Monday, May 18, 2009

User innovation at MIT

Workshop on Intellectual Property Law and Open & User Innovation, by the MIT Sloan School of Management and the Berkman Center for Internet and Society


Eric von Hippel: User, Collaborative, and Open Innovation are Increasingly Common

Users aren’t always the innovators; they do a lot of innovation in scientific instruments (77%), less in others, varying a lot by field. Sticky information: when users have hard-to-transfer information. Users tend to develop novel functional capability, like the first sports nutrition bar, but manufacturers tend to innovate in quality of delivery (like an improved power supply for a device). Each user responds to local needs using local situation information. Consequence: user innovation is widely distributed. Water vest for US troops was developed by a Texas biker/paramedic who was used to getting thirsty and used to hydrating people through IV bags; he combined those kinds of local knowledge.

Lead users innovate where no market yet exists, then collaborative user communities improve and filter innovations. Then manufacturers enter, usually nonincumbent manufacturers coming out of the user communities. Eventually incumbents decide that the market is big and secure enough to enter. Users can be big: Boeing is a user innovator when it makes machine tools to make its products.

The traditional linear model of innovation does not even show users as process actors; innovation ends with marketing. This is wrong! IP tends to ignore the user innovation side.

Case studies show that many users innovate, especially enthusiasts. 20-50% of firms develop or modify process equipment they use, at considerable expense. The most generally useful 25% is transferred to producers. IP claims are rare; they usually give it out for free.

Consumers: survey among UK consumers—have you created any products from scratch/modified any products you use in daily life to make them work better for you? 10% of ordinary consumers have done the former and 17% the latter in the past three years. Manufacturers patent product engineering; users don’t patent innovations.

What new policies are required? Infrastructure for distributed innovation. There is pressure in the market towards openness.

Carliss Baldwin: What Do the Designs Want: When Does Open, User, Collaborative Innovation Dominate Producer Innovation?

Marx: the hand mill produces the feudal lord, the steam mill the industrial capitalist, quoted by Heilbroner in “Do Machines Make History?” (1967). PCs/internet give you, potentially, open and user-based collaborative innovation. Scientific controversy: do we need strong IP/contract rights for incentives for wealth-seeking innovators to produce new designs? Or do we need to encourage communities with norms of openness and sharing to allow users and others to collaborate on the development of new designs?

We seem to agree that innovation is good, and that certain ways of organizing processes of innovation are “better”—in terms of greater social welfare, or in terms of winning in head-to-head competition. So, what kinds of designs are well matched to the social structure of open and user collaborative innovation?

Design space: some designs demand a lot of communication between maker and customer and others don’t. User innovating for her own/her group’s sake—will innovate if the cost is less than the value of the design; no external communication costs. Producers will add innovation, as long as the cost of communication and design is less than the aggregate value to the producer.

User innovation requires: a modular/task-divisible architecture; low design cost for individual pieces. Doesn’t suffer deadweight loss of monopoly pricing; permits easy recombination of ideas; makes design ideas accessible, promoting education and the creation of new designs.

Karim Lakhani: Given Micro-Contributions, Who is the Inventor?

We know OS aggregates lots of contributions from around the world. This is coming out also in music (always been true, more salient now). Data from PostgreSQL, an industrial strength database that is used by Sony and Skype, among others. Tracked every feature for a year: 55,000 lines of code written by the community. About 800 people participate, all users. They all work with different firms (except for 2 at the same firm). Example: one person identifies a problem; eight people participate in discussing the problem and various solutions, and within a day a possible solution is identified and one person works on it. In a month, another person improves on it. 23 people participate in the creation of the solution: 3 people wrote code; 1 reviewer; 2 testers; 17 discussants. The final worker got credit in the credit file, but the underlying work was critical.

On average, about 9 people participate in creating each feature. 3 participate in problem definition, 5 in development until source code committed; the contributions are pretty efficient, the vast majority of them ending up relevant to the solution.

Katherine Strandburg: A Review of the Traditional Justifications for Intellectual Property

Assumption of one traditional model that IP is necessary to avert free riding and allow people to recoup investments. Assumptions: (1) There’s a need to recoup the costs by beating out competitors; the motivation to create for oneself is insufficient—the innovator is a seller. (2) There aren’t enough first-mover/reputational/other advantages to recoup investment without IP. (3) Creator will prefer not to create at all rather than create and share.

Next increasingly popular justification for patents: inventors won’t disclose inventions if competitors can immediately copy them. Patents = early disclosure. Assumptions: (1) Trade secrecy is possible. (2) Early disclosure is preferable to trade secrecy plus independent invention/reverse engineering. (3) Creator prefers secrecy over free revealing, given a choice.

Prospect theory—controversial even on traditional terms. Broad exclusive rights promote efficient exploitation of inventions, avoid duplication. Assumptions: (1) Single right holder can coordinate optimal development of a particular line of technology. (2) Transaction costs of licensing follow-on innovation will not be prohibitive. (3) Inventors are interchangeable/they are invention managers.

Incentive to disseminate/commercialize: mostly for patents and also some copyrighted works. (I think Strandburg underestimates the extent to which this is a persuasive idea in copyright, especially for people who accept that a weakness of the basic incentive argument is that creators so obviously love to create.) Idea: there’s a “lab to market” gap—give exclusive rights to ensure investment in commercialization, and permit a “market” for ideas Assumptions: (1) Relatively large investment is required to bridge the lab to market gap. (2) Dissemination is costly. (3) The first mover/reputation advantage is insufficient.

IP law generally tries to balance incentives to invent, disclose and disseminate versus increased prices, reduced follow-on creation. Doctrinal handles include patent’s term, nonobviousness, utility requirement, claim scope doctrines; copyright’s substantial similarity test, fair use, and the idea/expression dichotomy. Bottom line: even under traditional IP view, we have no idea how to tailor these things.

Andrew Torrance: Empirical Evidence Challenging the Orthodoxy

Growing skepticism of traditional claims for IP, especially for patents. Very limited empirical literature. Moser: in 19th century, countries that offered patent protection did not have higher rates of innovation than countries without it. Lerner: 60 countries over 150 years, found that strengthening patent law didn’t seem to help innovation. Bessen and Meurer, Patent Failure: empirically, patent provides little incentive for innovation for most firms, and even drags on innovation, especially with software. May be different in biotech/pharma.

Online patent system simulation, with ability to change parameters like duration, difficulty of acquiring patent, and so on. Pure patent does slightly better than patent/open source, but pure commons does better than pure patent. Statistically significant at 5%. In total number of innovations, pure commons does a huge amount better, significant at .1%. And social utility is 10x higher in the pure commons than in the other two regimes.

Von Hippel: Not obvious that drugs/biotech are different. Drug trials, for example, are now being modularized.

Wendy Gordon: shocking to hear that mixed systems did worse in Torrance’s trial. Does this throw the legitimacy of all the hybrids we love into doubt?

Session 2: Theoretical Approaches to User and Collaborative Innovation

Yochai Benkler: "Intellectual Property" and Cooperative Human Systems Design

However ambiguous the total effect of IP on innovation, it’s unambiguous that strong IP benefits exclusion-based strategies at the expense of market- and non-market-based strategies that were not exclusion-based, whether based on knowhow or other things. So Benkler looked at the foundations of large-scale distributed production of innovation—low-cost distribution/modularization of work so as to allow small-scale contributions, coupled with diverse motivations and appropriation models. Allows for all sorts of mixed models, like IBM’s open source strategy, or YouTube, where market actors provide platforms for market- and non-market actors to distribute innovation.

Now he’s working on the microfoundations of cooperation. Once we get away from the rational self-interested actor as a sufficient model, what’s the evidence-based approach towards diverse people who are motivated to some extent by self-interest, to some extent by morals/social commitments, etc.? Trying to synthesize design levers, starting with the centrality of communication and how communication affects behavior. Moving to how we define our utility functions—with whom do we have empathy? How do we define norms (example of Wikipedia, which began with no technical constraints on defection but developed them over time)?

Experiments: he wants to build a web-based platform to run standard economics experiments on the web, identifying different effects of, for example, introducing empathy by adding a face or showing people within networks. Trying to use real-world interfaces to see how they affect user contributions. Example: voluntary music distribution sites that ask for donations; what configurations produce what level of payments? 150,000 transactions to date: levels of contribution are substantial—48% pay the typical rate of $8 an album, even though they could download for free. What happens if you randomize new subscribers to empahsize morality, show the artist’s face, etc.?

Terry Fisher: Why User Innovation Matters

Spectrum of types of innovation from centralized (pharma as we know it) to fully decentralized (like windsurfers in Maui, per von Hippel). Also have a variety of mechanisms for stimulating innovation and its dissemination: typical IP rights, grants, prizes, extralegal norms, systems of nonpecuniary rewards like prestige and satisfactions of sharing. It’s a mistake to begin by assuming a tight match between centralized/traditional IP and decentralized/no IP.

His project: Noneconomic/nonwelfarist reasons why user innovation matters. User innovation in the cultural context: fan fiction, real person slash, machinima—which tend to run up against copyright/IP hazards. Then there’s user innovation in the industrial context: hacked bicycles to run knife-sharpening machines in the developing world. These are connected in ways we haven’t often seen.

Why are these things good? The standard answer: because they’re welfare-enhancing. He’s all in favor of exploring that, but he has some non-welfarist reasons to care. (1) Cultural—semiotic democracy. Distributed innovation/creative engagement with mass-produced products leads to a more just and enriched culture and is good for the soul. Egalitarian and democratic. (2) Self-expression: commonly associated with opinion/artistic creativity. Same issues can be seen with user innovation in the industrial context, mixed with aesthetic rewards. We don’t just like things that work: we like elegant and graceful solutions to problems. (3) Communitarian: forming communities to produce and share innovation has functional advantages, but it also creates life-sustaining bonds between people. The communities around user innovation vary a lot. Woodworkers share things quite differently from climbers/windsurfers.

Methodology: Introspective aspect. He doesn’t know how to create mashups. But he does participate in industrial user innovation. Some of his happiest times are spent in the shop.

Brett Frischmann: Ongoing research projects on user innovation and commons. First project: infrastructure resources. Framework paper: research agenda for investigating commons where members of a defined community pool their contributions in a defined setting and distribute them in some way. Seek to extend Elinor Ostrom’s work on sustainable commons in the natural environment. Hoping to offer a set of questions that can be asked and answered in multiple contexts, so that different studies can be compared and contrasted.

Jonathan Barnett: Sharing in the Shadow of Property: Rational Cooperation in Innovation Markets

Game theory used to determine whether sharing can substitute for property under particular conditions. Players can cooperate, defect by claiming property (happened in semiconductors and in software, where claiming property was uncommon but then became standard), or defect by copying. Sharing regimes are viable but unstable. He looked at group size, capital intensity, asset values, and endowment heterogeneity: the last variable is the value of the innovation brought to the pool. The weakest and the strongest innovators threaten the stability of a sharing regime. Weakest: hard for them to meet contribution requirements. Strongest: they get less out of participation than others and have incentives to stop contributing/defect to property.

Sharing works best with low capital intensity requirements, small group size, lower asset values, and less endowment heterogeneity. Property is the opposite. He’s interested in mixed-form regimes, where the factors point in different directions. IP is everywhere, but show is sharing: every market that apparently supports innovation without IP uses some other instrumetn to regulate access at some point on the total bundle of products and services, and every market operates under a mixed regime where norm-intensive sharing arrangements are embedded within a property infrastructure.

Core/perimeter structure: a sharing regime at the core with a low-cost flow of information assets embedded in a property structure that regulates access, keeping out the low-endowment innovator and allowing special remuneration for the high-endowment innovator. Craft guilds; academic research (supported by reputational technology, the citation, but the university is an artificial creation supported by huge subsidies); open source as well, but access is regulated by reputation/talent and the market models are bundled with a proprietary element, like IBM’s hardware or packaging like Red Hat. Sharing works best when supported by property infrastructure.

Wendy J. Gordon: Gift Failure

Starting with Lewis Hyde’s notion of gift, married with Benkler’s work: voluntary, mutual cooperation not only produces user innovation but creates enough incentives—monetary, reputational, emotional—that people can stay alive and do it. How to frame the question we’re all interested in? Gordon is talking about gift, but gift may or may not work better than other ways of talking about these dynamics.

She was also inspired by an argument about art: she once argued that art is created by gratitude, the artist wanting to pay back what she gains when she sees the beauty of the sunset. Is this model a way to persuade, to capture the imagination?

Perfect gifts and gift failure: alternative to “perfect markets and market failure” as an analytic construct, shifting the burden of proof to people who want to start with the notion of perfect markets. A perfect gift is no more fanciful than a perfect market; before you institute IP rights, you should have to prove that there is a gift failure such that property rights are appropriate to solve the failure. A perfect gift would be: willingly given, with the needs of the recipient(s) in mind, reciprocated with good will (not anger, hierarchy, resentment) and reciprocated with money and emotional support as well as with new art. Hyde’s notion is one of cooperative community that within its own borders of artists (high culture rather than industry) gives to each other monetarily, emotionally, with community. Giving back and paying forward by creating new art. (Resentment can also be fruitful, as Harold Bloom reminds us.)

Range of incentives/inputs for cultural goods: control; money; fun/self-expression/satisfying the itch to affect the world, etc. The GPL explicitly says it’s not a gift—Gordon asked Eben Moglen why not. Answer: Moglen doesn’t want resentment; people should not feel that they are on the lower end of a hierarchy. But Gordon thinks that most people don’t feel resentment from gifts, but gratitude.

For high culture, money is a high-cost mode of incentive, when produced only through bureaucracy/advance permission, in contexts that require sponteneity. Gift is especially important when it forms the context for creativity—consider the expressionists, who worked for each other, to challenge, support, and teach each other.

She is proposing a comparative heuristic: a particular variant of commons/user innovation. When is gift a useful way to think?

Victoria Stodden: Free Revealing in Computational Science

Scientific output is changing: traditional view was hypothesisàexperimentàfinal paper, the last of which was what was shared. Now there are new communication mechanisms. Can have the same hypothesis, but now have the ability to share results, code, data, along with the final paper.

Why don’t scientists avail themselves of these communications technologies? Possible explanations: (1) Scientists are primarily motivated by personal gain/loss. (2) Scientists are worried about being scooped.

Survey of computational scientists in the subfield of machine learning, sampling American academics registered at the top Machine Learning conference, students and professors. (This allowed her to limit the inquiry to those subject to the same IP regime.) 290 surveys, 60 responses so far, still coming in.

Biggest reason not to share code/data: the time it takes to put it into a form they’re comfortable sharing and that they think that other people can use—documenting and cleaning it up. Dealing with questions from users is another significant reason to avoid sharing—also a private incentive. Less significant: worry of not receiving attribution, other private incentives.

Top reasons to share: communitarian norms—encourage scientific advancement, encourage sharing in others, improve the caliber of research, be a good community member. 82%/67% (code and data) also cited private incentive of attribution.

Surprise: scientists are not that worried about being scooped. Private incentives appear to be key to not sharing, while less important to sharing. (Hmm. The attribution reason to share was just as popular as several of the communitarian reasons; respondents didn’t have to pick just one reason to share, so it seems that private incentives could still be quite important to motivate sharing.)

Karim Lakhani: The Patterns of Innovation Generation in a Collaborative Community: Exploring the Relationship between Knowledge Novelty and Reuse

Friendly collaborative competition among software programmers, in a wiki-like setting where code can immediately be shared and evaluated. Objective evaluation of performance; a winner is declared at the end and the code is completely traceable. 11 contests, with over 100 participants in each and over 1500 entries. Question: how does individual action in writing code impact community reuse of the code?

The two faces of innovation: generating new knowledge and reusing existing knowledge. Sources of knowledge for the individual: generate it de novo, generate novel combinations of existing knowledge, and borrow existing knowledge of others to solve the problem. Structuring the resulting artifact can be looked at in terms of complexity/modularity, as well as in terms of conformance to standards that may exist.

Experiment: a one week programming contest; you can view anyone’s entry and take their code and resubmit it with your own changes. Three phases: darkness, when you work on your own without any idea of who’s competing with you; twilight, when you can see how you rank compared to others and see who’s currently contributed the best entry; daylight, when you can see all the other code and modify it. Example: winning entry came from Yi Cao, who participated in the entire contest but showed stunning improvement at the end. Of its 545 lines, his winning entry has only 12 new lines; the rest appeared at least once in other entries, from 30 other authors. Heterogeneity of endowments: many contributors never had the “best” entry, but their contributions were still part of the best entry at the end.

There were typically several leaders over time. 4402 entries, of which 181 were leaders at some point. But the nonleaders were important to the final result.

6% of the entries become top performers. De novo code is typically very limited over time. De novo code is statistically related to top performance and to social value (reused more often). Novel combinations of existing code: also true. But if you borrow code from others, that hurts top performance, but the social value of your new code increases because people can understand your code better because it’s more familiar.

Complexity: the more complex the code, the higher the performance and the social value—but this is a small contest with only about 800 lines of code. The less you conform to standards, the more likely you are to be a top performer, but that doesn’t correlate with social value: conforming code has more social value because it’s more understandable.

There is alignment between individuals generating new code and new combinations and the value of the collective: free riding hurts individual performance, but it’s still good for the community. Transparency—being able to see the code—may be more important than complexity/modularity. If you can see it, even if it’s a jumble, you may be able to break it up into workable chunks afterwards.


Carliss Baldwin: Two views of innovation: large independent chunks, or distributed and componentized.

Pam Samuelson: copyright/patent divide—for copyright, you need a work, while patent can be more divided up.

Baldwin: But there’s still the issue of whether the creative process is one of an individual creating a separate thing.

Samuelson: but patents often cover single components, so it’s more incremental.

Kathy Strandburg: Patents are often talked about in the mode of the romantic inventor, but the literature does recognize that patents are regularly combinations of existing things.

Brett Frischmann: When we talk about patent pools, we should recognize that sometimes sharing tacit knowledge, information about demand, and other features are more valuable results of pooling than the patents themselves.

No comments: