The Economics of Artificial Intelligence

Home > Other > The Economics of Artificial Intelligence > Page 28
The Economics of Artificial Intelligence Page 28

by Ajay Agrawal


  (Romer 1993, 71). In our model, A is a measure of a researcher’s human capital. Clearly, human capital depends on the existing technological and other knowledge and the researcher’s access to that knowledge. In turn, the production of new knowledge depends on the researcher’s human capital.

  156 Ajay Agrawal, John McHale, and Alexander Oettl

  Innovations occur as a result of combining existing knowledge to produce

  new knowledge. Knowledge can be combined a ideas at a time, where a =

  0, 1 . . . A. For a given individual researcher, the total number of possible combinations of units of existing knowledge (including singletons and the

  null set)3 given their knowledge access is

  A

  A

  (1)

  Z =

  = 2 A .

  i

  a

  a=0

  The total number of potential combinations, Z , grows exponentially with

  i

  A. Clearly, if A is itself growing exponentially, Z will be growing at a double i

  exponential rate. This is the source of combinatorial explosion in the model.

  Since it is more convenient to work with continuously measured variables in

  the growth model, from this point on we treat A and Z as continuously mea-i

  sured variables. However, the key assumption is that the number of potential

  combinations grows exponentially with knowledge access.

  The next step is to specify how potential combinations map to discover-

  ies. We assume that a large share of potential combinations do not produce

  useful new knowledge. Moreover, of those combinations that are useful,

  many will have already been discovered and thus are already part of A. This

  latter feature refl ects the fi shing- out phenomenon. The per- period transla-

  tion of potential combinations into valuable new knowledge is given by the

  (asymptotically) constant elasticity discovery function

  (2)

  1

  A =

  Zi

  =

  2 A

  ( ) 1

  for < ≤ 1

  i

  = ln Z = ln 2 A

  ( ) = ln(2) A for = 0,

  i

  where is a positively valued knowledge discovery parameter and use is

  made of L’Hôpital’s rule for the limiting case of = 0.4

  For > 0, the elasticity of new discoveries with respect to the number of

  possible combinations, Z , is

  i

  A Z

  1

  (3)

  i =

  Zi

  =

  Zi

  ,

  Z A

  Z

  1

  i

  ( Z

  1) /

  i

  i

  3. Excluding the singletons and the null set, total number of potential combinations would be 2A – A – 1. As singletons and the null set are not true “combinations,” we take equation (1) to be an approximation of the true number of potential combinations. The relative signifi cance of this approximation will decline as the knowledge base grows, and we ignore it in what follows.

  4. L’Hôpital’s rule is often useful where a limit of a quotient is indeterminate. The limit of the term in brackets on the right- hand side of equation (2) as goes to zero is 0 divided by 0 and is thus indeterminate. However, by L’Hôpital’s rule, the limit of this quotient is equal to the limit of the quotient produced by dividing the limit of the derivative of the numerator with respect to by the limit of the derivative of the denominator with respect to . This limit is equal to ln(2)A.

  Artifi cial Intelligence and Recombinant Growth 157

  which converges to as the number of potential combinations goes to infi n-

  ity. For = 0, the elasticity of new discoveries is

  A Z

  Z

  (4)

  i =

  i

  = 1 ,

  Z A

  Z

  lnZ

  lnZ

  i

  i

  i

  i

  which converges to zero as the number of potential combinations goes to

  infi nity.

  A number of factors seem likely to aff ect the value of the fi shing- out/

  complexity parameter, . First are basic constraints relating to natural

  phenomena that limit what is physically possible in terms of combining

  existing knowledge to produce scientifi cally or technologically useful new

  knowledge. Pessimistic views on the possibilities for future growth tend to

  emphasize such constraints. Second is the ease of discovering new useful

  combinations that are physically possible. The potentially massive size and

  complexity of the space of potential combinations means that fi nding useful

  combinations can be a needle- in-the- haystack problem. Optimistic views of

  the possibilities for future growth tend to emphasize how the combination

  of AI (embedded in algorithms such as those developed by Atomwise and

  DeepGenomics) and increases in computing power can aid prediction in the

  discovery process, especially where it is diffi

  cult to identify patterns of cause

  and eff ect in high- dimensional data. Third, recognizing that future oppor-

  tunities for discoveries are path dependent (see, e.g., Weitzman 1998), the

  value of will depend on the actual path that is followed. To the extent that

  AI can help identify productive paths, it will limit the chances of economies

  going down technological dead ends.

  There are L researchers in the economy each working independently,

  A

  where L is assumed to be measured continuously. (In section 5.4, we con-

  A

  sider the case of team production in an extension of the model.) We assume

  that some researchers will duplicate each other’s discoveries—the standing-

  on- toes eff ect. To capture this eff ect, new discoveries are assumed to take

  place “as if ” the actual number of researchers is equal to L , where 0 ≤ ≤ 1.

  A

  Thus the aggregate knowledge production function for > 0 is given:

  2 A

  ( ) 1

  (5)

  A = L

  .

  A

  At a point in time (with given values of A and L ), how does an increase A

  in aff ect the rate of discovery of new knowledge, A? The partial derivative

  of A with respect to is

  A

  L ( ln(2) A

  1)2 A

  A

  (6)

  =

  + LA .

  2

  2

  A suffi

  cient condition for this partial derivative to be positive is that that

  term in square brackets is greater than zero, which requires

  158 Ajay Agrawal, John McHale, and Alexander Oettl

  Fig. 5.2 Relationships between new knowledge production, ␪, and ␾

  1/

  (7)

  A >

  1

  .

  ln(2)

  We assume this condition holds. Figure 5.2 shows an example of how A

  (and also the percentage growth of A given that A is assumed to be equal to 100) varies with for diff erent assumed values of . Higher values of are

  associated with a faster growth rate. The fi gure also shows how and

  interact positively: greater knowledge access (as refl ected in a higher value

  of ) increases the gain associated with a given increase in the value of .

  We assume, however, that itself evolves with A. A larger A means a bigger and more complex di
scovery search space. We further assume that this

  complexity will eventually overwhelm any discovery technology given the

  power of the combinatorial explosion as A grows. This is captured by assum-

  ing that is a declining function of A; that is, = ( A), where ʹ( A) < 0. In the limit as A goes to infi nity, we assume that ( A) goes to zero, or

  (8)

  lim ( A) = 0.

  A

  This means that the discovery function converges asymptotically (given sus-

  tained growth in A) to

  (9)

  A = ln(2) L A .

  A

  This mirrors the functional form of the Romer/ Jones function and allows

  for decreasing returns to scale in the number of researchers, depending

  on the size of . While the form of the function is familiar by design, its

  combinatorial- based foundations have the advantage of providing richer

  motivations for the key parameters in the knowledge discovery function.

  Artifi cial Intelligence and Recombinant Growth 159

  We use the fact that the functional form of equation (9) is the same as that

  used in Jones (1995) to solve for the steady state of the model. More pre-

  cisely, given that the limiting behaviour of our knowledge production func-

  tion mirrors the function used by Jones and all other aspects of the economy

  are assumed to be identical, the steady state along a balanced growth path

  with constant exponential growth will be the same as in that model.

  As we have nothing to add to the other elements of the model, we here

  simply sketch the growth model developed by Jones (1995), referring the

  reader to the original for details. The economy is composed of a fi nal goods

  sector and a research sector. The fi nal goods sector uses labor, L , and inter-Y

  mediate inputs to produce its output. Each new idea (or “blueprint”) sup-

  ports the design of an intermediate input, with each input being supplied by

  a profi t- maximizing monopolist. Given the blueprint, capital, K, is trans-

  formed unit for unit in producing the input. The total labor force, L, is fully allocated between the fi nal goods and research sectors, so that L + L = L.

  Y

  A

  We assume the labor force to be equal to the population and growing at

  rate n(>0).

  Building on Romer (1990), Jones (1995) shows that the production func-

  tion for fi nal goods can be written as

  (10)

  Y = AL

  ( ) K 1 ,

  Y

  where Y is fi nal goods output. The intertemporal utility function of a rep-

  resentative consumer in the economy is given by

  (11)

  U = u( c) e tdt,

  0

  where c is per capita consumption and is the consumer’s discount rate.

  The instantaneous utility function is assumed to exhibit constant relative

  risk aversion, with a coeffi

  cient of risk aversion equal to and a (constant)

  intertemporal elasticity of substitution equal to 1 / .

  Jones (1995) shows that the steady- state growth rate of this economy

  along a balanced growth path with constant exponential growth is given by

  (12)

  g = g = g = g =

  n ,

  A

  y

  c

  k

  1

  where g = A/ A is the growth rate of the knowledge stock, g is the growth A

  y

  rate of per capita output y , (where y = Y / L), g is the growth rate of per c

  capita output c (where c = C / L) , and g is the growth rate of the capital labor k

  ratio (where k = K / L).

  Finally, the steady- state share of labor allocated to the research sector

  is given by

  (13)

  s =

  1

  1 + 1 /

  ( (1– ) / n) + (1/ ) –

  {

  }.

  160 Ajay Agrawal, John McHale, and Alexander Oettl

  We can now consider how changes in the parameters of knowledge pro-

  duction given by equation (5) will aff ect the dynamics of growth in the

  economy. We start with improvement in the availability of AI- based search

  technologies that improve a researcher’s access to knowledge. In the context

  of the model, the availability of AI- based search technologies—for example,

  Google, Meta, BenchSci, and so forth—should increase the value of and

  reduce the “burden of knowledge” eff ect. From equation (12), an increase

  in this parameter will increase the steady- state growth rate and also the

  growth rate and the level of per capital output along the transition path to

  the steady state.

  We next consider AI- based technologies that increase the value of the

  discovery parameter, . As does not appear in the steady state in equation

  (12), the steady- state growth rate is unaff ected. However, such an increase

  will raise the growth rate (and level) along the path to that steady state.

  The most interesting potential changes to the possibilities for growth

  come about if we allow a change to the fi shing- out/ complexity parameter,

  . We assume that the economy is initially in a steady state and then experi-

  ences an increase in as the result of the discovery of a new AI technology.

  Recall that we assume that will eventually converge back to zero as the

  complexity that comes with combinatorial explosion eventually overwhelms

  the new AI. Thus, the steady state of the economy is unaff ected. However,

  the transition dynamics are again quite diff erent, with larger increases in

  knowledge for an given starting of the knowledge stock along the path back

  to the steady state.

  Using Jones (1995) as the limiting case of the model is appealing because

  we avoid unbounded increases in the growth rate, which would lead to the

  breakdown of any reasonable growth model and indeed a breakdown in the

  normal operations of any actual economy. It is interesting to note, however,

  what happens to growth in the economy if instead of assuming that con-

  verges asymptotically to zero, it stays at some positive value (even if very

  small). Dividing both sides of equation (5) by A gives an expression for the

  growth rate of the stock of knowledge

  A

  (2 A )

  1

  (14)

  = ln(2) LA

  .

  A

  A

  The partial derivative of this growth rate with respect to A is

  ( A / A)

  (15)

  = LA 1+ 2 A

  ( ) ( ln(2) A 1) .

  A

  A 2

  The key to the sign of this derivative is the sign of the term inside the last

  round brackets. This term will be positive for a large enough A. As A is growing over time (for any positive number of researchers and existing knowledge

  stock), the growth rate must eventually begin to rise once A exceeds some

  threshold value. Thus, with a fi xed positive value of (or with converging

  Artifi cial Intelligence and Recombinant Growth 161

  asymptotically to a positive value), the growth rate will eventually begin to

  grow without bound.

  A possible deeper foundation for our combinatorial- based knowledge

  production function is provided by the work on “rugged landscapes” (Kauff -

  man 1993). Kauff man’s NK model has been fruitfully applied to question
s of

  organizational design (Levinthal 1997), strategy (Rivkin 2000) and science-

  driven technological search (Fleming and Sorenson 2004). In our setting,

  each potential combination of existing ideas accessible to a researcher is

  a point in the landscape represented by a binary string indicating whether

  each idea in the set of accessible knowledge is in the combination (a 1 in the

  string) or not (a 0 in the string). The complexity—or “ruggedness”—of the

  landscape depends on the total number of ideas that can be combined and

  also on the way that the elements of the binary string interact. For any given

  element, its impact on the value of the combination will depend on the value

  of X other elements.5 The larger the value of X the more interrelated are the

  various elements of the string, creating a more rugged knowledge landscape

  and thus a harder the search problem for the innovator.

  We can think of would-be innovators as starting from some already known

  valuable combination and searching for other valuable combinations in the

  vicinity of that combination (see, e.g., Nelson and Winter 1982). Purely

  local search can be thought of as varying one component of the binary

  string at a time for some given fraction of the total elements of the string.

  This implies that the total number of combinations that can be searched is

  a linear function of the innovator’s knowledge. This is consistent with the

  Romer/ Jones knowledge production function where the discovery of new

  knowledge is a linear function of knowledge access, A f. Positive values of

  are then associated with the capacity to search a larger fraction of the space

  of possible combinations, which in turn increases the probability of discov-

  ering a valuable combination. Meta technologies such as deep learning can

  be thought of as expanding the capacity to search a given space of potential

  combinations—that is, as increasing the value of —thereby increasing the

  chance of new discoveries. Given its ability to deal with complex nonlinear

  spaces, deep learning may be especially valuable for search over highly rug-

  ged landscapes.

 

‹ Prev