Books : reviews

Michael A. Nielsen.
Reinventing Discovery: the new era of networked science.
Princeton University Press. 2012

rating : 2.5 : great stuff
review : 17 April 2013

In this book Nielsen sets out his argument for a new collaborative, collective, open way of doing science. This new approach is facilitated by the internet. Others have decried the effect the Web is having, but Nielsen says that's because we don't know how to use it to betst advantage yet.

p20. Complex technologies, especially, often require considerable skill to use well. Automobiles are amazing tools, but we all know how learner drivers can terrorize the road. Looking at the internet and concluding that the main impact is to make us stupid is like looking at the automobile and concluding that it's a tool for learner drivers to wipe out terrified pedestrians. Online, we're all still learner drivers, and it's not surprising that online tools are sometimes used poorly, amplifying our individual and collective stupidity.

In arguing his thesis, he starts some fascinating examples of how a networked collaboration surpassed the best experts: a chess match with a team of players collaborating against Garry Kasparov, and an online collaboration to develop a mathematical proof. The chess match is particularly interesting, because Nielsen ses it to highlight the idea of "micro-expertise".

p26. Chess is so rich with possible variations that many of those players had their own individual areas of microexpertise where they too equaled or even surpassed Kasparov. ...
    ... the World Team had such a diverse collection of talent available that each time a problem arose, a member of the team rose to the occasion; someone with just the right microexpertise would leap in to fill the gap.

This forcefully reminded me of a cartoon that Jack Cohen has often shown during one of his talks, which shows a gathering where a professor is being introduced as an expert on crocodiles eyelids; he modestly demurs, saying he is only an expert on the upper eyelids.

single and multiple microexpertises One micro-expert is not enough. They would out-perform the expert in only one small place, leaving the expert to dominate everywhere else. But sufficient diverse micro-expertises, can combine to dominate the expert everywhere. The "expert" is then revealed to be merely a "Jack of all trades; master of none".

However, this simple linear patchwork combination of expertises does not capture the whole picture Nielsen lays out. There is an extra facet to a collaboration: other people's ideas spark new ideas we might not have had on our own, sparking new ideas in them, with a positive feedback that amplifies the overall effect.

p30. When we attempt to solve a hard creative problem on our own, most of our ideas go nowhere. But in a good creative collaboration, some of our ideas-ideas we couldn't have taken any further on our own-stimulate other people to come up with daughter ideas of their own. Those, in turn, stimulate other people to come up with still more ideas. And so on. Ideally, we achieve a kind of conversational critical mass, where the collaboration becomes self-stimulating, and we get the mutual benefit of serendipitous connection over and over again.

What is needed is the tools to harness and coordinate the relevant micro-expertises. This is where the internet comes in.

p32. the problem of amplifying collective intelligence is to direct microexpertise where it will be of most use. The purpose of the online tools is to help people figure out where they should direct their attention.

Nielsen analyses a range of collaborative internet projects, including Linux, Galaxy Zoo (a collaborative effort to classify images of galaxies), and Foldit (protein folding and other puzzles). He identifies:

p48. ... four powerful patterns that open source collaborations have used to scale. (1) a relentless commitment to working in a modular way, finding clever ways of splitting up the overall task into smaller subtasks; (2) encouraging small contributions, to reduce barriers to entry; (3) allowing easy reuse of earlier work by other people; and (4) using signaling mechanisms such as scores to help people decide where to direct their attention.

This networked structure has changed the way at least some disciplines work. So, in programming, for example, there is no longer the "heroic" single developer, reinventing everything from the ground up. Instead:

pp.58-59. Today, a great programmer isn't just someone who can quickly solve a problem from scratch. A great programmer is someone who is also a master of the information commons, someone who, when asked to solve a problem, knows how to quickly assemble and adapt code drawn from the commons, and how to balance that with the need to write additional code from scratch. Such a master can build on the work of others to solve problems faster and more reliably than other lesser programmers.

This new style leads to another concept: that of the "micro-contribution", where an individual's contribution to the overall project may be just a single line of code in an Open Source project, or a single line of text in a wikipedia article.

This all sounds marvellous, but does it really mesh with our experience of how the world works? Groups of people in committees and the like are notoriously bad at collective intelligence. ("A committee is a life form with six or more legs and no brain."):

p72. [groups tend to] focus on knowledge they hold in common, they focus on knowledge held by high-status members of the group, and they often ignore the knowledge of low-status members of the group. Because of this, they don't manage to convert individual insight into collective insight shared by the group.

Given this, what distinguishes the examples that Nielsen describes, where the collective effort demonstrably does work better?

p75. [they] are all examples of problems where there is a shared praxis.

That is, they are the kind of problem domains that have shared knowledge and techniques, that have a shared recognition of the right answer, and a shared agreement on that fact. Nieslen argues that if there is a shared praxis (such as in chess, mathematics, and science), then it is possible to have scalable collective intelligence.

Having discovered which problems can benefit from amplified collective intelligence, and talked about how collective tools can aid this process, the next obvious question is: will anyone bother? Is there enough effort available to perform the scaling? It takes a lot of "citizen science" effort to classify all those galaxies, fold all thos proteins, and so on. Or does it?

p153. The English soccer club Manchester United seats 76,000 at their home stadium, Old Trafford. Games take two hours, with stoppages, so the spectators at a game are spending roughly 150,000 hours of time in total, nearly a third of the amount of time the Zooites have spent classifying galaxies!

So there is plenty of effort available. The total effort invested in one major football match (not including the travel time, or all the people watching on TV) is the same order of magnitude as an entire citizen science project that has made multiple discoveries, and will make more.

p151. the Zookeepers have recently used the Zooites' galaxy classifications to train a computer algorithm to distinguish between spiral and elliptical galaxies.

So Nielsen has established that certain problem domains, including science, can benefit from scalable collective intelligence. This promises great benefits to the advancement of science. But there is a cultural problem. Open science requires a commitment, and effort, to providing infrastructure, from collaborative tools to open scientific data. Yet that effort is "invisible" to the traditional scientific evaluation process, which rewards only one specific form of openness: publications.

p186. the immediate obstacle to open science [is] a culture that only values and rewards the sharing of scientific knowledge in the form of papers.

So the final part of Nielsen's book is some suggestions of how the required culture change can be achieved. We can each play our individual part by making sure we credit infrastructure and data sources, but it will also require some top-down and collaborative initiatives. It won't be easy, but the benefits will be well worth it.

This is an important book, making crucial observations about how the practice of science can benefit from networking and openness. It is also very well written, and along the route of the argument is peppered with fascinating little insights, such as how the digital and linear DNA can encode complex shapes:

p145. The DNA-protein connection is Nature's way of making easy the seemingly impossible task of copying complex shapes.