Hello everybody. Today we’re talking to Doyne Farmer.

Doyne is Professor of Mathematics at Oxford

University in England and he co-directs the program

on complexity economics at the Institute for New Economic Thinking, also at Oxford.

In addition, Doyne is external professor at the Santa Fe Institute.

Doyne is a prominent scientist in so many fields that I’ve lost track.

But perhaps most prominently, he was a pioneer

in the study of dynamical systems and chaos, and in time series

analysis and in applying ideas from complex systems

to economics and also to studying the evolution of technology. And I’m sure I left something out.

But anyway, welcome Doyne. Thank you.

So let me ask, you’ve been studying economics for a long time, even though your background

is in theoretical physics.

In this unit, we’re studying self-organization in various systems

in nature and in technology. Would you say that the economy is a

self-organizing system? Definitely.

I would actually say that the thinking about self-organizing systems

came first from economics, when Adam Smith

wrote The Wealth of Nations, he marveled at

the invisible hand, such that people

could each pursue their own selfish purposes, and yet

somehow, things that are good for everybody emerge.

He didn’t put it in exactly those words, but that’s the basic idea.

And that was 1776.

In a certain sense, he was the first person to really articulate

what a self-organized complex system was about.

Do you think that

economics has anything in common with other systems in nature that we

look at as self-organizing systems? I know you’ve studied

the immune system and a number of other systems in nature.

What are the commonalities? Well, the commonalities are

that you have entities

interacting through interactions with each other

and blah blah blah, so

at that easy level there’s commonalities, but I think you can

view what happens in an economy

as a type of learning, very much like people in

machine learning often use -- it’s a self-organized kind of learning

where each individual is processing his or her information

and taking a certain set of actions

and communicating signals to other people through

crises, through consumption, through production

supply and demand, and the forces of those things are

in turn, altering the variables that everybody sees

and feeding back and so

I think that there’s actually

quite a lot of commonality. In fact, in simple

in some simple models in economics you ended up

with quantities like somebody’s wealth looking

exactly like probability in a Bayesian

updating algorithm. Dmitriy Cherkashin

and I wrote a paper on a simple model for

a kind of very, very simplified economy where exactly that happens. You

start out with players. Each of the players is making a bet on whether --

what the state of the world is

going to be in the next round. What is going to be the outcome

of a horse race or a coin flip, to make it really really simple.

It turns out that the wealth updating dynamics under a classic

paramutual betting algorithm

is exactly like that, with Bayesian updating.

So the system converging to a final state is just

like what you’d expect from some model

of distributed learning under a Bayesian

algorithm. So we’ve looked at some

ant colonies and we’ve looked at some

immunology a little bit, and in all those kinds of systems

biological systems, there’s adaptation, where we have the individual

components, but the whole system in some sense adapts

or improves to foster the

evolution of the entity, so does that happen

in economics? Does the whole system adapt in some sense?

Yeah. In the example I just gave you, the whole system is adapting because

the agent

with the strategy that most closely matches reality

is the one that, in the little game that I mentioned with all

the money -- in a more realistic model of ecnomy,

certainly the economy is adapting through time

We see it decade by decade.

As the economy changes, the set of goods we produce changes

and the set of the things we consume changes, constantly adapting

the conditions that are there and

in a co-evolving self-organized way, just like

biological systems. I think it was in

1990 that I wrote a paper called A Rosetta Stone

for Connectionism, where I discussed

in that paper models for autocatalytic networks

immune system, classifier system, neural networks, I kind of

waved my hands at game theory at the time because frankly I didn’t

understand game theory as well as I should have back then.

But were I to update that paper, I would definitely add game theory

in, and it fits -- in fact it can be viewed as

the general framework in which all those things sit.

And game theory is at the bottom

of most economic models, where you have actors that are

going about their own purposes, trying to do a little better

either competitively or cooperatively. There’s a

whole branch of game theory that’s cooperative game theory that says okay, let’s assume we’re playing on

the team, how do we best cooperate?

Within the confines of either kind of game theory, I think you could

very much build on those analogies and

those analogies in turn relate back to things like

everything from the origin of life to the immune system

at some level ant colonies and so forth, if you made them sufficiently elaborate

Okay, so let me switch gears a little bit and ask

how did you as a physicist end up studying economics?

And also, how can physics, or ideas from physics, contribute to

economics?

Well, I can answer that question in two different ways. One is, like many things

the course of my career is driven largely by chance and circumstance

When I was a graduate student, my friend

___ said hey, I bet we could figure out how to predict roulette using the laws of physics

and that set me off on a whole course for my career

where I became somebody whose career

has been based on challenges of trying to predict things that people say I shouldn’t

be able to predict. When I was doing time series

analysis I would give talks on time series

results we had on ice ages and sunspots and fluid

And some clown would always say, have you tried applying this

to the stock market? I got kind of sick of hearing that

and so I quit my job at Los Alamos, and started a company to predict the stock market

and I did that for eight years until we showed we really could predict the stock market

although not really with the same techniques I was using at the time

for ice ages and sunspots, but that’s a long story

and then when I emerged from my eight years in the wilderness

of commercial activity, I decided to go back in

to complex systems, and I figured I should make use of the domain knowledge

I’d learned about financial markets and economics that was

stimulated by the fact that I was reading papers on economics

which I’d gotten into deeper as a result of doing that I realized

I didn’t really believe the paradigms they were using at all, so using

Steven Weinberg’s statement that one should swim for

open water, that is, go to places where the problems aren’t solved

I thought, well, I need to do -- I want to do

complex systems, but in a domain where I can take advantage

of domain knowledge I used during my time at the prediction

company and so I went into economics.

Not quite realizing what fierce

social counterpressure

I would get from economists as a result of trying to suggest

that they should use different methodologies and ask different questions

and go about asking those questions in a different way than what they were doing already.

So what do you think physics can contribute to economics?

Well, I think it contributes mainly a different point

of view, a different epistemology. A different notion of

what an interesting question is, how

one should go about solving the question, and what a good answer to the question looks like

when you’e done. And I’ve been struck

in my now more than a decade of

interacting with economists

sometimes very cordially and sometimes a bit confrontationally

I’ve been just struck at what a difference

anthropology of the field makes.

How much I realized I got brainwashed to think

in a certain way when I was in graduate school and those guys

you know, that went to economic graduate school got

brainwashed in a different way, or maybe it’s also partly

a self-selection process, that

they selected themselves to do economic graduate school, I selected myself

to do physics. But I do think that

I do have several collaborations with economists, they’re very

productive -- I think thinking about the world differently is a good thing

and allows you to contribute providing you get other people to

listen to what you say when you get done.

What is this thing called complexity economics?

So complexity economics is using

the mindset, methods, and

point of view that we have in complex systems and applying it to economics.

And while it’s true that Adam Smith started the whole thing,

the intellectual evolution of the field

has substantially diverged from

that that is most represented in complex systems, which I think

is some kind of mixture of physics

computer science, biology, and cognitive

science, and other stuff. But economics is

not well represented.

In complex systems, we have a view

of the world that is you need to think about

coming up with rules for what the low-level interactions are

You need to think about how the entities interact with each other. You need to

capture the crucial elements of

the structural relationships between the entities, so

if you’re talking about an ecosystem

the nature of the environment, the nature of

the way the organisms live in that environment really matters

and you have to put enough realism in there to make it relevant to the real world.

And you shouldn’t be afraid of simulation.

Okay, that was my next question. Yeah, economists are

have not, in my opinion in general,

even come close to realizing the power of simulation

and because of the emphasis on

mathematical proof and

they’ve made in general excessively

stylized models that fail to capture key elements

of the real world. So that's

the sort of difference in key elements

at least to the difference in intellectual viewpoint that I think I have

with most of them. Not all of them, of course. There are many economists who are

a minority of economists who think differently. The 95%

that follow the mainstream.

So you’ve argued strongly for the use of agent-based modeling

in economics. Can you say a little bit about why you think that’s a good approach?

I think just for the reasons I gave

It’s at the end of the day imperative that

the economic agents be allowed

to exist in a world that contains the key institutional features

that are so important in the real world.

So if we want to understand the crisis

of 2007-8 and on, that we’re still kind of floundering

in in some way, we really have

to have institutions like banks

that can fail, can default

we have to have housing markets, we have to have those

things in there. The dynamic stochastic equilibrium

models that were prevalent in 2007-8 did not have those

features. There are -- it’s a

very serious effort in economics now to put them in, but

and while I think

those models may play a useful role, I think to really make those models

realistic in key ways, we really have to go with agent-based modeling.

So do you see agent-based modeling as

kind of a placeholder for more advanced mathematics

we’ll eventually be able to do without having to use simulation, or is it something that

is just essential, will always be essential?

I think it will always be essential, but

let me try and clarify my answer. First of all,

existing economics uses quite fancy mathematics. So it’s not about

how fancy the mathematics is. It’s about whether

the mathematics that we’re able to do is really relevant for the key features

of the world that we’re trying to capture. Okay, so replace

what I said with more relevant mathematics.

I think there’s a cycle that one goes through, and you should go through, where

you alternate between making

agent-based models that may have some level of realism

in them, and then stepping back and then asking what are the key features

of these models, can we make simpler models that capture

the essential features that give us a clearer

understanding of what happens. I think one should go back and forth

between those viewpoints. You may, in the course of formulating

your simplified model, realize something that you’re missing in your agent-based model

and put it in there, or put it in there in a different way. You may

extract the essence of what’s going on and change the way you think about what’s going on in your agent-based model.

It’s a little bit like if you look at

the study of fluid turbulence today. Fluid

turbulence is a field where there’s a lot of alternation between

experiements, simulations, and mathematical

simplified mathematical models. And all three of those

fields learn from each other. We, by the way, are

also trying to incorporate experiments as part of what we do

in the software that we’ve written in the

crisis project, we have the ability to run economic

experiments and unplug any of the algorithmic agents

in the model, replace that algorithmic agent with a person or

a set of people, and see what decisions they make, and

in the context build all the other algorithmic agents in our model

setting. So we have a group in Amsterdam that’s actually running experiments

along those lines, led by ____.

So we’re trying to do everything

we can try to calibrate the models in the most realistic

way possible.

One critique that has come up of very detailed agent-based models

is that you can keep throwing stuff in, more and more, until

you match the data, the empirical data

and that you get kind of hypercycle

not hypercycle, what do you call the

Right. Simulation that represents reality

and detail, but doesn’t really give you any clear abstractions

about what’s really going on. Well,

let me just say, if we could do that for the economy, I would say hallelujah

glory, brother, I would be tickled.

Because we’re far from being able to do that, and if we could do that

then we could run things at double speed and

run policy studies and make conditional

forecasts that we can’t do right now, and it would be extremely useful.

Secondly, if we were able to achieve

…. in 2001, and asking what are the key features

of our model that really give us the essence of the results

and in the course of uplugging those components we would learn a

hell of a lot about the real economy. And we could

replace things with stylized, dumbed-down versions, we could do all

kinds of experiments. So if we could ever achieve that

start point, then we would be most of the way there

in my view. Okay. Let me ask one last question, which is what are you

working on that you are really excited about these days?

I think the thing that, if you’re talking about

this month or these days,

Probably the thing I’m most excited about

is a project that I’m working on with James McNerney and Francesco Caravelli

where we’re trying to think about the

economy as a -- to

let’s say carry over a lot of ideas from ecology

and think about the structure of the economy

and try to understand something that we've

seen anecdotally, which is that some things

are -- don’t seem to be getting systematically cheaper

over time, like natural resources. Coal, oil, natural gas

iron, copper,

lots and lots of things that we’ve looked at that

are natural resources cost about the same now that they did

in 1900. In contrast

computers as we know are way, way cheaper,

by a factor of a billion. Than they were in 1900?

Than they were in 1900.

Not to mention performance.

Photovoltaic cells have dropped by a factor of

100 in price since they were introduced in the 50s

in a very steady and systematic way

Agriculture, food has even

gotten dramatically cheaper.

But some of these things have stubbornly, like oil

have stubbornly stayed at high prices. We have a theory for that

that we think it’s driven by the fact that the economy

grows by adding trophic levels

and we should really think of them like, much like food chains

in ecology and in some areas of economy these

trophic levels are quite deep, and so we

have an idea, we have a theory -- maybe I shouldn’t give away the details --

but we think that using those ideas we may be able to

explain why some things like computers improve a lot

faster than other things like oil prices

and we think that this is important in thinking about

how we ought to deal with global warming, because

it suggests that some energy technologies like photovoltaics

are on a trajectory where their prices are getting cheaper

and cheaper, whereas others like coal or

nuclear power

for reasons that we don’t understand,

coal we understand. Nuclear power we don’t. Don’t seem to be on such a trajectory.

So we think that if we’re placing bets on the future, we should place heavier bets

than people are placing on solar power.

Alright well, great. Thank you. This has been a really wonderful

conversation. Good. Thank you, Melanie. I had fun, too.