The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 94
The Economics of Artificial Intelligence Page 94

by Ajay Agrawal


  McKinsey Global Institute (MGI). Frey and Osborne (2017) attempt to

  determine what jobs may be particularly susceptible to automation and

  to provide an idea of how large an impact automation could have on the

  US labor force. The authors focus particularly on machine learning and its

  application to mobile robotics, and propose a model to predict the extent of

  computerization’s impact on nonroutine tasks, noting potential engineer-

  ing bottlenecks at tasks involving high levels of perception or manipula-

  tion, creative intelligence, and social intelligence. After categorizing tasks

  by their susceptibility to automation, Frey and Osborne map these tasks to

  the O*NET job survey, which provides open- ended descriptions of skills

  and responsibilities involved in an occupation over time. Integrating this

  data set with employment and wage data from the Bureau of Labor Statistics

  (BLS) allows the authors to propose certain subsets of the labor market that

  may be at high, medium, or low risk of automation. The study fi nds that

  47 percent of US employment is at high risk of computerization. It should

  be noted that this study is at an aggregate level and does not examine how

  fi rms may react, any labor saving innovations that could arise, or potential

  productivity or economic growth.

  Frey and Osborne’s work has also been applied by researchers in other

  countries—mapping Frey and Osborne’s occupation- level fi ndings to Ger-

  man labor market data, Brzeski and Burk (2015) suggest that 59 percent of

  German jobs may be highly susceptible to automation, while conducting

  that same analysis in Finland, Pajarinen and Rouvinen (2014) suggest that

  35.7 percent of Finnish jobs are at high risk to automation.

  The Organisation for Economic Co- operation and Development (OECD)

  similarly set out to estimate the automatability of jobs across twenty- one

  OECD countries applying Frey and Osborne to a task- based approach.

  The OECD report argues that certain tasks will be displaced and that the

  extent that bundles of tasks diff er within occupations and across countries

  may make certain occupations less prone to automation than Frey and

  Osborne predicted. Relying upon the task categorization done by Frey and

  Osborne, the authors map task susceptibility to automation to US data from

  the Programme for the International Assessment of Adult Competencies

  (PIAAC), a microlevel data source containing indicators on socioeconomic

  characteristics, skills, job- related information, job tasks, and competencies

  at the individual level. They then construct a model using the PIAAC to

  create a predicted susceptibility to automation based off of the observables

  AI, Labor, Productivity, and the Need for Firm- Level Data 557

  in the PIAAC data to mirror the automatability score that Frey and Osborne

  created. This model is then applied at the worker level across all the PIAAC

  data to predict how susceptible occupations may be to automation. By con-

  ducting the analysis at the individual level, the OECD argues that it is better

  able to account for task variation between individuals within the same occu-

  pation. As a result, the report suggests that Frey and Osborne overestimated

  the extent to which occupations would be susceptible to automation. The

  OECD Report argues that only 9 percent of jobs in the United States and

  across OECD countries will be highly susceptible to automation. The report

  continues to discuss variations across OECD countries, suggesting that the

  percent can range from 6 percent (in Korea) up to 12 percent (in Austria).

  Mann and Püttmann (2017) take a diff erent approach to analyze the

  eff ects of automation on employment. In their study, the authors rely on

  information provided from granted patents. They apply a machine- learning

  algorithm to all US patents granted from 1976 to 2014 to identify patents

  related to automation (an automation patent is defi ned as a “device that

  operates independently from human intervention and fulfi lls a task with rea-

  sonable completion”). They then link the automation patents to the indus-

  tries they are likely to be used in, and identify which areas in the United

  States that these industries are related in. By examining economic indicators

  in comparison to the density of automation patents used in an area, Mann

  and Puttman fi nd that though automation causes manufacturing employ-

  ment to fall, it increases employment in the service sector, and overall has a

  positive impact on employment.

  In June 2017, the McKinsey Global Institute published an independent

  discussion paper examining trends in investment in artifi cial intelligence, the

  prevalence of AI adoption, and how AI is being deployed by companies that

  have started to use the technology (MGI Report 2017). For the purpose of

  their report, the authors adopted a fairly narrow defi nition of AI, focusing

  only on AI technology that is programmed to conduct one set task. The

  MGI report conducted their investigation with a multifaceted approach: it

  surveyed executives at over 3,000 international fi rms, interviewed industry

  experts, and analyzed investment fl ows using third- party venture capital,

  private equity, and mergers and acquisitions data. Using the data collected,

  the MGI report attempts to answer questions regarding adoption by sector,

  size, and geography; to look at performance implications of adoption; and

  to examine potential impacts to the labor market. Though the fi ndings are

  presented at an aggregate level, much of the data, particularly the survey

  of executives, were collected at the fi rm level, allowing for further inquiry

  if one had access.

  In addition to these published works, other researchers have begun to

  examine the eff ect of AI on occupations by looking at its impact on indi-

  vidual abilities and skills. Brynjolfsson, Mitchell, and Rock (forthcoming)

  apply a rubric from Brynjolfsson and Mitchell (2017) that evaluates the

  558 Manav Raj and Robert Seamans

  potential for applying machine learning to tasks to the set of work activities

  and tasks in the Bureau of Labor Statistics’ O*NET occupational data-

  base. With this analysis, they create a “Suitability for Machine Learning”

  for labor inputs in the United States. Similar research by Felten, Raj, and

  Seamans (forthcoming) uses data- tracking progress in artifi cial intelligence

  aggregated by the Electronic Frontier Foundation (EFF) across a variety

  of diff erent artifi cial intelligence metrics and the set of fi fty- two abilities

  in the O*NET occupational database to identify the impact of artifi cial

  intelligence on each of the abilities, and create an occupation- level score

  measuring the potential impact of AI on the occupation. Because the data

  from the EFF is separated by AI metric, this work allows for the investiga-

  tion and simulation of progress in diff erent kinds of AI technology, such as

  image recognition, speech recognition, and ability to play abstract strategy

  games among others.

  The current body of empirical literature surrounding robotics and AI

  adoption is growing, but is still thin, and despite often trying to answe
r

  similar questions, diff erent studies have found disparate results. These dis-

  crepancies highlight the need for further inquiry, replication studies, and

  more complete and detailed data.

  22.3 The Need for Firm- Level Data

  While there is generally a paucity of data examining the adoption, use,

  and eff ects of both AI and robotics, there is currently less information avail-

  able regarding AI. There are no public data sets on the utilization or adop-

  tion of AI at either the macro or micro level. The most complete source of

  information, the MGI study, is proprietary and inaccessible to the general

  public or the academic community.

  The most comprehensive and widely used data set examining the diff usion

  of robotics is the International Federation of Robotics Robot Shipment

  Data. The IFR has been recording information regarding worldwide robot

  stock and shipment fi gures since 1993. The IFR collects this data from its

  members, who are typically large robot manufacturers such as FANUC,

  KUKA, and Yaskawa. The data are broken up by country, year, industry,

  and technological application, which allows for analysis of the industry-

  specifi c impacts of technology adoption. However, the IFR data set has

  shortcomings. The IFR defi nes an industrial robot as an “automatically con-

  trolled, reprogrammable, multipurpose manipulator, programmable in three

  or more axes, which can be either fi xed in place or mobile for use in industrial

  automation applications.”5 This defi nition limits the set of industrial robots

  and ensures that the IFR does not collect any information on dedicated

  industrial robots that serve one purpose. Further, some of the robots are

  5. https:// ifr .org/ standardization.

  AI, Labor, Productivity, and the Need for Firm- Level Data 559

  not classifi ed by industry, detailed data is only available for industrial robots

  (and not robots in service, transportation, warehousing, or other sectors),

  and geographical information is often aggregated (e.g., data exist for North

  America as a category rather than the United States, or an individual state

  within the United States).

  Another issue with the IFR data is the diffi

  culty of integrating it with

  other data sources. The IFR utilizes its own industry classifi cations when

  organizing the data, rather than relying on broadly used identifi ers such as

  the North American Industry Classifi cation System (NAICS). Mapping

  IFR data to other data sets (such as BLS or census data) fi rst requires cross-

  referencing IFR classifi cations to other identifi ers. Industry- level data also

  cannot be used to answer micro- oriented questions about the impacts and

  reaction to technology adoption at the fi rm level.

  While the IFR data are useful for some purposes, particularly examining

  the adoption of robotics by industry and country, its aggregated nature

  obscures diff erences occurring within industries and across regions, mak-

  ing it diffi

  cult to uncover when and how robots might serve as substitutes

  or complements to labor, and obscuring the diff erential eff ects of adoption

  within industries or countries. Additional data is needed to answer the issues

  raised above and to replicate existing studies. In particular, the National

  Academy of Sciences Report (NAS 2017) highlights the need for computer

  capital broken down at the fi rm and occupation level, skill changes over time

  by fi eld, and data on organizational processes as they relate to technology

  adoption.

  The European Manufacturing Survey (EMS) has been organized and exe-

  cuted periodically by a number of research organizations and universities

  across Europe since 2001, and is currently one of the only fi rm- level data

  sets examining the adoption of robotics. The overall objective of the EMS is

  to provide empirical evidence regarding the use and impact of technological

  innovation in manufacturing at the fi rm level. The EMS accomplishes this

  via a survey of a random sample of manufacturing fi rms with at least twenty

  employees across seven European countries (Austria, France, Germany,

  Spain, Sweden, Switzerland, and the Netherlands). While some aspects of

  the survey vary across countries, the core set of questions inquire about

  whether the fi rm uses robots, the intensity of robot usage, and reinvestment

  in new robot technology. Data currently exists for fi ve survey rounds: 2001–

  2002, 2003– 2004, 2006– 2007, 2009– 2010, and 2012– 2013, and has been

  used in reports created by the European Commission to analyze the use of

  robotics and its impact on labor patterns, including wages, productivity,

  and off shoring.

  As of now, the EMS appears to be one of the few data sources that are

  capturing the use of robots and automation at the fi rm level. This provides

  opportunities to analyze microeff ects of robotics technology on fi rm pro-

  ductivity and labor, and to analyze fi rm decision- making following adop-

  560 Manav Raj and Robert Seamans

  tion. However, the EMS has its own limitations. The survey only consid-

  ers industrial robots, and the core questionnaire only asks three questions

  regarding the use of robots in a factory setting. The survey is performed

  at the fi rm rather than establishment level, and the sample size of 3,000

  is quite small. In contrast, the Census’s Annual Survey of Manufacturers

  (ASM) surveys 50,000 establishments annually and 300,000 every fi ve years.6

  Finally, similar to many other existing data sets, the EMS is purely focused

  on the manufacturing industry and does not address technology adoption

  at smaller fi rms with less than twenty employees.

  22.4 Additional Firm- Level Research Questions

  Firm- level data on the use of AI would allow researchers to address a

  host of questions including, but not limited to: the extent to which, and

  under what conditions, AI complement or substitute for labor; how AI aff ect

  fi rm- or establishment- level productivity; which types of fi rms are more or

  less likely to invest in AI; how market structure aff ects a fi rm’s incentives to

  invest in AI; and how adoption is eff ecting fi rm strategies. As the nature of

  work itself changes with increased adoption, researchers can also investi-

  gate how fi rm management has been aff ected, particularly at the lower and

  middle level.

  Additionally, there are many important policy questions that cannot be

  answered without disaggregated data. Some of these questions are related

  to the need to reevaluate how individuals are trained prior to entering the

  workforce. Without an understanding of the changes in worker experience

  resulting from technology adoption, it will be diffi

  cult to craft appropriate

  worker education, job training, and retraining programs. Further, issues

  related to inequality could be examined, particularly with relation to the

  “digital divide” and the eff ects of technology adoption on diff erent demo-

  graphics. There are also unanswered questions regarding the diff erential

  eff ects of adoption on regional economies. For example, the eff ects of AI

  on labor may b
e pronounced in some regions because industries, and even

  occupations within those industries, tend to be geographically clustered

  (Feldman and Kogler 2010). Thus, to the extent that AI or robots substitute

  for labor in certain industries or occupations, regions that rely heavily on

  those industries and occupations for jobs and local tax revenue may suff er.

  Moreover, following the recent fi nancial crisis, unemployment insurance

  reserves in some states have been slow to recover (Furman 2016a). Data on

  the regional adoption of AI could be used to simulate the extent to which

  future adoption may increase unemployment and whether unemployment

  insurance reserves are adequately funded.

  6. The census surveys all 300,000 manufacturing establishments every fi ve years, and a rotat-ing subsample of about 50,000 every year. See: https:// www .census .gov/ programs- surveys/ asm

  / about .html.

  AI, Labor, Productivity, and the Need for Firm- Level Data 561

  Finally, these new technologies may have implications for entrepreneurs.

  Entrepreneurs may lack knowledge of how best to integrate robotics with a

  workforce and often face fi nancing constraints that make it harder for them

  to adopt capital- intensive technologies. In the case of AI, entrepreneurs may

  lack data sets on customer behavior, which are needed to train AI systems.

  Firm- level surveys on the use of AI will help us develop a better understand-

  ing of these and related issues.

  22.5 Strategies for Collecting More Data

  Micro- level data regarding the adoption of AI, robots, and other types

  of automation can be created in a variety of ways, the most comprehensive

  of which would be via a census. Census data would provide information for

  the entire population of relevant establishments, and while the information

  provided would be narrow, quality is likely to be high. Additionally, data

  from the Census Bureau would be highly integrable with other government

  data sources, such as employment or labor statistics from the BLS. Data

  could be collected as a stand- alone inquiry, similar to the Management

  and Organizational Practices (MOPS) survey (see Bloom et al. 2017), or by

  adding questions to existing surveys, similar to work done by Brynjolfsson

  and McElheran (2016), which involved adding questions on data- driven

  decision- making to an existing census survey.

 

‹ Prev