The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 49
The Economics of Artificial Intelligence Page 49

by Ajay Agrawal

to rule their countries many decades after they had lost the cooperation

  of their masses. And they did not have super- smart robots to help them.

  If the future elite of countries that are willing to protect their rents from

  owning the economy’s productive assets (machines) study history’s success-

  ful autocrats well enough (or their machines do), this could go on for quite

  a while.

  In contrast, where the machines are nationally owned, and where the

  rents are shared by all society’s members, what I will call inclusive societies,

  there is no reason that we cannot have equality in consumption. The very

  good, incentive- based reasons for inequality to exist under capitalism will

  no longer apply.

  The Political Economic Source of Future Human Work

  What will humans do for work in a world where machines are better at

  doing everything than humans? It would seem that the obvious answer is

  nothing. We will have to learn to create meaning from non- work- related

  activities, and hopefully overcome our evolved proclivity toward equating

  personal value with social productivity. I am going to argue that this obvious

  286 Patrick Francois

  answer is wrong. There will actually be vital and important work for humans

  to do in this world, and that the amount of it to be done will be greatest in

  the most inclusive societies.

  Managing the Machines Will Be the Source of Human Work

  Why would machines need managing? The machines will be self-

  replicating, self- maintaining, self- creating, self- repairing, self- improving,

  so what else needs to be done? What is not so clear is which ends the

  machines are pursuing.

  Usually we tend to think in terms of well- defi ned human objectives, and

  for most of these it is a nonquestion as to what machines should do. For

  example, oncology machines will read MRIs, diagnose potential cancers,

  order more tests, or operations, or drugs, and so forth, based on protocols

  they have learned by being run millions of times on training data. They can

  learn what to do because objectives here are relatively simple, and success

  in meeting them can be used to determine optimal actions easily. So these

  machines with very narrow objectives need relatively little managing.

  But machines will be producing all output and services in our economy,

  and while doing this will all the while continually reinvent and modify them-

  selves in pursuit of objectives that were programmed in to them by their

  human masters. So we will have a complex set of evolving machines who are

  not only running all production, but doing all inventing as well. We could

  think of these machines as designed, but through the process of machine

  learning and machine- based innovation the designs would become far

  removed from anything imagined by the last generation of human design-

  ers that worked on them. Even understanding what they are doing will be

  diffi

  cult for us humans. Perhaps we will develop intuitions about them, a

  richer human language, or narratives about what they do that will give us

  some vague understandings of what they are about, but it is reasonable to

  suppose that no human will fully understand them.

  The question is, Will we be willing to let this design direction simply con-

  tinue without human interjection? I would argue that we will not. We (our

  societal “we”) will be greatly concerned about the direction that this design

  takes, and managing this direction will require immense human oversight.

  The more so, the more inclusive a society is. But why would we need to

  manage it if we have already programmed in to these advanced machines

  a set of objectives that are human centred? If we have already delegated

  that to the machines? I am assuming that, as part of this programming, we

  will fi nd fail- safes to short- circuit rogue machines following objectives that

  do not advance human welfare, as interestingly sketched by Nick Bostrom

  (2014), so I am explicitly excluding that particular dystopia.

  But even with such fail- safes, additional human involvement will be

  required. This is because we cannot delegate a particular objective function

  to machines and be done with it, because whatever delegation that we imple-

  Comment 287

  ment at time t, based on an objective articulated with the knowledge we have

  at time t, may well be outdated by time tʹ > t because either our knowledge or our values have changed by tʹ. We will need people (obviously greatly

  aided by machines) charged with working out what our social consensus is

  at time tʹ, informing other citizens at tʹ what relevant information they need to make their decisions then, and then implementing those changes at time

  tʹ. These actions, which would of course be simple for machines to do since

  they will be so much smarter than us, will be inherently nonimplementable

  by the machines that are doing all our inventing and production at time

  tʹ, because those machines will have been programmed with the objective

  functions of time t society, which is precisely what we wish to countenance

  changing at time tʹ.

  The whole problem is that writing objectives at time t may lead machines

  to evolve capacities based on those objectives that become outdated at tʹ. In

  order for us to know whether they are outdated at tʹ, we have to fi rst develop a conception of what the machines should be doing at tʹ, and how that diff ers

  from what we thought at t, and we need to somehow have a sense of what the

  machines are actually doing at tʹ and how it diff ers from t. All of these things are collective human decisions, and will require immense human eff ort.

  For example, suppose we program in to these advanced machines an

  objective of maximizing human welfare defi ned in a utilitarian way in the

  year 2035. The designing machines will then set off to come up with machine

  improvements that advance our utilitarian human objectives. But in doing

  so, they may end up doing some violence to other objectives which, on the

  whole we were ready as a society to subordinate to sound utilitarian ones in

  2035, but are no longer willing to countenance in 2050. For instance, it may

  be the case that the utilitarian- based inventing machines put no weight on

  animal welfare, other than how it indirectly advances the utilitarian goal.

  But it could be that our societal objectives, beliefs, views and so forth have

  evolved in the intervening years. Maybe we come to learn something more

  about animal neurology, or maybe we just change our values as we become

  richer. And then people, on the whole, start to want to privilege other mam-

  mals as much as ourselves. Or alternatively perhaps we become so impressed

  with the complexity of machines that we want to countenance nonorganic

  life as of value in itself. In either such case, we will need to, as human deci-

  sion makers, understand enough of what machines are doing in pursuit of

  some of our earlier objectives to be able to see whether the societal objectives

  unstated in 2035 are being trammelled upon or not in 2050. They may not

  be, and in that case nothing much needs to change. But how will we know

  without checking?

  That will be very complicated to
do. It fi rstly requires some humans trying

  to understand just what it is that the machines are doing in 2050: How they

  are evolving and what they have been up to? We then need to work out what

  the relevant parts of that information are for our societal decision makers

  288 Patrick Francois

  to know, and in inclusive societies “societal decision makers” are a lot of

  people. We then need to fi nd a way of communicating this perhaps highly

  sophisticated information to these decisions makers, some, and perhaps

  many, of whom have very little technical training about machine function,

  so that they can make their decisions based on the knowledge and training

  that they do have.

  This process also, of course, begs the question as to who “we” as a set

  of societal decision makers are in this context, and what “we” want. Some

  humans must be involved in making these ethical and social decisions. And

  here I do not mean decisions of the form whether a car should collide with

  and kill three old citizens instead of a pregnant mother, which is of course

  diffi

  cult, but which we at least implicitly grapple with every day. But I mean

  the more basic decisions as to what is the societal objective that the network

  of machines that are not only producing everything for us, but also designing

  and inventing everything for us are trying to attain. One could argue that

  we also implicitly engage in such decisions today as a society, for example,

  when we elect politicians or parties with competing platforms. However,

  in the future it will be much more explicit, as our collective stance on these

  things will be needed to determine precisely what direction we will orient

  our machine inventors to head towards every single day.

  It will not be possible (or prudent if it were possible) to delegate this set

  of conversations and tasks to machines alone. Even though they may be

  demonstrably smarter and hence better at making those decisions given

  a well- defi ned objective function, the point is that there is and never will

  be such a well- defi ned social objective function (we have known this since

  Arrow’s impossibility theorem). We need to modify it via our political

  processes in a continual way, and the objective function followed by the

  machines will need to be adjusted in refl ection of a social conversation that

  occurs amongst humans. In inclusive societies, where presumably all citi-

  zens will have a voice in those decisions, this will involve a lot of people, all

  of whom will have to be informed so that they can weigh in on that social

  consensus.

  Managing that conversation, reporting back to “us” what is relevant for

  that conversation emerging from the self- directed world of machines, and

  then adjusting the trajectory of the machines in light of what “we” decide

  via whatever social mechanisms we come up with to express as our collective

  will, must require humans at certain critical points. Human decision making

  will not be replicable or replaceable by machines here almost by defi nition.

  So, to summarize, I am describing a world that we are admittedly far from

  today. A world in which most human labor is involved in the set of essentially

  political tasks related to managing the machines that will be doing all the

  production in our economy, and hence determining much of our societies’

  directions. A set of people will need to work at determining just what our

  current machines are doing and making that intelligible to social decision

  Comment 289

  makers (which in inclusive societies will be a lot of citizens). Another set of

  people will need to work out how the diverse sets of opinions manifested

  by citizens maps back to a consensus about what our machines should be

  doing, and what directions they should be heading toward. All of these

  workers will be helped by machines, but the machines helping them will

  need human guidance since they will not be using objective protocols that

  could ever be unchanging. This is because it is the very protocols that the

  machines are using that we humans must be constantly discussing changing.

  Humans, though immeasurably dumber than machines, will be essential and

  nonsubstitutable in that process.

  References

  Acemoglu, Daron, and Pascual Restrepo. 2016. “The Race between Man and

  Machine: Implications of Technology for Growth, Factor Shares and Employ-

  ment.” Unpublished manuscript, Massachusetts Institute of Technology.

  Bostrom, Nick. 2014. Superintelligence: Paths, Dangers, Strategies. Oxford: Oxford University Press.

  10

  Artifi cial Intelligence and Jobs

  The Role of Demand

  James Bessen

  There is widespread concern today that artifi cial intelligence technologies

  will create mass unemployment during the next ten or twenty years. One

  recent paper concluded that new information technologies will put “a sub-

  stantial share of employment, across a wide range of occupations, at risk in

  the near future” (Frey and Osborne 2017).

  The example of manufacturing decline provides good reason to be con-

  cerned about technology and job losses. In 1958, the broadwoven textile

  industry in the United States employed over 300,000 production workers,

  and the primary steel industry employed over 500,000. By 2011, broadwoven

  textiles employed only 16,000, and steel employed only 100,000 production

  workers.1 Some of these losses can be attributed to trade, especially since the

  mid- 1990s. However, overall since the 1950s, most of the decline appears to

  come from technology and changing demand (Rowthorn and Ramaswamy

  1999).

  But the example of manufacturing also demonstrates that the eff ect of

  technology on employment is more complicated than a simple story of

  “automation causes job losses” in the aff ected industries. Indeed, fi gure 10.1

  shows how textiles, steel, and automotive manufacturing all enjoyed strong

  employment growth during many decades that also experienced very rapid

  productivity growth. Despite persistent and substantial productivity growth,

  these industries have spent more decades with growing employment than

  James Bessen is Executive Director of the Technology & Policy Research Initiative at Boston University School of Law.

  For acknowledgments, sources of research support, and disclosure of the author’s material fi nancial relationships, if any, please see http:// www .nber .org/ chapters/ c14029.ack.

  1. These fi gures are for the broadwoven fabrics industry using cotton and manmade fi bers, SIC 2211 and 2221, and the steel works, blast furnaces, and rolling mills industry, SIC 3312.

  291

  Fig. 10.1 Production employment in three industries

  Artifi cial Intelligence and Jobs: The Role of Demand 293

  with job losses. This “inverted- U” pattern appears to be quite general for

  manufacturing industries (Buera and Kaboski 2009; Rodrik 2016).2

  The reason automation in textiles, steel, and automotive manufac-

  turing led to strong job growth has to do with the eff ect of technology on

  demand, as I explore below. New technologies do not just replace labor with

  machines, but, in a competitive market, automation wi
ll reduce prices. In

  addition, technology may improve product quality, customization, or speed

  of delivery. All of these things can increase demand. If demand increases

  suffi

  ciently, employment will grow even though the labor required per unit

  of output declines.

  Of course, job losses in one industry might be off set by employment

  growth in other industries. Such macroeconomic eff ects are covered by

  other articles in this volume (chapter 13, chapter 9). This chapter explores

  the eff ect of technology on employment in the aff ected industry itself. The

  rise and fall of employment poses an important puzzle. While a substan-

  tial literature has looked at structural change associated with technology, I

  argue that the most widely accepted explanations for deindustrialization are

  inconsistent with the observed historical pattern. To explain the inverted-

  U pattern, I present a very simple model that shows why demand for these

  products was highly elastic during the early years and why demand became

  inelastic over time. This model forecasts the rise and fall of employment in

  these industries with reasonable accuracy: the solid line in fi gure 10.1 shows

  those predictions. I then explore the implications of this model for the future

  impact of artifi cial intelligence over the next two decades.

  10.1 Structural

  Change

  The inverted- U pattern in fi gure 10.1 is also seen in the relative share of

  employment in the whole manufacturing sector, shown in fi gure 10.2. Logi-

  cally, the rise and fall of the sector as a whole in this chart results from the

  aggregate rise and fall of separate manufacturing industries such as those in

  fi gure 10.1. Yet, explanations of this phenomenon based on broad sector-

  level factors face a challenge because individual industries show rather dispa-

  rate patterns. For example, employment in the automotive industry appears

  to have peaked nearly a century after textile employment peaked. Data on

  individual industries are needed to analyze such disparate responses.

  The literature on structural change provides two sorts of accounts for

  the relative size of the manufacturing sector, one based on diff erential rates

  of productivity growth, the other based on diff erent income elasticities of

 

‹ Prev