Andrej Karpathy mentioned The Idea Factory by Jon Gertner 2 times

https://www.goodreads.com/review/show/1448792342

— Andrej Karpathy

The Idea Factory is a fascinating book for anyone interesting in the process of innovation. The book follows the particular example of Bell Labs, which at its peak in ~1940-70 was a 1000+ PhD behemoth conducting full-stack research/development/deployment in communications, with decades of runway enabled by its parent company’s (AT&T) massive government-regulated monopoly. Through conscious top down design/strategy the lab was able to sustain a golden age of innovation: it developed a vast array of transformative technologies including transistors, photovoltaics, digital transmissions, fiber optics, radio astronomy, CCDs, lasers, UNIX, C, and even supported a range of pure research results such as Shannon’s information theory. For me, the core question evoked by the book is the following: What is the best approach to fostering innovation? Universities are one particular model (though they are not full stack). Private institutions or prizes are another. The model studied in detail in this book is the singular (and arguably not easily reproducible) example of the Bell Labs model: a massive industrial research lab living comfortably within a huge government-regulated monopoly. The core idea was to pack thousands of researchers into one physical location, give them a vertical as a constraint (communications in this case), and otherwise set them loose. This quantity of talent at one physical location was (and still is) historically unprecedented. Many aspects of the lab were explicitly designed top down with the goal of supporting innovation in mind. For instance, the physical space was arranged along a large corridor so that chemists, physicists, mathematicians and electrical engineers, with focus on both theory/applications were forced to mingle in serendipitous interactions. As another example, the labs created courses to diffuse the knowledge through the organization, effectively creating a small college. Another benefit of the organizational structure was the so called “going to the person who wrote the book” - the idea being that among so many people there was an expert for everything, so everyone had instant access to the best people in any particular vertical of interest. However, among so many benefits of the organization structure, the Bell Labs model came with its own set of problems. On one hand, AT&T was a massive monopoly ensuring the continued existence of Bell Labs, but this was a double-edged sword because the executives had to constantly fight pressure from the government who repeatedly tried to break up the monopoly. The need to stay on the good side of government introduced a whole layer of complexities and compromises and constrained the research. For instance, as part of a settlement AT&T had to agree to stay out of the quickly emerging computing business. Additionally, they had to work on several projects that were not necessarily related to AT&T’s profits, but pleased various government incentives or justified the lab’s existence as beneficial to the wider society. Also, their patents had to be released and licensed at low prices. Therefore, innovations developed at AT&T’s Bell Labs through a long and expensive process formed the basis of several companies that would build on this work and later compete with AT&T. For reasons not made entirely clear (only few unconvincing hints are given), the Bell Labs started to slowly die down near the end of 1970’s. For instance, they missed the development of the Integrated Circuit, which instead came from startups in the nascent Silicon Valley such as Texas Instruments and Fairchild semiconductors (these companies were often staffed by ex-Bell Labs employees). An argument could be made that perhaps Silicon Valley came up with a better (or faster) model of innovation, based around smaller competing units instead of one large, monolithic lab whose continued existence does not necessarily depend on the speed of its research. Eventually, the lab was severely downsized as a result of the monopolistic breakup of the Bell System in 1982. With it came the end of an era, concluding one of the most interesting large-scale experiments on innovation. A lot of these topics are very nicely covered in the book, but if I had to pick on something it would be that the book missed a chance to zoom out and provide a more concrete analysis of pros/cons of different funding models of innovation from the historical perspective. Bell Labs is just one experiment, but we have data from many other forms of institutions, and these were not explicitly compared or contrasted. Unsorted additional interesting points: - Not all Bell Labs projects along the way were successful. For example, AT&T made a large bet on a picture phone, for which technology existed already back in 1960s. However, despite market research indicating that people wanted a picture phone, the actual product was a decisive flop. It turned out that people were content (and even preferred) non-visual communication over the phone, and it also turned out that people don’t really know what they want if you survey them, especially when it has something to do with a new, not yet existing product (Steve Jobs knows something about this too). Indeed, I have FaceTime on my iPhone but rarely ever use it. In fact, I rarely ever conduct ordinary phone calls too, preferring instead a simple, asynchronous chat. Sometimes technology offers a solution to a problem that isn’t. - The book drew an interesting distinction between two types of innovations: one is a “if I don’t do this someone else will within the next few years”, for ideas that are “in the air”, and it’s mostly a race to who does it first (the transistor is argued by its inventors to be of this form). On the other end of the spectrum are innovations that present a much larger and less anticipated leap (Shannon’s information theory being an example). It seems that in practice one is mostly rewarded for being first without regard to this distinction, which might sometimes unfortunately incentivize work on the first type of problems. For example, the Bell Labs scientists still got the Nobel Prize for the transistor, even though they practically rushed it to publication in fear of being scooped by one university research lab that was suspected to be on the verge of discovering it as well. - Many ideas about what the future might look like are not necessarily wrong, just early. The book highlights several examples of people struggling with this dilemma, e.g. with the example of satellites in space used for communications. Although this point, along with its common analogy of catching a wave during surfing is widely appreciated. - Another insight offered by the book was the image of progress looking somewhat like a staircase, but the steps are relatively small and involve small leaps and myriads of small innovations and improvements. Therefore, zooming out innovation across many disciplines looks mostly like a smooth line, and the edges of the stairs are only visible on short time scales. - Another distinction drawn by the book is between “invention” and “innovation”. The former is about the act of coming up with a new, novel idea. Meanwhile the latter is taking that idea and scaling it into a marketable, mass producible product (which can in fact be quite a lot more complex and time demanding). Usually these two pieces are done by different institutions, but Bell Labs was careful to draw the distinction since they were involved with the full vertical stack, from theory on the bottom to marketable products at the end. - I enjoyed seeing the transistor as a very nice example of creative destruction at work. The transistor deprecated almost two decades of research on perfecting vacuum tubes. To the credit of AT&T’s executives, they encouraged its development despite the imminent cannibalization of much of their vacuum tube business. An amusing case study of the reverse (and wrong) strategy is Kodak, who pioneered the creation of the digital camera but failed to invest due to their fear of eating into their traditional (and profitable) business of selling chemicals for film. - Another fun (though brief) anecdote was Bell Labs’ large-scale experiment on productivity of its employees. They looked at a number of possible predictors for a productive employee (measured by number of filed patents), and discovered that most variables showed relatively unconvincing correlations. Amusingly, the one that showed strongest correlation was that the most successful employees had lunches with Harry Nyquist. The running theory was that Nyquist was a great person to talk to for his talent of asking very good questions. There are many (too many!) other fun anecdotes and thoughts along the way, which I cannot hope to serialize here. In summary, a great book not only for its content, but also for the wider thoughts it elicits and teases. I have to read this again at some point. 5/5

— Andrej Karpathy

2015-12-29 on goodreads.com