Friday, November 28, 2014

A Brain’s View of Economics

CAMBRIDGE – In his pathbreaking 2005 book On Intelligence, Jeff Hawkins proposed an alternative paradigm of how the human brain works. In his view, the brain is not a Turing machine that manipulates symbols according to a table of rules, which is the model on which computers and artificial intelligence have been based. Instead, the brain is a giant hierarchical memory that is constantly recording what it perceives and predicting what will come next.

The brain makes predictions by finding similarities between patterns in recent sensory inputs and previous experiences stored in its vast memory. It matches current fragmentary sounds in a sea of noise with a known song, or the face of a person in disguise with that of your child. The idea is similar to the auto-complete function in, say, the Google search box – constantly guessing what you will enter next based on what you have already typed in.

To see the hierarchy in this mechanism, consider that by perceiving just a few letters, you can predict the word; by looking at a few words, you can predict what the sentence means, or even the paragraph. In fact, right now you must be guessing where it is that I am going with this entire commentary. The hierarchy allows you to understand meaning, whether the input got to your brain by reading or listening. The brain is thus an inductive machine that predicts the future based on finding similarities, at many different levels, between the present and the past.

Hawkins’ alternative model of how the brain works has important implications for many fields, including the one that I spend most of my time thinking about: economic-development strategy.

By definition, development is not just more of the same, just as an adult is not just a big baby. The process involves adding and combining new and existing capabilities to support more diverse and complex activities.

But finding new things that can be done successfully is tricky, because it requires knowing what you will need and whether you will be able to procure it. This is why Jeffrey Sachs’s Millennium Villages project has faltered, as the journalist Nina Munk’s recent book shows. In trying to move farmers from subsistence to commercial agriculture, Munk argues, there are just too many missing pieces.

Traditional thinking in economic development has followed a Turing-like approach, trying to specify a general model of the world – based on first principles – and then use that model to think about a country’s predicament or a policy’s potential impact. But the world is often too complex and nuanced for such an approach.

Would it not be a great improvement if, when looking at a particular place, we could have in mind all of the world’s previous experiences and automatically identify the most relevant ones, in order to infer what to do next? Would it not be useful to see the development possibilities just as our brain, according to Hawkins, sees the world?

An alternative, Hawkins-like approach to economic development would take massive amounts of data about the world and ask what is likely to succeed next in a country or a city at a given point in time, given what is already present and in light of the experience there and everywhere else. It would be like Amazon’s recommendation system, proposing books you may like based on your and everybody else’s experience.

In a recent paper, my colleagues and I showed that such an approach to economic development actually works. In a particular city or country, you can predict, even a decade in advance, which industries will appear or disappear or grow or wane just by knowing the history of what has been there and everywhere else.

Countries tend to move into industries that are related to the ones they already have or that are present in locations that are similar to them. We have made the approach user-friendly for countries in our recent Atlas of Economic Complexity.

The idea of looking at previous experiences to inform future action is as old as civilization. Following this intuition, Justin Yifu Lin, the former chief economist of the World Bank, has suggested that when countries choose what to do next, they should look at a successful country that was similar to them two decades ago.

But we should be able to do much better than that by looking at many more experiences in much more detail, using a much bigger memory that can find many more patterns across much more of the relevant human experience. Imagine that Sachs’s Millennium Villages project had known the sequence of all previous successful moves out of subsistence agriculture, rather than relying only on guesswork or deduction. Would it not be useful to understand the paths to industrial development – and the dead ends – that are most relevant to a particular country today?

This alternative approach can empower many more people to seek successful routes to prosperity by lowering the perils and risks involved in the search – in the same way that maps empower people to get to where they want to go with much more information than they would otherwise have. Just as augmented-reality technologies make our experience of the world richer (imagine a sports match today without instant replay), putting the development experience of the world at the fingertips of those engaged in promoting development is now perfectly feasible. We should seize this opportunity.

  • Contact us to secure rights


  • Hide Comments Hide Comments Read Comments (15)

    Please login or register to post a comment

    1. CommentedDavid Donovan

      Instead, the brain is a giant hierarchical memory that is constantly recording what it perceives and predicting what will come next.

    2. CommentedRavindra Muley

      This is very insightful article, thought provoking, predicting economic development and economic future of various countries coupled with data of numerous variable will help to find the competitive advantage of the region and thus would give guidance in which ways the economy must tread, though human brain makes predictions and relies on past data, we must not ignore the fallacies and traps of thinking too.

    3. Commentedfrancesco totino

      i wrote something similar almost ten years ago... unfortunately only in Italian Language .. i ve to traslate still .. Culture , Education, Ethic undervalued factors of competitive development …

    4. CommentedMark Greidanus

      The dream you have is exactly what historical scholarship does in an ideal situation. It is limited by many flaws, but once in a long while a mind comes along that succeeds in predicting development. For example, James Madison predicted the American Civil War at the Constitutional Convention.

    5. Portrait of Ricardo Hausmann

      CommentedRicardo Hausmann

      Hi Eduardo: You make two points. First, we have much less data than the brain. Second, we will keep on making mistakes. To the first point I would say that we already have much more information than we actually use. We can do much more with the Big Data revolution, using government administrative data on taxes, trade, social security and the like, as well as private data from cell phones, credit cards, financial statements, etc. On the second, we will always err because induction is what it is, given that we live in an open non-ergodic changing world. But as David Hume said, induction is our only basis of knowledge about the world.

    6. Portrait of Ricardo Hausmann

      CommentedRicardo Hausmann

      Hi Joshua: Thanks for your comment. Pattern recognition leads to hierarchies of meaning. The word dog is a useful generalization of many instances of certain kinds of animals. Newton's Law of Universal Gravitation is the recognition of a higher level pattern on a set of motions (the apple and the moon). The overarching framework that you mention is a higher order pattern on patterns in the data. At least, that is what Jeff Hawkins thinks it is. It is pattern recognition turned on itself. The paper "Implied Comparative Advantage" that I link to in the piece tries to show that traditional trade theories (both Ricardian and Heckscher-Ohlin) can provide interpretation and meaning to why the approach I propose actually works in the data, but highlight some possible channels that need to be clarified through future research. (e.g. inter-industry knowledge spillovers).

    7. CommentedJoshua Soffer

      It is indeed an improvement over Turing-based algorithms to recognize the creative role of rich historical data in formulating predictive strategies for the future. But we must also appreciate that human understanding is not simply about the accumulation of data to be pattern-matched against incoming information. The meaning of a set of data is in the way it is interpreted via an overarching framework of understanding. Meaning is a gestalt, a schema, and significant development of meaning-making has to do with the ability to transform the organization of our frameworks of understanding. Otherwise we repeat the same interpretation of experience over and over again , regardless of how rich our database becomes. Amazon and Google will never produce what's not already in its database, but human progress depends on foreseeing what has never been before, and this only happens as a result of reconceiving the implications of historical data.

    8. Portrait of Eduardo Moron

      CommentedEduardo Moron

      Ricardo, the main problem with your suggested approach for economic development is the lack of massive data compared to what our brain normally processes prior to send us a new guess. It seems that we will keep failing in our task to move our countries to development.

    9. Commentedfrancesco totino

      completely agree on your last article A Brain’s View of Economics that i read on Projects Syndicate !!! please take a look at what i wrote about 10 years ago on this subject ...sorry if it need a translator... the title is Culture , Education, Ethic undervalued factors of competitive development .. you can find of my blog

    10. CommentedProcyon Mukherjee

      Why projects on poverty reduction falter and whether inductive reasoning could have helped those projects, Jeffrey Sachs’ project, for example, could have taken a cue from the work done by Verghese Kurien in India, which formed cooperatives among farmers that bound them together in a journey that made mobility from destitution possible for millions. Some of the key takeaways are that poor farmers when pulled into a cooperative become a formidable force to reckon with. Dolling out subsidies and free aid actually enfeebles them into inaction that go against them as they do not understand the true economic value. True value can only be understood when they come on terms with losses and only when the cooperative spirit is combined to offset losses by dealing with the inherent challenges (inductive reasoning makes tremendous sense), the farmers’ initiatives to combat poverty becomes a sustainable proposition; no other backstop works but the combined trust of 'cooperative farming' for marginal farmers, but it needs the passion, grit and the guts of a relentless leader like Kurien.

    11. CommentedJose araujo

      Amazes me that a professor @Harvard doesn't know the scientific method and is proposing some aberrant datamining tool for economic knowledge, and acting like economics is all about gess work and hunches...

        CommentedJose araujo


        Since the dawns of times inEconomy, that's how you do science. Actually that how you do science in all human sciences, since its hard to replicateand make controlled experiences.

        Most of economic data that validates or wrongs an hypothesis are empricaldata you collect.

        Actually this article explains Haussman's points of views very well. For him economy is based on hunches and prejudices..

        Just take his attack on a initiative to reduce hunger in the world, and his implicit praise on the work of Nina Munk's who's conclusion is that we should let the hunger problem be solved by private, markets, initiative....

        CommentedRobert Snashall

        Awesome article once again Hausmann. I have been thinking for a while that it would be interesting to combine AI weather modelling whereby the program looks at current data, makes thousands of predictions based on varying theories, waits for the next batch of data, decides which theory worked best, learns, and updates itself for the next set of predictions.

    12. CommentedKen Presting

      Prof. Hausmann describes valuable research, but we should all keep in mind that economics is among the "Sciences of the Artificial," as Herbert Simon once called them. By far the greatest volume of transactions in any economy are not between individual human beings with live brains, but instead are between corporations which make decisions by applying mathematical formulas. These formulas range from simply balancing their books with simple arithmetic, to stock trading strategies running on supercomputers.

      One of the great satisfactions (and great utilities) in studying economics is to realize that mathematical relations such as arbitrage or Dutch books can identify bad choices *before* we make the error of trying them.

      Let's not lose sight of the fundamental role of abstraction in economics. Money itself is an abstraction. Our brains may not be Turing machines, but both a corporate balance sheet and a household budget are exactly what Turing machines handle best - a big list of numbers.