The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 21
The Economics of Artificial Intelligence Page 21

by Ajay Agrawal


  Intelligence on Innovation

  An Exploratory Analysis

  Iain M. Cockburn, Rebecca Henderson, and Scott Stern

  4.1 Introduction

  Rapid advances in the fi eld of artifi cial intelligence have profound implica-

  tions for the economy as well as society at large. These innovations have the

  potential to directly infl uence both the production and the characteristics of

  a wide range of products and services, with important implications for pro-

  ductivity, employment, and competition. But, as important as these eff ects

  are likely to be, artifi cial intelligence also has the potential to change the

  innovation process itself, with consequences that may be equally profound,

  and which may, over time, come to dominate the direct eff ect.

  Consider the case of Atomwise, a start-up fi rm that is developing novel

  technology for identifying potential drug candidates (and insecticides) by

  using neural networks to predict the bioactivity of candidate molecules. The

  company reports that its deep convolutional neural networks “far surpass”

  the performance of conventional “docking” algorithms. After appropri-

  ate training on vast quantities of data, the company’s AtomNet product

  is described as being able to “recognize” foundational building blocks of

  Iain M. Cockburn is the Richard C. Shipley Professor of Management at Boston University and a research associate of the National Bureau of Economic Research. Rebecca Henderson is the John and Natty McArthur University Professor at Harvard University, where she has a joint appointment at the Harvard Business School in the General Management and Strategy units, and a research associate of the National Bureau of Economic Research. Scott Stern is the David Sarnoff Professor of Management and chair of the Technological Innovation, Entrepreneurship, and Strategic Management Group at the MIT Sloan School of Management, and a research associate and director of the Innovation Policy Working Group at the National Bureau of Economic Research.

  We thank Michael Kearney for extraordinary research assistance. For acknowledgments,

  sources of research support, and disclosure of the authors’ material fi nancial relationships, if any, please see http:// www .nber .org/ chapters/ c14006.ack.

  115

  116 Iain M. Cockburn, Rebecca Henderson, and Scott Stern

  organic chemistry, and is capable of generating highly accurate predictions

  of the outcomes of real- world physical experiments (Wallach, Dzamba,

  and Heifels 2015). Such breakthroughs hold out the prospect of substantial

  improvements in the productivity of early stage drug screening. Of course,

  Atomwise’s technology (and that of other companies leveraging artifi cial

  intelligence to advance drug discovery or medical diagnosis) is still at an

  early stage: though their initial results seem to be promising, no new drugs

  have actually come to market using these new approaches. But whether or

  not Atomwise delivers fully on its promise, its technology is representa-

  tive of the ongoing attempt to develop a new innovation “playbook,” one

  that leverages large data sets and learning algorithms to engage in precise

  prediction of biological phenomena in order to guide design- eff ective inter-

  ventions. Atomwise, for example, is now deploying this approach to the

  discovery and development of new pesticides and agents for controlling

  crop diseases.

  Atomwise’s example illustrates two of the ways in which advances in arti-

  fi cial intelligence have the potential to impact innovation. First, though the

  origins of artifi cial intelligence are broadly in the fi eld of computer science,

  and its early commercial applications have been in relatively narrow domains

  such as robotics, the learning algorithms that are now being developed sug-

  gest that artifi cial intelligence may ultimately have applications across a very

  wide range. From the perspective of the economics of innovation (among

  others, Bresnahan and Trajtenberg 1995), there is an important distinction

  between the problem of providing innovation incentives to develop tech-

  nologies with a relatively narrow domain of application, such as robots

  purpose- built for narrow tasks, versus technologies with a wide—advocates

  might say almost limitless—domain of application, as may be true of the

  advances in neural networks and machine learning often referred to as “deep

  learning.” As such, a fi rst question to be asked is the degree to which devel-

  opments in artifi cial intelligence are not simply examples of new technolo-

  gies, but rather may be the kinds of “general purpose technologies” (GPTs)

  that have historically been such infl uential drivers of long- term technologi-

  cal progress.

  Second, while some applications of artifi cial intelligence will surely consti-

  tute lower- cost or higher- quality inputs into many existing production pro-

  cesses (spurring concerns about the potential for large job displacements),

  others, such as deep learning, hold out the prospect of not only productivity

  gains across a wide variety of sectors, but also changes in the very nature

  of the innovation process within those domains. As articulated famously

  by Griliches (1957), by enabling innovation across many applications,

  the “invention of a method of invention” has the potential to have much

  larger economic impact than development of any single new product. Here

  we argue that recent advances in machine learning and neural networks,

  through their ability to improve both the performance of end- use technolo-

  The Impact of Artifi cial Intelligence on Innovation 117

  gies and the nature of the innovation process, are likely to have a particularly

  large impact on innovation and growth. Thus the incentives and obstacles

  that may shape the development and diff usion of these technologies are an

  important topic for economic research, and building an understanding of

  the conditions under which diff erent potential innovators are able to gain

  access to these tools and to use them in a procompetitive way is a central

  concern for policy.

  This chapter begins to unpack the potential impact of advances in arti-

  fi cial intelligence on innovation, and to identify the role that policy and

  institutions might play in providing eff ective incentives for innovation, dif-

  fusion, and competition in this area. We begin in section 4.2 by highlighting

  the distinctive economics of research tools, of which deep learning applied

  to research and development (R&D) problems is such an intriguing example.

  We focus on the interplay between the degree of generality of application

  of a new research tool and the role of research tools not simply in enhanc-

  ing the effi

  ciency of research activity, but in creating a new “playbook” for

  innovation itself. We then turn in section 4.3 to briefl y contrast three key

  technological trajectories within artifi cial intelligence (AI)—robotics, sym-

  bolic systems, and deep learning. We propose that these often confl ated fi elds

  will likely play very diff erent roles in the future of innovation and techni-

  cal change. Work in symbolic systems appears to have stalled and is likely

  to have relatively little impact going forward. And while developments in

  robotics have the potential t
o further displace human labor in the production

  of many goods and services, innovation in robotics technologies per se has

  relatively low potential to change the nature of innovation itself. By contrast,

  deep learning seems to be an area of research that is highly general purpose

  and has the potential to change the innovation process itself.

  We explore whether this might indeed be the case through an examina-

  tion of some quantitative empirical evidence on the evolution of diff erent

  areas of artifi cial intelligence in terms of scientifi c and technical outputs

  of AI researchers as measured (imperfectly) by the publication of papers

  and patents from 1990 through 2015. In particular, we develop what we

  believe is the fi rst systematic database that captures the corpus of scientifi c

  paper and patenting activity in artifi cial intelligence, broadly defi ned, and

  divides these outputs into those associated with robotics, symbolic systems,

  and deep learning. Though preliminary in nature (and inherently imperfect

  given that key elements of research activity in artifi cial intelligence may

  not be observable using these traditional innovation metrics), we fi nd strik-

  ing evidence for a rapid and meaningful shift in the application orientation

  of learning- oriented publications, particularly after 2009. The timing of

  this shift is informative, since it accords with qualitative evidence about the

  surprisingly strong performance of so-called “deep learning” multilayered

  neural networks in a range of tasks including computer vision and other

  prediction tasks.

  118 Iain M. Cockburn, Rebecca Henderson, and Scott Stern

  Supplementary evidence (not reported here) based on the citation pat-

  terns to authors such as Geoff rey Hinton, who are leading fi gures in deep

  learning, suggests a striking acceleration of work in just the last few years

  that builds on a small number of algorithmic breakthroughs related to multi-

  layered neural networks.

  Though not a central aspect of the analysis for this chapter, we further fi nd

  that, whereas research on learning- oriented algorithms has had a slow and

  steady upward swing outside of the United States, US researchers have had

  a less sustained commitment to learning- oriented research prior to 2009,

  and have been in a “catch-up” mode ever since.

  Finally, we begin to explore some of the organizational, institutional,

  and policy consequences of our analysis. We see machine learning as the

  “invention of a method of invention” whose application depends, in each

  case, on having access not just to the underlying algorithms, but also to

  large, granular data sets on physical and social behavior. Developments in

  neural networks and machine learning thus raise the question of, even if the

  underlying scientifi c approaches (i.e., the basic multilayered neural networks

  algorithms) are open, prospects for continued progress in this fi eld—and

  commercial applications thereof—are likely to be signifi cantly impacted by

  terms of access to complementary data. Specifi cally, if there are increasing

  returns to scale or scope in data acquisition (there is more learning to be

  had from the larger data set), it is possible that early or aggressive entrants

  into a particular application area may be able to create a substantial and

  long- lasting competitive advantage over potential rivals merely through

  the control over data rather than through formal intellectual property or

  demand- side network eff ects. Strong incentives to maintain data privately

  has the additional potential downside that data is not being shared across

  researchers, thus reducing the ability of all researchers to access an even

  larger set of data that would arise from public aggregation. As the competi-

  tive advantage of incumbents is reinforced, the power of new entrants to

  drive technological change may be weakened. Though this is an important

  possibility, it is also the case that, at least so far, there seems to be a signifi cant amount of entry and experimentation across most key application sectors.

  4.2 The Economics of New Research Tools: The Interplay between

  New Methods of Invention and the Generality of Innovation

  At least since Arrow (1962) and Nelson (1959), economists have appreci-

  ated the potential for signifi cant underinvestment in research, particularly

  basic research or domains of invention with low appropriability for the

  inventor. Considerable insight has been gained into the conditions under

  which the incentives for innovation may be more or less distorted, both in

  terms of their overall level and in terms of the direction of that research.

  As we consider the potential impact of advances in AI on innovation, two

  The Impact of Artifi cial Intelligence on Innovation 119

  ideas from this literature seem particularly important—the potential for

  contracting problems associated with the development of a new broadly

  applicable research tool, and the potential for coordination problems aris-

  ing from adoption and diff usion of a new “general purpose technology.”

  In contrast to technological progress in relatively narrow domains, such as

  traditional automation and industrial robots, we argue that those areas of

  artifi cial intelligence evolving most rapidly—such as deep learning—are

  likely to raise serious challenges in both dimensions.

  First, consider the challenge in providing appropriate innovation incen-

  tives when an innovation has potential to drive technological and organiza-

  tional change across a wide number of distinct applications. Such general

  purpose technologies (David 1990; Bresnahan and Trajtenberg 1995) often

  take the form of core inventions that have the potential to signifi cantly

  enhance productivity or quality across a wide number of fi elds or sectors.

  David’s (1990) foundational study of the electric motor showed that this

  invention brought about enormous technological and organizational change

  across sectors as diverse as manufacturing, agriculture, retail, and residential

  construction. Such GPTs are usually understood to meet three criteria that

  distinguish them from other innovations: they have pervasive application

  across many sectors, they spawn further innovation in application sectors,

  and they themselves are rapidly improving.

  As emphasized by Bresnahan and Trajtenberg (1995), the presence of a

  general purpose technology gives rise to both vertical and horizontal exter-

  nalities in the innovation process that can lead not just to underinvestment

  but also to distortions in the direction of investment, depending on the

  degree to which private and social returns diverge across diff erent appli-

  cation sectors. Most notably, if there are “innovation complementarities”

  between the general purpose technology and each of the application sectors,

  lack of incentives in one sector can create an indirect externality that results

  in a system- wide reduction in innovative investment itself. While the private

  incentives for innovative investment in each application sector depend on

  its the market structure and appropriability conditions, that sector’s innova-

  tion enhances innovation in the
GPT itself, which then induces subsequent

  demand (and further innovation) in other downstream application sectors.

  These gains can rarely be appropriated within the originating sector. Lack

  of coordination between the GPT and application sectors, as well as across

  application sectors, is therefore likely to signifi cantly reduce investment

  in innovation. Despite these challenges, a reinforcing cycle of innovation

  between the GPT and a myriad of application sectors can generate a more

  systemic economy- wide transformation as the rate of innovation increases

  across all sectors. A rich empirical literature examining the productivity

  impacts of information technology (IT) point to the role of the microproces-

  sor as a GPT as a way of understanding the impact of IT on the economy as

  a whole (among many others, Bresnahan and Greenstein 1999; Brynjolfsson

  120 Iain M. Cockburn, Rebecca Henderson, and Scott Stern

  and Hitt 2000; Bresnahan, Brynjolfsson, and Hitt 2002). Various aspects

  of artifi cial intelligence can certainly be understood as a GPT, and learning

  from examples such as the microprocessor are likely to be a useful founda-

  tion for thinking about both the magnitude of their impact on the economy

  and associated policy challenges.

  A second conceptual framework for thinking about AI is the economics

  of research tools. Within the research sectors some innovations open up new

  avenues of inquiry, or simply improve productivity “within the lab.” Some of

  these advances appear to have great potential across a broad set of domains

  beyond their initial application: as highlighted by Griliches (1957) in his clas-

  sic studies of hybrid corn, some new research tools are inventions that do

  not just create or improve a specifi c product—instead, they constitute a new

  way of creating new products with much broader application. In Griliches’s

  famous construction, the discovery of double- cross hybridization “was the

  invention of a method of inventing.” (IMI) Rather than being a means of

  creating a single new corn variety, hybrid corn represented a widely appli-

  cable method for breeding many diff erent new varieties. When applied to

  the challenge of creating new varieties optimized for many diff erent locali-

  ties (and even more broadly, to other crops), the invention of double- cross

 

‹ Prev