Three Seats Beat One Genius

The Core Insight

A search on Amazon for the phrase the man who invented returns 1,860 book results. Walter Isaacson opens The Innovators with that number, because it measures the habit the book exists to correct. The computer and the Internet came out of teams, institutions, and public money.

The dispute is staged in the first pages with two histories of the same machine. Grace Hopper wrote the history of Harvard's Mark I around Howard Aiken. IBM commissioned a rival history featuring the teams of engineers who contributed the incremental advances. Isaacson refuses to pick, and puts the advances down to creative individuals paired with teams that build their ideas.

Most people read technology history as a sequence of solitary minds, and the industry sells that reading back on magazine covers. Isaacson argues the lone inventor is an artifact of how the story gets reported. Ideas were handed from one cohort to the next, and the credit belongs to whoever combined the parts and shipped.

ENIAC shows the shape. Mauchly consulted and supplied the ideas. Goldstine ran the Army's budget and operations. Eckert was the chief engineer who checked every resistor in the machine. The machine ran in November 1945 with 17,468 vacuum tubes, close to thirty tons, and five thousand additions a second. No one of the three built it.

The Framework

Isaacson states the structure once, in the Bell Labs chapter, and then demonstrates it for the rest of the book. Three roles have to be present in the same place, and the solitary genius is one of them.

• A creative genius generates the idea, as John Mauchly did at Penn and Nolan Bushnell did at Atari.
• A practical engineer turns the concept into a working object, as Presper Eckert, Walter Brattain, and Al Alcorn did.
• Entrepreneurs and technicians build the market, as Gordon Teal and Pat Haggerty did for the transistor.
• Remove one of the three and the work stops, whatever else the building contains.

The removals are the proof. Babbage and Atanasoff had the concepts and no engineers. Penn after Mauchly and Eckert left, Princeton after von Neumann, and Bell Labs after Shockley kept the teams and lost the engine.

Teams form three ways in the book, and none of the three wins outright. Government money built Colossus, ENIAC, and ARPANET. Companies built Bell Labs, Xerox PARC, Intel, and Microsoft. Peer production built Wikipedia, the Web, Linux, and GNU. Isaacson's verdict is that the combination beats any one of them alone.

Key Ideas

Isolation Is The Reliable Predictor Of Failure

Every failure case in the book carries the same shape: a correct idea with no institution around it.

• Leibniz had the theory and no engineer near him, so his calculating device never worked reliably.
• The British government sank more than 17,000 pounds into Babbage, twice the cost of a warship, and no engine was ever built.
• Konrad Zuse finished the first working programmable digital computer in 1941, then lost his machines and designs to the 1943 bombing of Berlin.
• John Atanasoff left for the Navy in 1942, and nobody at Iowa State remembered what his machine was.

Atanasoff is the cleanest case. He sketched regenerative memory on a napkin in December 1937, after a night drive of 189 miles. A private foundation gave him 5,330 dollars. By September 1942 the machine held close to three hundred vacuum tubes and a card reader that never worked. Iowa State had no machinists or engineers to hand the problem to. In 1948 a graduate student dismantled the machine without knowing what it was.

Alan Turing nearly joined that list. Max Newman warned in a recommendation that Turing worked with no supervision or criticism. The remedy was contact with other workers, before he became a confirmed solitary. Turing went to Bletchley Park instead. His team ran two bombes by August 1940, and they broke 178 coded messages.

The Institution Is The Invention

The transistor appeared on December 16, 1947, out of gold foil, a chip of semiconductor, and a bent paper clip. Bell Labs earns as much credit as the three men in the room.

Mervin Kelly became research director in 1936 and made basic science a routine industrial activity, where until then it belonged to universities. The building was policy. Murray Hill sat on two hundred acres with every building connected, so departments held no fixed border. The corridors ran longer than two football fields, which pushed people of different specialties past each other every day. John Bardeen had no office and worked inside Walter Brattain's lab.

War work supplied the scale. In the first years after Pearl Harbor the Labs took on nearly a thousand military projects, and the staff doubled to nine thousand.

Shockley showed the reverse. Passed over on the point-contact patent, he worked alone in a Chicago hotel. He produced the junction transistor on January 23, 1948, and kept it secret for almost a month. Bardeen resigned and wrote that Shockley had used the group to exploit his own ideas. The atmosphere that produced the transistor did not survive him.

Noyce built the opposite structure at Intel: identical cubicles, and an org chart drawn as one X among other Xs. Neither Noyce nor Moore settled disputes, so Andy Grove supplied the confrontation. Grove found the pattern later in Drucker. The outside person, the inside person, and the person of action work as three people rather than one.

Public Money Bought The Long Horizon

Vannevar Bush delivered Science, the Endless Frontier in July 1945, and the National Science Foundation came out of it. From the 1950s through the 1980s, the Defense Department and the NSF spent as much on basic research as all of private industry.

Military demand did more than fund the microchip. It set the price. Minuteman II needed two thousand chips for its guidance system in 1962, and by 1965 seven Minutemen were built each week. Fairchild won the Apollo contract, and seventy-five guidance computers carried five thousand identical chips apiece. The Apollo program bought more than a million microchips by July 1969.

Prices collapsed under that volume. An Apollo prototype chip cost 1,000 dollars and cost 20 dollars in production. Minuteman chips averaged 50 dollars in 1962 and 2 dollars in 1968. Consumer electronics arrived on a price the taxpayer established.

Noyce refused the standard contractor arrangement, because it handed research direction to people less competent to set it. He paid for chip development with company money instead. In 1964 he sold simple chips below cost to force volume.

The pattern repeats. ARPA money built the network. Paul Allen wrote the Altair emulator on Harvard's DARPA-funded PDP-10, and the Gore Act of 1991 paid for the center where Mosaic was built.

Two Contract Clauses Commoditized Hardware

Grace Hopper called the split first. She built the A-0 compiler by 1952 and sent early versions to friends, asking for improvements. Her claim was that hardware turns into a commodity while programming holds the value.

Gates collected on that twice, and he did it with contract terms. Allen bought the January 1975 issue of Popular Electronics for seventy-five cents, and the Altair BASIC deal followed. The MITS license ran ten years and paid 30 dollars a copy. Two clauses carried it: Gates and Allen kept ownership, and MITS carried a duty to sublicense the software to other hardware makers.

The IBM contract of early November 1980 ran thirty-two pages and paid something like 186,000 dollars. Microsoft that year had forty employees and 7.5 million dollars of revenue against IBM's 30 billion dollars. The license was nonexclusive and Microsoft kept the source code, so every clone maker had to buy MS-DOS. The business model carried it, and Windows held 80 percent of the market by 1990 and 95 percent by 2000.

Noyce wrote the clause first. Intel gave Busicom a low price on the 4004 and kept the right to license the chip to anyone outside calculators. The 4004 was unveiled in November 1971 at 200 dollars.

Open And Closed Both Won

The open cases are the ones that spread. Bell Labs licensed the transistor to any company for a flat 25,000 dollars and taught manufacturing seminars on top of it. Von Neumann's First Draft went out as twenty-four mimeographed copies at the end of June 1945. That distribution counted as prior publication and voided the Eckert and Mauchly patent claims on the stored program. Stephen Crocker mailed RFC 1 on April 7, 1969, under a title chosen to claim no authority. CERN gave up all rights to the Web code at Berners-Lee's request.

Closed control paid too. Proprietary systems funded the hardware spurt of the 1950s, Apple took margin from integration, and Microsoft took share from licensing breadth. Torvalds paid 169 dollars for MINIX because UNIX cost 5,000 dollars a copy. He chose the GPL because he wanted help with the kernel.

The patent fights settled ownership and changed nothing else. The ENIAC trial ran more than nine months and entered 32,600 exhibits. Kilby and Noyce spent a decade and more than a million dollars in legal fees, after their companies had already cross-licensed in 1966.

Isaacson's verdict is that government funding, private companies, and peer production all had to coexist. The work stays most alive where open systems compete with proprietary ones.

Partnership Beats Artificial Intelligence

The Dartmouth conference of 1956 put a thinking machine about twenty years away. It remains about twenty years away.

The machine wins are real and narrow. Deep Blue beat Kasparov in 1997 by evaluating 200 million positions a second against 700,000 grandmaster games. Watson won at Jeopardy in 2011 on 200 million pages in four terabytes, of which Wikipedia was 0.2 percent.

Kasparov drew the useful conclusion after losing. Computers are strong where people are weak, so he proposed partnered play. At the 2005 freestyle tournament neither the best grandmaster nor the strongest computer won. Two American amateurs running three computers took it, on their skill at coaching them.

That result is the closing argument. Licklider published Man-Computer Symbiosis in 1960, and Engelbart aimed at augmenting people rather than replacing them. Page and Brin built a ranking system out of billions of human judgments in links.

IBM pointed Watson at oncology and had to program humility into it, so the system reports a percentage likelihood instead of an answer. Isaacson proposes the Licklider Test: whether people and machines in partnership beat any artificial intelligence working alone.

Practical Applications

Name the three seats on your current project, with a person in each. An unfilled seat is the next hire, and it names the failure you are walking into.

Treat a departure as a structural event rather than a staffing one. When the idea person leaves, the team keeps shipping increments and stops making leaps. Penn and Bell Labs each ran that experiment.

Put the theorist and the builder in one room, and treat the floor plan as a decision you own. The transistor came out of a daily blackboard argument between a quantum theorist and a man who built with sealing wax and paper clips.

Write the licensing terms before writing the code. Answer three questions: what you own, who can resell it, and whether the buyer gets exclusivity. Gates answered them twice and outran a decade of hardware engineering.

Who This Is For

Founders assembling a first team get the most from this. The book is a long catalogue of team compositions with outcomes attached. Anyone deciding what to license and what to keep gets the second most.

Skip it for method. The book supplies patterns and cases, no procedure, and long stretches carry no mechanism. The Byron family melodrama is the clearest example.

Read it as popular history assembled from secondary sources. Isaacson is a biographer working from other people's archives, published accounts, and interviews, and he selects the cases that fit his argument. The isolation cases are in the book because they failed, and the institutional cases are in it because they worked. A book arguing against great-man history still runs chapter by chapter on famous names, with Ada Lovelace at both ends. The mechanism holds up. The sample is the author's.

The Decision

The test runs on your own team and takes ten minutes. List the people who decide what gets built. Write the seat each one holds: idea, engineering, market.

One person in two seats is a warning, and an empty seat is a forecast. A missing engineer produces Atanasoff's card reader, correct on paper and broken on the bench, with nobody left to repair it.

A missing market person produces Galaxy Game. Pitts and Tuck built a better version of Spacewar at Stanford in 1971, on a 20,000 dollar PDP-11, and charged ten cents a game. Bushnell stripped the same game into hardware, made Computer Space for 1,000 dollars a unit, and sold 1,500 of them.

Both teams had the engineering. One of them worked out the price. Run the seat count this week, and hire against the gap.