Book summary “Reinventing discovery”: summary

Introduction

You face a frustrating reality when tackling complex scientific problems using traditional, isolated methods. Progress often moves too slowly because institutions lock away data and researchers guard their discoveries in silos. You find that current academic practices fail to harness the vast intellectual capacity of the global scientific community.

Michael A. Nielsen argues in Reinventing discovery that networked online tools and open collaboration can radically accelerate scientific breakthroughs. You will discover how collective intelligence transforms research when communities share data, modularize tasks, and build open platforms.

The book explores online attention architectures, open source collaboration patterns, and citizen science initiatives. You will also examine the cultural barriers and incentive structures that currently hinder open science.

• How can online networks solve hard problems faster than traditional labs?
• What collaboration patterns scale effectively across large groups?
• How do you overcome academic incentives that discourage open research?

1. Amplifying Collective Intelligence

Online tools and massive collaboration can solve complex problems much faster than traditional isolated research. Improving the method of scientific discovery matters far more than making any single isolated breakthrough.

Galileo observed the moons of Jupiter and Newton formulated gravitational laws during the 17th century. Over the next twenty years, online tools will transform science just as those early methods did.

The Polymath Project

Mass online collaboration can solve extremely difficult, unsolved mathematical problems much faster than traditional isolated research.

Mathematician Tim Gowers launched the Polymath Project in January 2009. Seven hours after opening the blog, no one had commented.

Over 37 days, 27 people wrote 800 mathematical comments containing over 170,000 words. They successfully solved a hard mathematical problem by using the blog as a short-term collective working memory. Diverse minds built upon partial progress and applied new cognitive tools.

GenBank and Citizen Science

Centralized online databases integrate data from thousands of researchers. Computer algorithms then find hidden meanings and correlations across vast fields of knowledge.

Scientists worldwide upload biological data to shared repositories like GenBank, which launched in 1982. The human genome was completed in 2003, and the HapMap mapping human genetic variations was completed in 2007.

Online citizen science projects leverage large groups of untrained volunteers to classify vast amounts of data. Robotic telescopes automatically capture data that professionals lack the time to analyze by hand.

Galaxy Zoo recruited over 200,000 online volunteers to classify galaxy images. They produced over 150 million galaxy classifications, leading to the discovery of green pea galaxies and a quasar mirror, plus 22 scientific papers published in three years.

Mandatory collective agreements tied to funding are necessary to overcome individual career incentives. The 1996 Bermuda conference and the March 14, 2000 joint statement by Bill Clinton and Tony Blair made immediate data sharing the mandatory norm.

Fierce competition for rare university positions drives young scientists to work eighty-hour-plus workweeks. Academic reward structures focus entirely on the conservative goal of writing scientific papers, penalizing time spent on public goods like Wikipedia.

2. Restructuring Expert Attention

Expert attention is the fundamental scarce resource in creative problem solving, and online tools create value by restructuring it. When you design an open architecture of attention, you connect problem posers with the right experts who can use their expertise in high-leverage ways.

Online architectures of attention harness latent microexpertise and transform lucky coincidences into designed serendipity.

ASSET India used InnoCentive to post a challenge for a solar-powered wireless router, putting up twenty thousand dollars in prize money provided by the Rockefeller Foundation. The challenge received four hundred downloads in the two months it was posted, resulting in twenty-seven solutions submitted. Zacary Brown possessed a massive comparative advantage and solved it quickly.

Harnessing Latent Microexpertise

Large and cognitively diverse groups contain vast amounts of ordinarily latent microexpertise that can be uncovered to match elite performers. Garry Kasparov reached a peak rating of 2851 in 1999 and far outranked Irina Krush. Individual contributors nonetheless possessed narrow areas of microexpertise where they equaled or surpassed him.

In 1999, Garry Kasparov played against 50,000 people from 75 countries in a chess game organized by Microsoft. Coordinated significantly by 15-year-old Irina Krush and Smart Chess, the World Team played one of the strongest games in chess history before resigning on move 62. Key moves like move 10 and move 26 were contributed by Irina Krush and a junior player named Yasha respectively.

Designed Serendipity and Division of Labor

Scaling up creative collaboration turns occasional lucky coincidences into a routine mechanism called designed serendipity. Encounters with the exact right person stop being rare when problems are exposed to a broad range of expertise.

Albert Einstein worked on general relativity from 1907 to 1915, getting stuck trying to connect gravity to a nonstandard type of geometry. In 1912, his friend Marcel Grossmann directed him to the pre-existing work of Bernhard Riemann.

Online collaboration utilizes a dynamic division of labor rather than the static division found in conventional organizations. The 2009 movie Avatar employed 2,000 people in fixed roles, contrasting with the fluid participation seen in open projects. Online tools absorb and expand the capabilities of markets, organizations, and small-group conversations.

3. Patterns of Online Collaboration

Open source methodologies scale effectively through relentless modularity, microcontributions, radical reuse, and automated scoring signals.

You can apply open source as a general design methodology beyond programming to any project involving digital information, such as architecture, art, and biology. Sharing digital designs freely allows innovative ideas to spread more quickly and undergo rapid iterative enhancement across diverse fields.

Cameron Sinclair and Kate Stohr launched the Open Architecture Network in 2006, an online community for open source architecture containing over 4,000 projects by early 2010, including a primary school design for Gando, Burkina Faso.

The Open Source Model

Open source software projects rely on making code publicly available and encouraging widespread community contributions to build software far more complex than any individual could create alone. These projects feature two main attributes: public availability allowing anyone to experiment with code, and active encouragement for others to submit improvements ranging from bug reports to major new modules.

Linus Torvalds announced his free operating system Linux on an online forum in 1991, recruiting a volunteer army of programmers. By early 2008, the Linux kernel grew to nearly 9 million lines of code written by over 1,000 people, and is widely used by companies like Google, Amazon, Pixar, and Sony.

Four Patterns for Scaling Collaboration

Open source collaborations scale by implementing four powerful patterns to amplify collective intelligence, increase cognitive diversity, and leverage a wider range of microexpertise. You can incorporate these patterns into any architecture of attention.

To succeed, you must adopt specific operational steps:

• Commit relentlessly to working in a modular way by finding clever ways to split up tasks into smaller subtasks.
• Encourage small contributions to reduce barriers to entry.
• Allow easy reuse of earlier work by other people.
• Use signaling mechanisms such as scores to help people decide where to direct their attention.

Large collaborations must consciously and relentlessly break down their overall problem into modular subtasks so you do not need to understand the entire project in detail. For Linux release 2.0, the entire code base was substantially rewritten to add loadable kernel modules. Similarly, Wikipedia utilizes Talk pages, Village Pump, and special community pages to break down the massive task of running an encyclopedia.

Issue trackers organize participation around single problems and conversational threads, enabling projects to benefit from specialized expertise. Developers and users utilize the Firefox web browser issue tracker to report bugs like favicon mix-ups, discuss subtle code problems, and propose new features.

Programmers build and draw from a publicly shared information commons, changing the definition of a great programmer from someone who writes code from scratch to a master of reuse. A study by Amit Deshpande and Dirk Riehle shows the open source information commons contains over a billion lines of publicly available code, growing by more than 300 million lines per year.

Competitions that allow immediate code reuse and automated leaderboard scoring function as large-scale collaborations, producing heavily optimized solutions beyond the capability of any single individual. In the 1998 MathWorks CD packing problem contest, the winning program left only 20 seconds of unused space compared to six minutes in early submissions, representing a 20-fold improvement sculpted from contributions by at least nine people.

Microcontributions lower barriers to entry, spark inspiration, and allow collaborations to explore a much broader range of ideas by letting participants make tiny changes like altering a single line of code. Studies by Oliver Arafat and Dirk Riehle demonstrated that the most common change in open source software projects is to a single line of code, followed by changes to two lines.

Automated scoring systems act like market prices, providing immediate feedback signals that help participants direct their expert attention to where it will make the most impactful improvements. In MathWorks competitions, organizers run submitted programs against secret test inputs like Beatles, jazz, and dance music catalogs, instantly calculating scores that populate a public leaderboard.

For online collaborations to succeed, you must effectively convert individual insights into collective insights so the group knows what it knows and avoids squandering available expertise. In Kasparov versus the World, Irina Krush manually reviewed thousands of chess moves, but the system broke down at move 51 when participant Jose Unodos manipulated the voting system to play a suboptimal personal choice.

4. The Limits and Potential of Collective Intelligence

Group discussions often fail to utilize collective knowledge and can actively make decisions worse by reinforcing preconceived ideas rather than pooling unique information. You see that people spend most of their time talking about information they hold in common while ignoring unique information. Furthermore, groups give disproportionate attention to information held by high-status members and ignore low-status members.

A shared praxis consisting of common knowledge, techniques, and standards is a fundamental requirement for scaling up collective intelligence.

Psychologists Garold Stasser and William Titus showed this failure in a 1985 experiment using fictional presidential candidate profiles. Groups given partial information chose an undesirable candidate more often after discussion, dropping the best candidate support from 25 percent down to 20 percent.

Without a shared praxis, groups experience unresolvable disagreements based on different values or aesthetic standards, causing the community to fragment. When a shared praxis exists, participants can rapidly agree on what constitutes progress and collaborate effectively.

Shared Praxis and Group Limits

There is a significant gap between the difficulty of having an insight and the ease of recognizing that insight once it is explained. Many problem-solving projects succeed because participants can easily evaluate and verify a valuable contribution even if they could not have generated it independently. This asymmetry fuels collective progress.

Consider an empty eight-by-eight chessboard puzzle. Covering the board with one-by-two dominoes while leaving the bottom-left and top-right squares uncovered is impossible because those two squares are the same color, but people easily understand the proof once it is explained.

Fields like economics sit on the cusp of collective intelligence because they possess partial shared reasoning methods alongside fundamental disagreements. President Harry Truman famously asked for a one-armed economist because economists constantly disagree on fundamental questions, giving rise to the joke that five economists in a room will produce six opinions.

Science is exceptionally well suited for collective intelligence because it maintains a large repository of shared techniques and widely agreed standards for correctness. Albert Einstein published special relativity in 1905 as an unknown 26-year-old clerk, and the physics community rapidly accepted his compelling arguments. Similarly, James Watson and Francis Crick recognized a basic chemistry error in Linus Pauling's proposed DNA structure.

Collaboration Markets in Science

Online tools can create a collaboration market that matches scientists seeking help with experts who have a comparative advantage to solve specific bottlenecks. Ranking algorithms can prioritize millions of daily assistance requests to connect theoretical researchers, students, and experimentalists based on their specialized expertise, significantly lowering costs and accelerating discovery.

A theoretical physicist at Caltech can overnight receive a prioritized list of assistance requests, such as helping a materials scientist in Budapest solve a specialized diffusion problem on a triangular latticework or guiding a student in Bangalore.

For networked science to reach its full potential, it must evolve into open science supported by a culture of transparently sharing data and knowledge. Although cultural obstacles and competitive secrecy currently limit the sharing of key ideas, realizing the full vision requires an interconnected online web of scientific knowledge, code, reasoning chains, and social tools.

5. All the World’s Knowledge

Data-driven intelligence and massive open datasets transform scientific inquiry by shifting the primary constraint to human ingenuity.

Information scientist Don Swanson reasoned that scientific knowledge had grown so vast that important connections went unnoticed. Using Medline, he searched for conditions associated with migraines and found eleven two-stage connections pointing to magnesium deficiency.

Don Swanson investigated migraine headaches in 1988 using Medline and discovered evidence suggesting migraines are caused by magnesium deficiency. Traditional medical therapeutic trials later confirmed this idea.

The U.S. Centers for Disease Control and Prevention relies on participating doctors to report flu-like symptoms. This creates a one-to-two-week time lag for tracking diseases that kill between 250,000 and 500,000 people around the world each year.

Google and CDC scientists correlated historical flu data with common user search queries to solve this delay. They found forty-five predictive queries and built a model that tracked influenza across 29 countries.

Google and CDC scientists used historical CDC data from 2003 to early 2007 to find 45 search queries correlated with flu outbreaks. Their model achieved a 97 percent agreement when tested against the 2007-2008 flu season data.

Mining Literature and Digital Sky Surveys

The Sloan Digital Sky Survey uses a 2.5-meter telescope in New Mexico to systematically scan the sky. Since beginning operation in 2000, the 25 participating institutions and 200 official scientists have imaged 930,000 galaxies.

The collaboration released over 70 terabytes of image data publicly on the web through SkyServer. The original SDSS paper received 3,000 citations by 2011 because open data invites outsiders to perform independent analyses.

Astronomers J. Richard Gott III and collaborators used SDSS data in 2003 to discover the Sloan Great Wall. This giant chain of galaxies is 1.37 billion light-years long and about 1 billion light-years away from Earth.

Todd Boroson and Tod Lauer used SDSS image data in 2009 to discover two supermassive black holes orbiting one another. They searched for a subtle double-coloring signature caused by relativistic velocity differences.

The paired black holes have masses of 20 million and 800 million relative to the sun. They sit 1/3 of a light-year apart and possess an approximate orbital period of 100 years.

Building the Data Web

Large-scale mapping projects are immediately placing their datasets into a growing global information commons. The Ocean Observatories Initiative began building a high-speed fiber-optic network across the Pacific Ocean floor in September 2009. This underwater internet will extend 100 kilometers offshore with a planned cable length of 1,200 kilometers.

The Allen Institute for Brain Science uses automated robots to map gene expression in human and mouse brains. Each Allen Institute robot analyzes 192 brain slices daily to complete maps expected by 2012.

Data-driven intelligence excels at automated pattern matching across superhuman datasets. In freestyle chess tournaments, hybrid teams pairing human operators with algorithmic search engines consistently defeat both unaided grandmasters and standalone supercomputers.

In a 2005 Playchess.com freestyle tournament, teams pairing human grandmasters with computers trounced standalone world-class chess computers like Hydra. Amateurs using standard software and superior process won the tournament.

Complex biological sequencing problems are solved through specific laboratory steps:

• Make many copies of long DNA strands.
• Cut the copies up at random locations.
• Directly sequence the resulting short fragments using laboratory chemistry.
• Use computers to figure out where different fragments overlap and reassemble them.

Following a presidential memorandum on transparency issued in January 2009, the U.S. government launched data.gov to share over 1,200 open datasets. Wikipedia featured 100 entries on its list of biological databases as of April 2011.

6. Democratizing Science

Citizen science platforms and open access movements engage the public and bridge the societal ingenuity gap.

Citizen Science and Foldit

Online platforms leverage the cognitive surplus of internet volunteers to classify vast datasets. Humans outperform computers at pattern recognition for visual tasks. This hybrid approach processes massive datasets while allowing amateurs to spot anomalous phenomena that automated programs miss.

Galaxy Zoo invites members of the public to classify galaxy images:

• 200,000 volunteers in Galaxy Zoo
• 150 million galaxy images classified
• 70,000 classifications per hour at peak rate

On August 7, 2007, a 25-year-old Dutch schoolteacher named Hanny van Arkel discovered a mysterious blue blob on Galaxy Zoo, which was later dubbed Hanny's Voorwerp and recognized as a potential quasar mirror or active galactic nucleus mirror. Zooites also identified green pea galaxies.

Foldit turns the protein folding problem into an interactive game where players manipulate protein shapes to find the lowest energy configuration. Supported by community forums, shared scripts, and scoring systems, amateur players achieve results comparable to or outperforming fully automated computational servers in global CASP competitions. 60,000 complete protein structures are known via X-ray diffraction, and 75,000 people signed up for Foldit. Foldit players competed in the 2008 and 2010 CASP competitions, finishing near or at the top of the challenges despite having no biochemistry PhDs.

Open Access and Science Blogging

The open access movement uses internet infrastructure to dismantle paywalls and make scientific research freely available to the public. Traditional scientific publishing relies on a subscription and pay-per-download model, locking humanity's knowledge behind expensive paywalls. Open access initiatives like the arXiv preprint server and PLoS change this by making research instantly downloadable or shifting costs to author publication fees. In 2009, Elsevier made 1,100 million US dollars profit, while the NIH operates with a 30 billion dollars annual budget. Physicists use the arXiv preprint server to upload and share papers immediately upon completion, bypassing traditional journal delays and making cutting-edge research globally accessible.

Science blogs establish a direct two-way dialogue between researchers and the public while enabling rapid peer-led investigations of controversial claims. Prominent science blogs offer millions of readers a transparent glimpse into how scientists think and work. Paul Myers' blog Pharyngula received 100,000 visits per day, and Terence Tao's mathematical blog is read by 10,000 people. Ad hoc groups of science bloggers can quickly scrutinize and dismantle dubious technical claims presented in public disputes through wiki litigation. After author Simon Singh was sued for libel by the British Chiropractic Association, a group of science-bloggers investigated and discredited the BCA's cited evidence in less than a day.

Online tools act as institution-generating machines that help society bridge the ingenuity gap between complex global challenges and our capacity to solve them. Civilizations can collapse when they fail to generate solutions to the problems they create. The Polynesian society of Easter Island experienced a catastrophic population crash and cultural collapse after consuming all their natural resources and failing to solve their internal ingenuity gap, dropping 90 percent from an estimated peak population of 10,000 to 30,000. By making it easier to coordinate collective action without monetary hierarchies, online tools enable new institutions that redefine the role of science in society.

7. The Challenge of Doing Science in the Open

Academic reward structures and the lack of career incentives for tool building create severe barriers to open science.

In the seventeenth century, scientists faced intense competition and risks of plagiarism without institutional systems for sharing. Galileo, Newton, Huygens, and Hooke used anagrams and ciphers to claim credit for discoveries like the rings of Saturn while retaining time to build upon them. Galileo announced his observation of Saturn's rings via the anagram smaismrmilmepoetaleumibunenugttauiras, which Kepler incorrectly tried to decode as moons of Mars. Wealthy patrons and leaders like the Medici subsidized science for prestige and public benefit, tying career rewards to the sharing of discoveries rather than secrecy. Michael Faraday summarized his success as Work, Finish, Publish, reflecting the norm that unpublished work was incomplete.

Historical Roots and Academic Hyper-Competition

An oversupply of PhD graduates relative to available faculty positions creates a logjam, driving researchers to devote all their time to activities that yield papers, grants, and letters of recommendation. US universities produce 1,300 physics PhDs annually while only 300 faculty positions open up, and mid-level university job openings easily draw over 100 applicants. Projects aimed at building comprehensive online references face a collective action problem. The qwiki, launched in 2005 by John Stockton for quantum computing, failed to gain community contributions beyond vanity researcher pages and was largely maintained by its creators and successors. User-contributed comment sites for science fail because scientists are reluctant to publicly critique peers who might referee their next paper. Nature's 2006 open peer review trial and physics sites like Quick Reviews and Physics Comments were discontinued due to lack of substantive community contributions.

The Stigma Against Tool Builders

Pioneering online tool creators often receive little institutional support or funding and are viewed privately with condescension by peers who benefit from their systems every day. A well-known physicist remarked that Paul Ginsparg, creator of the arXiv, had wasted his talent for physics and was doing garbage collecting. Achieving the full potential of networked science requires extreme openness, moving all scientific information onto machine-readable networks. Laws granting universities ownership of intellectual property from government grants led institutions to pursue applied research and commercial spin-offs, eroding the traditional division between open basic science and secret applied work. Between 2001 and 2003, a quantum computing research center required scientists to sign nondisclosure agreements and sign dated notebooks for future patent applications.

Systemic cultural change happens through fifty-year horizons built from small individual actions. You can take practical steps toward open science right now:

• Upload some of your old data and old code online, document it, encourage other people to use it, and specify how you would like to be cited.
• Try out blogging and use your blog to develop ideas you have had in the back of your head.
• Make a small contribution to others' open science projects, such as commenting on a science blog or contributing to a wiki.
• Be generous in giving other scientists credit when they share their scientific knowledge in new ways by citing their online ideas, data, and code.
• Promote your open work by highlighting it on your curriculum vitae and grant applications, demonstrating its impact.
• Talk with colleagues holding old-fashioned scientific values about the value and courage of working in the open.

10 Key Ideas

1. Mass online collaboration can solve extremely difficult, unsolved mathematical problems much faster than traditional isolated research.
2. Online architectures of attention harness latent microexpertise and transform lucky coincidences into designed serendipity.
3. Open source methodologies scale effectively through relentless modularity, microcontributions, radical reuse, and automated scoring signals.
4. A shared praxis consisting of common knowledge, techniques, and standards is a fundamental requirement for scaling up collective intelligence.
5. Data-driven intelligence and massive open datasets transform scientific inquiry by shifting the primary constraint to human ingenuity.
6. Citizen science platforms and open access movements engage the public and bridge the societal ingenuity gap.
7. Academic reward structures and the lack of career incentives for tool building create severe barriers to open science.
8. Centralized online databases integrate data from thousands of researchers so computer algorithms can find hidden meanings across vast fields of knowledge.
9. Open source software projects rely on making code publicly available and encouraging widespread community contributions to build software far more complex than any individual could create alone.
10. The open access movement uses internet infrastructure to dismantle paywalls and make scientific research freely available to the public.