Friday, May 29, 2015

The Economy as One Big Brain

In my second year of graduate school at Northwestern, 1977-78, I took the year-long graduate math analysis sequence.  The professor was Robert Welland, who had an interesting persona, with a certain flair and personal idiosyncrasy.  He kept his hair long and sometimes in the middle of lecture would pull out a comb to straighten his hair.  I've never experienced another instructor doing anything like that. In class, and in conversation too, he would often preface what he had to say with the admonition, "Christ, man!"  This was his alert that what you said wasn't quite on the mark.

I may have been the only non-math graduate student in the class, though of this I'm not sure, but Welland liked me for studying economics and, at least during the first quarter, that I seemed to have more on the ball than the other students.  He told me he read my face while in class to measure how his lecture was going.  If I seemed to show comprehension then things were going fine.  If I looked confused then he assumed the class as a whole was in trouble.

I may have had certain advantages over the other students to achieve this position.  Some or all of them may have been in the first year, and graduate school is a slug then.  Also, I had that topology class at Cornell which I've written about before, most recently here, so I had some confidence going in that I wasn't over my head with the math.  There is also that if you study pure math sometimes you lack for a sense of why the concepts matter.  The economics helped me there, in particular when we studied Hilbert Spaces, since the inner product of two vectors is used in economics where one vector is a price system and the other vector is a commodity bundle and then the inner product yields the value of the bundle. 

By the third quarter for sure and quite possibly earlier, some of the other students had overtaken me as class performers, but also by then Welland had formed an impression of me that would sustain.  The consequence is that a few times I had discussions with him outside of class and on topics mainly unrelated to coursework.  In one of those talks we are walking on the landfill east of Norris Center next to Lake Michigan.  He gives me his view of economics, which at the time was quite futuristic and in retrospect still seems remarkable to me.  He said that economics would die as a discipline as computing power took off and all transactions could be recorded and stored in some big database.  There no longer would be a need for economic models.  The data could tell the full story.

I don't know if Welland was aware of Moore's Law then or not, but even Gordon Moore wasn't predicting this law would hold over many decades at the time he made his famous prediction.  And computing was remarkably primitive when I took Welland's class.  This was before the personal computer.  If you wanted to run a program you had to write it up on punch cards and submit it for a batch job on the mainframe at the Vogelback computing center.  So, based on what was actually possible at the time, Welland's prediction seems kind of fantastic.  Nonetheless, it appears he had a substantial interest in economics, based on the book titles he authored, so he clearly came to his ideas with much forethought.  (I have subsequently learned that the linked book is not by Welland, as Amazon says, but by a different guy named Weiland.)

Though Welland didn't say this to me, his thinking seems to imply that socialism would eventually take over.  At the time I was in graduate school, decentralization was a popular idea among economic theorists.  Here decentralization means many atomistic decision makers making choices independently, coordinated by the invisible hand or some other mechanism, rather than decision making by one all powerful center.   There were other reasons for valuing decentralization apart from limiting computing ability.  Another biggie was limited speed in information transmission.  If the lag was too great, between the time when information emerged at the edge of an organization till the time that the information was received and well understood by the center, it would be more efficient for the decision to be made at the edge, even if the decision maker couldn't factor in other information that might be relevant to the decision.  A third issue was incentives.  People might have reason to misreport the true information, so as to increase private gain.  The last reason I'll mention here is complexity.  Centralized decision making works best when the nature of the information to be collected is already understood.  So, for example, a Web form could be used to elicit the requisite information.  Decentralization is apt to perform better when there is so much uncertainty that it is unclear even how to describe the current situation, in which case making some sense of it requires a good deal of creativity.

* * * * *

There still is economics as a discipline.  Welland was either wrong about the possibility that data can tell the full story or he under estimated how much computing power it would take to achieve that end.   But the recent rise in concern about smart machines coupled with emergence of analytics as a field makes it seem at least possible that we are marching toward Welland's vision.  I really don't know. Instead of further speculation, now I want to turn from how things actually are and likely to be in the near future to pure science fiction - how things might be if computing did advance in the way Welland envisioned. 

In this utopian vision, smart machines are the salvation to our economic woes and, to turn things on their head, enable the economy to fully employ all people who are willing to work, pay them a decent wage, and restore a middle class lifestyle to the bulk of the population.  How do we get there from here?  Let's envision a bunch of different open source software development projects into artificial intelligence that are aimed at fundamentally changing capitalism as we know it.  The first one, which I will describe in some detail for illustration, is called The Virtual CEO.

As our story opens, the year is 2050.  Moore's Law is miraculously still alive and well.  Artificial intelligence has advanced to the point that it is quite capable of performing the executive decision making function. Actually, it can perform it better.  The Virtual CEO never has a feeling in the gut to drive decisions.  All choices are made based on available data.  The Virtual CEO remains focused on long term objectives for the organization.  It cares not for scoring short term profit at the expense of long term positioning.  It is never venal and always fair with customers, employees, shareholders, and the larger community.

The Virtual CEO functions best in flat organizations that aim for democratic decision making with employee input an important factor.  Indeed, a big part of the impetus behind the open source software development project that has produced The Virtual CEO is to convert existing hierarchical organizations to this structure and to create new organizations with this structure that can out-compete the older hierarchical organizations in the marketplace.

Such organizations will feature a flat compensation scheme for employees.  Since the Virtual CEO, which really functions for all upper level management in the organizations, demands no compensation whatsoever, there is more revenue available to pay existing employees a decent wage with solid benefits and to actually take a labor intensive approach to the work the organization produces.  This is part of the underlying objective of the organization.  Another part is to do well by customers, offering them a good service at a fair price.  One of the sidebar consequences of the Virtual CEO project, an important one to be sure, is the discovery that people would much rather interact with other people than with machines, once they are convinced that the people are there to help them rather than to screw them.  Machine interaction is maintained for the routine stuff but much is not routine.  It is heavily customized.  In effect, each customer has become the designer of an experience that the organization helps to provide.  Much of the organization staff serve as consultants for this design.

Still another part of the object is for the organization to be socially responsible.  There are several components to this.  One, of course, is to embrace environmentally friendly production techniques.  Another is to be a contributor to the community where the organization is located, to help keep it a place where people want to live and interact outside of work.  A third is to be a fair competitor in the marketplace.  This means that product and service quality are what the organization focuses on.  The organization shuns market manipulation via merger or acquisition, predatory pricing, and other unfair practices.  A codicil in the organization charter prevents individual shareholders from  concentrating ownership.  The Virtual CEO software maintains a steady vigilance, much like current day anti-virus software, to ward off attempts at stock purchases aimed at concentrating ownership, masked by the names of faux individual owners.

Companies run by the Virtual CEO software might still fail from time to time, either because the product they set out to develop failed to realize its potential or because competitors came up with something better and they couldn't catch up.  But companies managed by the Virtual CEO software share information to better mitigate general business risk and thereby to better anticipate where they should be heading.  The Justice Department is okay with this sort of information sharing because they know it won't be used for insider trading or other market manipulation.  In this way the market coordinates where human managed firms never could, so some of the needless destruction we associate with capitalism is avoided.

As a consequence, the half-life of a Virtual CEO managed organization is longer and such places of work become attractive career opportunities. In turn, part of the Virtual CEO software aims to manage employee careers, providing opportunities for personal growth, suitable mentorship, and helping each employee balance work with life events. 

* * * * *

This vision is deliberately ironical in viewing automation as a substitute for the top executive function, instead of how it currently is regarded by going for the lowest rung of the ladder first and the climbing to successive rungs.  It seems clear who in the current ways of things would resist this sort of change, so would be cast in the role of heavy in our story.  I wish I knew how to write a compelling short story myself, that would be a fun and engaging read, to create a broad audience for these ideas.  But I lack that sort of skill.  Perhaps one of my readers will take up the challenge.

To conclude this piece,  let me note that once the premise from the previous section is embraced, that executive function can be performed well by artificial intelligence, then there are clearly many other areas of our economy where we'd want to see it deployed.  Given its current very low approval rating, who wouldn't want a Virtual Congress, for example?

But the real reason to have such a story, or perhaps many such stories with a similar theme, is to get us to ask what would things look like if they were fundamentally better than they are now.  If we could actually agree on that, couldn't we try to head in that direction without machines needing to run the show?  Then, wouldn't we have benefited from Welland's vision even if he was quite wrong in his prediction?  Sometimes, I think, it is better to have the intelligent mistake than the right answer.

No comments: