The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 43
The Economics of Artificial Intelligence Page 43

by Ajay Agrawal


  t

  rate of 0.4 percent per year. Regimes last for thirty years. Period 100 is highlighted with a black circle. At this point in time, the capital share is relatively high and growth is relatively low.

  the capital share in the future is lower than in period 100 instead of higher

  can be reversed.

  Summing Up

  Automation—an increase in —can be viewed as a “twist” of the capital-

  t

  and labor- augmenting terms in a neoclassical production function. From

  250 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones Uzawa’s famous theorem, since we do not in general have purely labor-augmenting technical change, this setting will not lead to balanced growth.

  In this particular application (e.g., with < 0), either the capital share or the

  growth rate of GDP will tend to increase over time, and sometimes both.

  We showed one special case in which all tasks are ultimately automated that

  produced balanced growth in the limit with a constant capital share less than

  100 percent. A shortcoming of this case is that it requires automation to be

  the only form of technological change. If, instead, the nature of automa-

  tion itself improves over time—consider the plow, then the tractor, then the

  combine- harvester, then GPS tracking—then the model is best thought of

  as featuring both automation and something like improvements in A . In

  t

  this case, one would generally expect growth not to be balanced. However,

  a combination of periods of automation followed by periods of respite,

  like that shown in fi gure 9.3 does seem capable of producing dynamics at

  least superfi cially similar to what we have seen in the United States in recent

  years: a period of a high capital share with relatively slow economic growth.

  9.3 Artifi cial Intelligence in the Idea Production Function

  In the previous section, we examined the implications of introducing AI in

  the production function for goods and services. But what if the tasks of the

  innovation process themselves can be automated? How would AI interact

  with the production of new ideas? In this section, we introduce AI in the

  production technology for new ideas and look at how AI can aff ect growth

  through this channel.

  A moment of introspection into our own research process reveals many

  ways in which automation can matter for the production of ideas. Research

  tasks that have benefi ted from automation and technological change include

  typing and distributing our papers, obtaining research materials and data

  (e.g., from libraries), ordering supplies, analyzing data, solving math prob-

  lems, and computing equilibrium outcomes. Beyond economics, other ex-

  amples include carrying out experiments, sequencing genomes, exploring

  various chemical reactions and materials. In other words, applying the same

  task- based model to the idea production function and considering the auto-

  mation of research tasks seems relevant.

  To keep things simple, suppose the production function for goods and

  services just uses labor and ideas:

  (19)

  Y = A L .

  t

  t

  t

  But suppose that various tasks are used to make new ideas according to

  1

  1/

  (20)

  A = A

  X di

  where

  < 0.

  t

  t

  it

  0

  Artifi cial Intelligence and Economic Growth 251

  Assuming some fraction of tasks have been automated—using a similar

  t

  setup to that in section 9.2—the idea production function can be ex-

  pressed as

  (21)

  A = A ( B K ) + ( C S )

  (

  )1/ A F BK , C S

  (

  ),

  t

  t

  t

  t

  t

  t

  t

  t

  t

  t

  t

  where S is the research labor used to make ideas, and B and C are defi ned t

  t

  t

  as before, namely, B ≡ (1 ) and C ≡ (1

  )(1 )/ .

  t

  t

  t

  t

  Several observations then follow from this setup. First, consider the case

  in which is constant at some value but then increases to a higher value

  t

  (recall that this leads to a one- time decrease in B and increase in C ). The t

  t

  idea production function can then be written as

  BK

  (22)

  A

  = A S F

  t , C

  t

  t

  t

  St

  ~ A CS ,

  t

  t

  where the “~” notation means “is asymptotically proportional to.” The

  second line follows if K / S is growing over time (i.e., if there is economic t

  t

  growth) and if the elasticity of substitution in F(·) is less than one, which we have assumed. In that case, the CES function is bounded by its scarcest

  argument, in this case researchers. Automation then essentially produces a

  level eff ect but leaves the long- run growth rate of the economy unchanged

  if < 1. Alternatively, if = 1—the classic endogenous growth case—then

  automation raises long- run growth.

  Next, consider this same case of a one- time increase in , but suppose the

  elasticity of substitution in F(·) equals one, so that F(·) is Cobb- Douglas. In this case, as in the Zeira model, it is easy to show that a one- time increase in

  automation will raise the long- run growth rate. Essentially, an accumulable

  factor in production (capital) becomes permanently more important, and

  this leads to a multiplier eff ect that raises growth.

  Third, suppose now that the elasticity of substitution is greater than one.

  In this case, the argument given before reverses, and now the CES function

  asymptotically looks like the plentiful factor, in this case K . The model will t

  then deliver explosive growth under fairly general conditions, with incomes

  becoming infi nite in fi nite time.9 But this is true even without any automa-

  tion. Essentially, in this case researchers are not a necessary input and so

  standard capital accumulation is enough to generate explosive growth. This

  is one reason why the case of < 1—that is, an elasticity of substitution

  less than one—is the natural case to consider. We focus on this case for the

  remainder of this section.

  9. A closely related case is examined explicitly in the discussion surrounding equation (27) below.

  252 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones 9.3.1 Continuous Automation

  We can now consider the special case in which automation is such that the

  newly automated tasks constitute a constant fraction, q, of the tasks that

  have not yet been automated. Recall that this was the case that delivered a

  balanced growth path back in the Balanced Growth section

  In This Case, B → 1 and (C / C ) → g = – [(1 – )/ ]· > 0 Asymptotically t

  t

  t

  c

  The same logic that gave us equation (22) now implies that

  B K

  (23)

/>   A

  = A C S F

  t

  t ,1

  t

  t

  t

  t

  C S

  t

  t

  ~ A C S ,

  t

  t

  t

  where the second line holds as long as BK / CS → ∞, which holds for a large class of parameter values.10

  This reduces to the Jones (1995) kind of setup, except that now “eff ective”

  research grows faster than the population because of AI. Dividing both sides

  of the last expression by A gives

  t

  A

  S

  (24)

  t = Ct t .

  A

  A 1

  t

  t

  In order for the left- hand side to be constant, we require that the numerator

  and denominator on the right side grow at the same rate, which then implies

  + g

  (25)

  g = gC

  S .

  A

  1

  In Jones (1995), the expression was the same except g = 0. In that case, the

  C

  growth rate of the economy is proportional to the growth rate of research-

  ers (and ultimately, the population). Here, automation adds a second term

  and raises the growth rate: we can have exponential growth in research eff ort

  in the idea production function not only because of growth in the actual

  number of people, but also as a result of the automation of research implied

  by AI. Put another way, even with a constant number of researchers, the

  number of researchers per task S/ (1 – ) can grow exponentially: the fi xed

  t

  number of researchers is increasingly concentrated onto an exponentially

  declining number of tasks.11

  10. Since B → 1, we just require that g > g . This will hold—see below—for example if > 0.

  k

  c

  11. Substituting in for other solutions, the long- run growth rate of the economy is g =

  y

  {– [(1 – )/]· + n}/(1 – ), where n is the rate of population growth.

  Artifi cial Intelligence and Economic Growth 253

  9.4 Singularities

  To this point, we have considered the eff ects of gradual automation in

  the goods and idea production functions and shown how that can poten-

  tially raise the growth rate of the economy. However, many observers have

  suggested that AI opens the door to something more extreme—a “techno-

  logical singularity” where growth rates will explode. John Von Neumann is

  often cited as fi rst suggesting a coming singularity in technology (Danaylov

  2012). I. J. Good and Vernor Vinge have suggested the possibility of a self-

  improving AI that will quickly outpace human thought, leading to an “intel-

  ligence explosion” associated with infi nite intelligence in fi nite time (Good

  1965; Vinge 1993). Ray Kurzweil in The Singularity is Near also argues for

  a coming intelligence explosion through nonbiological intelligence (Kurz-

  weil 2005) and, based on these ideas, cofounded Singularity University with

  funding from prominent organizations like Google and Genentech.

  In this section, we consider singularity scenarios in light of the produc-

  tion functions for both goods and ideas. Whereas standard growth theory is

  concerned with matching the Kaldor facts, including constant growth rates,

  here we consider circumstances in which growth rates may increase rapidly

  over time. To do so, and to speak in an organized way to the various ideas

  that borrow the phrase “technological singularity,” we can characterize two

  types of growth regimes that depart from steady- state growth. In particular,

  we can imagine:

  • a “Type I” growth explosion, where growth rates increase without

  bound but remain fi nite at any point in time; and

  • a “Type II” growth explosion, where infi nite output is achieved in fi nite

  time.

  Both concepts appear in the singularity community. While it is common

  for writers to predict the singularity date (often just a few decades away),

  writers diff er on whether the proposed date records the transition to the

  new growth regime of Type I or an actual singularity occurring of Type II.12

  To proceed, we now consider examples of how the advent of AI could

  drive growth explosions. The basic fi nding is that complete automation of

  tasks by an AI can naturally lead to the growth explosion scenarios above.

  However, interestingly, one can even produce a singularity without relying

  on complete automation, and one can do it without relying on an intelligence

  explosion per se. Further below, we will consider several possible objections

  to these examples.

  12. Vinge (1993), for example, appears to be predicting a Type II explosion, a case that has been examined mathematically by Solomonoff (1985), Yudkowsky (2013), and others. Kurzweil (2005), by contrast, who argues that the singularity will come around the year 2045, appears to be expecting a Type I event.

  254 Philippe Aghion, Benjamin F. Jones, and Charles I. Jones 9.4.1 Examples of Technological Singularities

  We provide four examples. The fi rst two examples take our previous mod-

  els to the extreme and consider what happens if everything can be auto-

  mated—that is, if people can be replaced by AI in all tasks. The third ex-

  ample demonstrates a singularity through increased automation but without

  relying on complete automation. The fi nal example looks directly at “super-

  intelligence” as a route to a singularity.

  Example 1: Automation of Goods Production

  The Type I case can emerge with full automation in the production for

  goods. This is the well- known case of an AK model with ongoing techno-

  logical progress. In particular, take the model of section 9.2, but assume that

  all tasks are automated as of some date t . The production function is there-0

  after Y = A K and growth rates themselves grow exponentially with A .

  t

  t

  t

  t

  Ongoing productivity growth—for example, through the discovery of new

  ideas—would then produce ever- accelerating growth rates over time. Spe-

  cifi cally, with a standard capital accumulation specifi cation ( K = sY – K ) t

  t

  t

  and technological progress proceeding at rate g, the growth rate of output

  becomes

  (26)

  g = g + sA

  ,

  Y

  t

  which grows exponentially with A .

  t

  Example 2: Automation of Ideas Production

  An even stronger version of this acceleration occurs if the automation

  applies to the idea production function instead of (or in addition to) the

  goods production function. In fact, one can show that there is a mathe-

  matical singularity: a Type II event where incomes essentially become infi -

  nite in a fi nite amount of time.

  To see this, consider the model of section 9.3. Once all tasks can be auto-

  mated, that is, once AI replaces all people in the idea production function,

  the production of new ideas is given by

  (27)

  A = K A .

  t

  t

  t

  With > 0, this diff erential equation is �
�more than linear.” As we discuss next,

  growth rates will explode so fast that incomes become infi nite in fi nite time.

  The basic intuition for this result comes from noting that this model is

  essentially a two- dimensional version of the diff erential equation A = A 1+

  t

  t

  (e.g., replacing the K with an A in equation [27]). This diff erential equation can be solved using standard methods to give

  1/

  (28)

  A =

  1

  .

  t

  A

  t

  0

  Artifi cial Intelligence and Economic Growth 255

  And it is easy to see from this solution that A( t) exceeds any fi nite value before date t* = (1 / A). This is a singularity.

  0

  For the two dimensional system with capital in equation (27), the argu-

  ment is slightly more complicated but follows this same logic. The system of

  diff erential equations is equation (27) together with the capital accumulation

  equation ( K = sY – d K , where Y = A L). Writing these in growth rates gives t

  t

  t

  t

  t

  A

  (29)

  t = Kt A ,

  A

  A

  t

  t

  t

  K

  A

  (30)

  t = sL t

  .

  K

  K

  t

  t

  First, we show that ( A / A ) > ( K / K ). To see why, suppose they were equal.

  t

  t

  t

  t

  Then equation (30) implies that ( K / K ) is constant, but equation (29) would t

  t

  then imply that ( A / A ) is accelerating, which contradicts our original t

  t

  assumption that the growth rates were equal. So it must be that

  ( A / A ) > ( K / K ).13 Notice that from the capital accumulation equation, this t

  t

  t

  t

  means that the growth rate of capital is rising over time, and then the idea

  growth rate equation means that the growth rate of ideas is rising over time

  as well. Both growth rates are rising. The only question is whether they rise

  suffi

  ciently fast to deliver a singularity.

  To see why the answer is yes, set = 0 and sL = 1 to simplify the algebra.

  Now multiply the two growth rate equations together to get

  A

  K

  (31)

  t

  t = A .

  A

  K

  t

  t

  t

  We have shown that ( A / A ) >

‹ Prev