Book Read Free

The Economics of Artificial Intelligence

Page 80

by Ajay Agrawal

elasticity of substitution (CES) production function so that there is love

  of variety. At any date t there is a measure N( t) of varieties. The marginal returns to new varieties are positive, but diminishing. The key “building

  on the shoulders of giants” externality is that the cost of developing a new

  variety is inversely proportional to the measure of varieties. As a result, inno-

  vation costs fall over time, generating endogenous growth. A one- sector,

  two- country extension appears in Rivera- Batiz and Romer (1991). A two-

  sector, two- country extension appears in Grossman and Helpman (1991).

  This brief review leads to a number of observations. As in the previous

  section, the benefi t of trade policy depends on whether the externality

  operates at the national or international levels; Q of the previous section is replaced here by either n( t) or N( t). Hence, if each fi rm builds on the international frontier n( t) or the international number of varieties N( t), then there are no implications for comparative advantage; however, if each fi rm builds

  on its national n( t) or national N( t) then the frontier country will develop an increasingly strong comparative advantage in the quality or expanding-varieties sector. With national- level externalities one country will capture

  the lion’s share of the quality/ varieties sector. Further, a country can capture

  this sector by using R&D and trade policies.

  Endogenous growth models provide important insights into the details

  of R&D and trade policies. Research and development policies directly tar-

  get the knowledge externality and so are preferred to (second- best) trade

  policies. One R&D policy avenue is to promote knowledge diff usion. This

  can be done through subsidies to nonprofi t organizations targeting local

  within- industry interactions and industry- university collaborations. A sec-

  ond R&D policy avenue is to promote knowledge creation through R&D

  subsidies that are available to all fi rms, universities, and students. There

  is a tension between these two avenues; knowledge diff usion can discour-

  age knowledge creation since knowledge diff usion to competitors reduces

  the returns to innovation. However, the tension is sometimes constructive:

  Silicon Valley emerged from the shadows of Massachusetts’ Route 128

  partly because of an “open- source attitude” (Saxenian 1994) and Califor-

  16. Placing endogenous growth into a two- sector model so as to facilitate a discussion of comparative advantage is not easy because the sector with improving quality slowly takes over the entire economy unless other price or nonprice “congestion” forces prevent this.

  Artifi cial Intelligence and International Trade 479

  nian restrictions on noncompete clauses (Marx and Fleming 2012). It is less

  likely that diff usion of knowledge to foreign countries will be as benefi cial

  domestically.

  This class of models discourages policies that target individual fi rms

  or that “pick winners.” To understand why industry leaders should not be

  advantaged by policy, note that counterintuitively, industry leaders will be

  the least innovative fi rms due to the “market- stealing” eff ect. If an entrant

  innovates, it steals the market from the leader. If a leader innovates, it canni-

  balizes itself. Leaders therefore have less of an incentive to innovate. Aghion et al. (2001, 2005) address this counterintuitive result by developing a model

  in which leaders innovate in order to escape the competition. Aghion et al.

  (2017) and Lim, Trefl er, and Yu (2017) are currently developing international

  trade models featuring escape the competition.

  In the context of AI, none of the above endogenous growth models is

  ideal, leading us to conjecture about what an appropriate model might look

  like. The advantage of endogenous growth models is that they emphasize

  knowledge creation and diff usion. Thinking more deeply about AI develop-

  ment and commercialization, it is useful to distinguish two aspects of what

  is done in the AI research departments of large fi rms. First, they improve

  AI algorithms, which have the fl avor of quality ladders. (Recall that qual-

  ity can be something that is perceived by consumers or, as is relevant here,

  something that reduces marginal costs.) Second, AI research departments

  develop new applications of existing AI; for example, Google uses AI for

  its search engine, autonomous vehicles, YouTube recommendations, adver-

  tising network, energy use in data centers, and so forth. This suggests an

  expanding- varieties model, but one that operates within the fi rm. We are

  unaware of any endogenous growth models that have both these features.

  Grossman and Helpman (1991) have the fi rst and Klette and Kortum (2004)

  have the second. Combining them in one model is not trivial and analytic

  results would likely have to be replaced with calibration.

  19.3.4 New Economic Geography and Agglomeration

  The discussion in the previous section points to the possibility that

  knowledge spillovers are subnational, and this leads naturally to a theory

  of regional clusters such as Silicon Valley. New economic geography or

  NEG (Krugman 1980) does not typically consider knowledge spillovers, but

  it does consider other local externalities that drive regional clusters. Three

  mechanisms have been particularly prominent: (a) demand- side “home-

  market eff ects,” (b) upstream- downstream linkages, and (c) labor- market

  pooling. All of these theories feature two key elements: costs of trading

  across regions (e.g., tariff s) and increasing returns to scale at the fi rm level

  (which can be thought of as the fi xed costs of developing a new product). We

  explain the role of these two elements in the context of home- market eff ects.

  480 Avi Goldfarb and Daniel Trefl er

  Consider a model with CES monopolistic competition and two regions

  ( j = 1, 2). There are varieties of machines and the larger the set of machines to choose from, the more productive are the producers. Let N be the measure

  j

  of machine varieties available in region j. Then with CES production func-

  tions, productivity is proportional to N .17 The fundamental factor push-

  j

  ing for agglomeration is the strength of this love- of-variety/ productivity

  externality. (This is related to the externality in Romer’s expanding varieties

  model, which is also proportional to N .) As in previous models, the exter-

  j

  nality operates at the local level rather than at the international level. This

  externality encourages fi rms to colocate or agglomerate since the agglomera-

  tion of fi rms drives up N and productivity. The fundamental factor pushing

  j

  against this agglomeration is trade costs: a fi rm can avoid trade costs by

  locating close to consumers rather than close to other producers. The main

  insight of this model is that in equilibrium a disproportionate share of the

  world’s fi rms will locate in a single region, and this region will thus have

  higher productivity. As a result, this region will be richer. Notice that fi rms

  are choosing to set up where the competition is greatest and where wages

  and property values are the highest.

  The above model of agglomeration has been extended in countless ways

  (e.g., Krugman and Venables 1995; Fajgelba
um, Grossman, and Helpman

  2011; Duranton and Puga 2001) and it is easy to think of applications where

  the force for agglomeration is not the variety of machines, but the variety

  of knowledge held by fi rms. If this knowledge is tacit (meaning it cannot

  be codifi ed and transmitted in a document), then knowledge spillovers are

  only transmitted locally via face- to-face interactions. In this case, knowl-

  edge externalities lead fi rms to agglomerate. The result is regions like Silicon

  Valley.

  19.3.5 Cluster

  Policies

  Cluster policies have long been the politician’s best friend, yet economists

  remain highly critical of them. In surveying the evidence for the success of

  these policies, Uyarra and Ramlogan (2012) write “There is no clear and

  unambiguous evidence that over the long term clusters are able to gener-

  ate strong and sustainable impacts in terms of innovation, productivity or

  employment.” One of the world leaders in the economics of clusters, Gilles

  Duranton, titled his 2011 survey “‘California Dreamin’: The Feeble Case

  for Cluster Policies.” Yet clusters remain fashionable.

  In light of what we have described, the fi rst question is: When are cluster

  policies likely to succeed? The answer is that they are most likely to succeed

  when there is clear evidence of scale economies and of knowledge creation

  together with local knowledge diff usion. Artifi cial intelligence displays these

  17. More precisely, productivity is proportional to N 1/ (– 1) where > 1 is the elasticity of substitution between varieties.

  Artifi cial Intelligence and International Trade 481

  characteristics, though the extent of international knowledge diff usion can-

  not be ignored.

  The second question is: What policies are likely to work? To answer this

  question we turn to the insights of Ajay Agrawal, Director of Rotman’s

  Creative Destruction Lab (CDL), and Michael Porter, the business guru

  of cluster policies. We start with Agrawal. Agrawal identifi es two problems

  with developing AI in the Canadian context. First, there is a shortage of

  people with the skills to scale up companies. Agrawal calls these people

  1000Xers. Second, the cost of information about a start-up’s quality is so

  high that capital markets cannot identify the best and the brightest start-ups.

  Agrawal’s CDL addresses both of these problems by linking start-ups with

  serial entrepreneurs who can identify a good start-up, tap into 1000Xers for

  growth, and pass on valuable information about start-up quality to inves-

  tors globally.

  Another approach to the question of what policies are likely to work uti-

  lizes Porter’s (1990) diamond, which emphasizes four features of clusters:

  ( a) factor conditions such as universities and an abundant supply of AI sci-

  entists, ( b) home- market- demand externalities for AI, ( c) externalities fl owing from suppliers of specialized intermediate inputs into AI such as fi nan-

  cial services, and ( d ) a competitive environment. Items b– d involve eff ects that have already been described in our discussion of knowledge spillovers

  and lie at the heart of local agglomeration. Item a is a more conventional

  economic factor, that is, drive down the price of the key input by subsidizing

  its supply. Yet Porter’s research shows that many clusters are driven primarily

  by a. That is to say, the single most important policy in practice is simple:

  follow Hinton’s advice in training a large number of AI scientists locally.

  Our models also suggest two diffi

  culties with Hinton’s advice that must

  be shored up. First, there is international rather than national knowledge

  diff usion due to the fact that, for example, Canadian- trained scientists are

  likely to leave Canada for Silicon Valley, China, and other AI hotspots. This

  suggests value in programs like those used successfully in Singapore that

  require student loans to be repaid if the student does not work in Singapore

  for a minimum number of years.

  Second, scale in data is a huge problem for a small country like Canada.

  To understand appropriate solutions for this, we now turn to the details of

  national regulatory environments that aff ect data and the use of AI.

  19.4

  Behind- the- Border Trade Barriers:

  The Domestic Regulatory Environment

  Given these models, we next turn to the specifi c regulatory issues that

  are likely to impact trade policy. Many of the core trade issues around AI

  involve access to data. Data is a key input into AI, and there are a number

  of government policies that aff ect data access and data fl ows. To the extent

  482 Avi Goldfarb and Daniel Trefl er

  these regulations vary across countries, they can advantage some countries’

  AI industries. The models above suggest that this advantage can have con-

  sequences if there are economies of scale, local externalities, and/or rents.

  We highlight fi ve policies in particular. The fi rst three involve data: domes-

  tic privacy policy, data localization rules, and access to government data.

  The others are development of the regulation of AI application industries

  (such as autonomous vehicles) and protection of source code. Privacy policy,

  data localization, and source code access have already become signifi cant

  trade issues. For example, the TPP addresses all three of these, as do the US

  Trade Representative’s NAFTA renegotiation objectives. The US position

  is that strong Canadian and Mexican privacy rules, localization require-

  ments, and access to foreign source code are all impediments to US exports

  of AI- related goods. In other words, the emphasis on trade policy in these

  areas is that regulation could be disguised protection that helps domestic

  fi rms and hurts foreign fi rms. In the discussion below, we explore the extent

  to which this starting assumption is appropriate.

  Privacy Regulation. Privacy regulation involves policies that restrict the

  collection and use of data. Such regulation diff ers across locations. Privacy

  policy has the power to limit or expand the ability of fi rms to use AI eff ec-

  tively. Restrictions on the use of data mean restrictions on the ability to use

  AI given the data available; however, restrictions on the use of data may also

  increase the supply of data available if it leads consumers to trust fi rms that

  collect the data. Although the theory is ambiguous, thus far, the empirical

  evidence favors the former eff ect on balance. Stricter privacy regulations

  reduce the ability of fi rms and nonprofi ts to collect and use data and there-

  fore leads to less innovative use of data (Goldfarb and Tucker 2012). Thus,

  fi rms in some countries may benefi t from favorable privacy policy.

  We believe the most useful analogies for privacy policy in trade relate

  to labor and environmental regulations. Such regulations also diff er across

  countries for a variety of reasons. They could refl ect diff erences in prefer-

  ences across countries, or could be perceived as normal goods that wealthier

  countries are willing to pay for but poorer countries are not (Grossman and

  Krueger 1995). There is room for reasonable disagreement on how data

  might b
e collected or used. Some countries will restrict the information

  used in prediction while others will not. For example, for insurance, the data

  that can be used varies by state, with diff erent states providing a variety of

  restrictions on the use of race, religion, gender, and sexual orientation in

  insurance.18 Even with such restrictions, if other variables provide surrogates

  for such categories, it is possible that fi rms may be forced to abandon AI

  methods entirely for more transparent prediction technologies. In terms of

  18. http:// repository.law.umich .edu/ cgi/ viewcontent.cgi?article=1163&context=law_econ _current.

  Artifi cial Intelligence and International Trade 483

  privacy policy, we think it is useful to take as given that there are diff erences

  across countries in their preferences for policies that restrict the collection

  and use of data.

  Given these diff erences in preferences, what are the implications for trade?

  Suppose that the optimal privacy policy for growing an AI industry involves

  relatively few restrictions on data. Artifi cial intelligence requires data, and

  so the fewer government restrictions on data collection, the more rapidly the

  industry grows.19 To the extent that young fi rms tend to grow by focusing

  on the domestic market, this will advantage the growth of AI fi rms in some

  countries relative to others. Thus, lax privacy policies may help domestic

  industry relative to countries with strict policies just as lax labor and envi-

  ronmental regulation may help the domestic industry.

  This suggests the potential of a “race to the bottom” in privacy policy.

  Evidence for such races has been found in enforcement of labor policies

  (e.g., Davies and Vadlamannati 2013) and in environmental policies (e.g.,

  Beron, Murdoch, and Vijverberg 2003; Fredriksson and Milliment 2002).

  There is evidence that privacy regulation does disadvantage jurisdictions

  with respect to their advertising- supported software industries. In par-

  ticular, Goldfarb and Tucker (2011) examined a change in European privacy

  regulation (implemented in 2004) that made it more diffi

  cult for European

  internet fi rms to collect data about their online customers. This regulatory

  change was particularly likely to reduce the eff ectiveness of advertising on

 

‹ Prev