. And I find the graph VERY explicative in what is the relative positions of the people. Somehow it makes it all obvious. While if not people can hide behind silence, not to offend anyone. But that does not help.

[Victor Vorski]

"If the group is very small it usually does not need Vilfredo to decide."

But would using Vilfredo make the decision process more efficient or better? Also giving documentation to how a decision was arrived at is very valuable in many circumstances.

[Victor Vorski]

"Indeed most decisions take less than 10 people, but I wonder if this is because so few are necessary or because it is clumsy to decide in more, or for other reasons. "

It takes more energy to gather and organise a bigger group for sure. we have lost our communities where we were truly democratic. Top down has become the default organisation mode...

Changing the ease of decision making in groups could thus have a huge effect on what organisation mode is chosen by default...

[Pietro Speroni di Fenizio]

Hi Victor, thanks for stepping in.

It surely would make it better in the sense that the result is what the whole group want after considering all the options. It might be not the best option, as it could be beaten by a lone genius unable to explain himself, who happens to have top down power. In other words a community decision can be better than the community creating it. But not much better. If people do not understand a topic, and there is a need to have a technical decision being taken, and the people who do not understand are not willing to step aside but instead vote randomy, then the result is not that good.

Victor, you mentioned me that you wanted to use Vilfredo for a small community you have. Do you think you could expand on that? Maybe in another thread?

[David Bovill]

Have some time now to respond to some issues. The point you raise here is a nice explanation to my earlier question about why you think Vilfredo does not scale - is there an real mathematical / theoretical reason behing this or is it simply your experience of using Vilfredo? From your post below it would seem to be based on practice, not theory, and I thank you for a very clear and honest exposition.

My take on your experience would most definitely not to draw the lessons you draw. Yes Vilfredo does not scale, but not because of any hard wall / fundamental reason. I have had similar experiences to you, but I feel the reasons while real, and hard to overcome, are much more trivial, than deep mathematical or philosophical reasons.

First on the point about Parkinson Law - this seems to be an observation without an explanation - the observation is common sense, and a fundamental critique of the cooperative movement (not as practised but as preached), or direct democracy. I think it is worth reading Dunbar. 1997. Grooming, Gossip and the Evolution of Language' - not for the headline idea that evolution has resulted in a limit to our cognitive ability in terms of cooperating with other individuals to around 150 - but the more nuanced argument which from memory includes a survey of group size, from a wide range of military units, to company and political organisation. I distinctly remember that there is a common group number of around 12-15 as a smaller unit and 150 for the larger organisational structure.

For Dunbar this is about biological RAM. I suspect there is a point here - but would expect there to be a much more significant contribution with regard to comparative costs of collaboration and a mathematical tipping point where if the cost rise very slightly over a threshold we get a type of colony collapse disorder :)

In terms of Vilfredo - even basic experiments with teh drawing algorhythms provided by graphviz would lead you to the same conclusion - not because there is something fundamental with regard to human communication going on here - but simply because the graphs suck. To illustrate the point try and find any graph, using any software algorithm that creates a meaningful layout for more than 20 nodes (and less than 100 nodes). I've been looking at this problem for 15 years and not found anything much that does not use hand drawn simplification. Layout algorithms seem promising for small graphs, and very large graphs - my conclusion was that this is a visual design problem, not a mathematical, or social / cognitive problem. I believe the solution is better graphic design, and shifted my interest towards the area of augmented graphing tools which allow the author to "paint" their argument in social space - there is a place for algorithm - but not in the way that I feel Vilfredo currently places them in this problem space.

--
--
You received this message because you are subscribed to the Google
Groups "Vilfredo Develops in Athens" group.
To post to this group, send email to
vilfredo-deve...@googlegroups.com
To unsubscribe from this group, send email to
vilfredo-develops-i...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/vilfredo-develops-in-athens?hl=en
 
Vilfredo Develops in Athens is the mailing list for the developers of Vilfredo goes to Athens. The website can be found at
http://Vilfredo.org
the code can be found at
https://github.com/pietrosperoni/Vilfredo
On twitter we are @Vg2A
---
You received this message because you are subscribed to the Google Groups "Vilfredo Develops in Athens" group.
To unsubscribe from this group and stop receiving emails from it, send an email to vilfredo-develops-i...@googlegroups.com.
To post to this group, send email to vilfredo-deve...@googlegroups.com.
Visit this group at http://groups.google.com/group/vilfredo-develops-in-athens.
To view this discussion on the web visit https://groups.google.com/d/msgid/vilfredo-develops-in-athens/24599142-cfb1-4c9d-afc5-d5655fb4fdff%40googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Pietro Speroni di Fenizio]

Hi David,

if it was just a matter of human social RAM then Vilfredo would be in the perfect position to make a truly great advance for the whole of humanity. Let's hope it is so, but I don't think so.

Also I disagree that it is only a matter of how badly the graph get represented. Yes, I am convinced there are purely mathematical reasons why the system does not scale. Even though there isn't a hard limit. If you take n people voting randomly over m proposals you will find that on average the Pareto front will be of P_{n,m}%. Now, although I have never measured that P, I am convinced that as n grows, P approaches 1. This could easily be proven. What cannot be proven and what I have observed in the tests we did was that even with agents not voting at random P tends to grow with n approaching 1. What does this mean? That as n grows the filter becomes less and less efficient and it filters out less and less proposals. No there is no hard limit. No it does not scale. Yes, it does not scale for mathematical reasons that have nothing to do with the graph (you notice I never even spoke about it).

I shall make the graph of how the pareto front grows with the growth of n, one of those days.

Cheers,

Pietro