The foundation of strong economic systems is intelligence that emerges through the participation of individuals in the economy. The intelligence may come from groups of people, organized deliberately or arising from contemporary fashions or concerns. The economy itself is a by product of a multitude of transactions and negotiations between individuals or groups they participate in. The economy is a living intelligence in that it is always active in the present in order to make deals with whatever opportunities are present at the time.
One persistent human goal is to comprehend economic systems like we comprehend mechanical machinery. Human’s make their economy. Granting my above assertion, the economy owes its existence to human intelligence. It seems reasonable that we should be able to control an economy like we are able to control machines. We should be able to construct economies with precision to provide a warranty of its operation withing specification. We should be able to operate the economic machinery with as much confidence and predictability as we operate other machines in our lives. We should be able to drive an economy as if it were a car.
That is the dream of central planners and of the various forms of socialism. The goal is both appealing and very reasonable from the analogy of other human successes of mastering their creations. The logic of this argument compels us to keep attempting an engineered economy despite recurring and frequently disastrous failures. Even within economic science, the statistical models make an implicit assumption that an economy is something that eventually we will learn how to turn over to some individuals to drive in order to deliver future promises of distributions of wealth. The theories that reject this notion of controllable economies often assert this rejection as a law to restrain the models from the predictions they would otherwise naturally make. There is no proof that economies are necessarily uncontrollable. The theories that argue against central planning merely assert that such planning will fail. Coincidentally, their predictions are accurate, but the predictions of unmanageable economies are just an assertion of belief. There is no scientifically-proven law of nature that demands that economies be uncontrollable in the way we dream of controlling them.
Much of the current appeal of big data is ultimately a resurgence of this desire to control economies. With the recent access to technologies that can master the three Vs (volume, velocity, variety) of big data, we find new hope that we finally have the tools to realize our dream of a managed economy at both micro and macro levels.
Political or sociological theories provided the basis of past attempts at managed economies. Because human thinkers constructed these theories, other humans had authority to argue against those theories. Ultimately the concepts for controlling an economy were just some human idea that is subject to suspicion that the theory may be wrong. Even when theories are practiced, there remains reasonable disagreement by competent thinkers.
With the new era of big data, we found a basis for theories that goes beyond human ideas. We imagine we can control economies using data alone. The new promises of big-data driven management of economy come from the facts that that we can trust the observed data. Through design and past performance, we also trust the algorithms we use on these data. In fact, we are observing many successes in using data this way to obtain beneficial results. There is a growing enthusiasm that our data technologies can finally deliver us the economy of our dreams.
I fear we are setting ourselves up for another disappointment. We are missing something fundamental about how economies operate. Economies operate based on negotiations. Individual transactions involving persuasion to find some mutually beneficial arrangement. There is nothing new about either or both parties to a transaction to use data to persuade desirable terms. However, ultimately, the agreement involves an intelligent decision that evaluates the terms in terms in order to accept or reject the offer. The intelligent actors are integral to the healthy economy.
Our appreciation of the power of science convinces us of the possibility that economies can operate on some mechanistic principles. Economies are just another human invention. Like all human inventions, we ought to be able to control their parameters to engineer better versions that operate with better performance and predictability. As a result, we keep trying to apply the latest knowledge to the project of engineering an economy. Today’s promising tool is big data.
An alternative model of intelligence is that it is analogous to living intelligence instead of mechanical devices. Unlike machinery that comes from application of human intelligence onto relatively unintelligent materials, economies emerges as an aggregation of actively participating intelligence.
The difference between machines (with predictable engineered qualities) and economies is the difference in their component parts. The parts of the machine do not argue back, at least not in the time it takes for the machine to operate according to its design. The parts of the economy involves quarrelsome components.
The scope of economies is human experience. That scope provides ample opportunity for participating humans to argue back. The goals of a successfully managed economy is one that delivers benefits over human lifespans. We will evaluate the success of any economy by comparing our present conditions with the conditions of our past. If the economy fails to deliver value to the participating humans, the humans will withdraw their participation and eventually find some other economy.
The promise of 3Vs is that data can be acted upon faster than humans can think. This is the model of program trading. Operating on data at this speed can demote humans to be more like material substances in engineered machinery. Our intelligence is inert at the time-scales of the operation of the algorithms. I agree with this concept. As data systems operate at rates far faster than humans can think (let alone argue), then we may be able to conceive of a mechanistic version of an economy. Ultimately, the humans will evaluate the economy on terms of their experiences. Eventually, we will judge the resulting economy as we would judge any economy. If we do not find it working for ourselves individually, we’ll reject it either through withdrawal of participation, redirection into underground economies, or outright rebellion. Big data can achieve its goals of managing something, but that something may turn out to be an autocratic economy out of necessity because too few are voluntarily participating.
Big data can’t engage in argument with humans. Humans can not negotiate with algorithms. Simple example is machine learning that involves training on prior data (supervised or unsupervised) and rigid application of its learning on the present environment. In IT, we make a distinction between production and staging where production is where things operate consistently until they are replaced by something new coming out of staging. Only the production operates on the present tense real world but we demand that it does not change its principles until we replace it with something new. The IT concept of production is similar to the AI concept of trained machine learning: the operational stuff is blocked from negotiation or argument.
In IT, the concepts of cyber-security come from the goals of assuring that production remains secure in its defined intention. In effect, what we call malicious hacking is a form of negotiation that is an inherent part of economy. The cyber economy criminalizes natural human economy. We popularly reject the actions of the cyber-criminal, but his activities are similar to every-day economic transactions. We demonize the cyber-criminal because of the scale of the negotiations possible and the fact the negotiations involved a non-human partner (the computing system). In this concept, cyber-crime is everyday economy engaging with cyber components. As computing becomes more engaged in economy, it is natural for human individuals to attempt to negotiate economic terms directly with the computers. As computing becomes more involved with increasing size of the overall economy, we can expect more people to begin to redirect their economic behaviors to make deals with computers.
Normal successful negotiation of favorable terms from a production system will involve obtaining something that the designers of that system did not intend. Normal economic successes like these are criminalized. The only non-criminal transactions with computing systems are those that were intended when those systems obtained approval for operating in production.
A data-driven economy is not a free economy. While there remains promise that algorithms acting on vast amounts of rapidly arriving data can produce a better economy, I am suspicious that such an economy will eventually languish because it robs the human actors of their ability to negotiate. The vitality of a free economy derives from individual freedom to negotiate terms of engagement. Eventually, A data-driven economy may prove to be superior but it will succeed only by suppressing natural human negotiation. Human actors negotiating in a data-driven economy must negotiate with machines. Applying approaches that work for other humans to machines instead is criminalized as cybercrime. Human negotiation involves coming to terms with weaknesses as well as strengths. Everyday economy involves negotiation and persuasion to get the best deal, but any attempts to persuade a machine is a crime.
Big data is increasingly controlling everyday economy. As a mundane example, grocery stores use loyalty cards to measure the purchasing patterns for the patrons of the store. The analysis involves aggregating the patrons into clusters based on shopping patterns observed from the habits of the individuals within the group. Even though I do not know exactly what analysis or techniques they use, I am confident that if I notice a change in the store’s inventory, it is because of what they are learning from their data.
One example I noticed is in a change in the size of cut of meat packaged in the meat aisle. For a long time, I would be able to get a package of 4 cuts that would range between .9 and 1 pound. This suited me perfectly. Then, the same 4 cuts started to show up around 1.5 pounds. Now, I never see anything less than 1.25 pounds unless I ask for it specifically. As usual for my personality, I rather just go along with the flow, buying the bigger cuts and slicing to my satisfaction at home.
That’s my personality. I don’t want to ask for anything special for myself. I always try to adapt to what is available instead of demanding custom services. I like living off-the-shelf. Coincidentally, this personality trait is also one that is ideal for observing the data system at work. The store is stocking the shelves based on what their analytics are telling them about their customers. I can learn a lot about my peers by just looking at the contents of the store shelves. In the past, I liked this particular store because it seemed to cater to single people living alone. Over time, I’ve noticed that the store is now catering to people who have larger dining tables. However, the size is due to entertaining friends rather than feeding family. When stock changes, it is generally in the direction where the replacement item that is more likely to be impressive to guests.
My paying attention to the stock of the store is in reality my private attempt to do data analysis of the store’s data analysis. I use the stock as a proxy for the data they must be acting upon. I reverse engineer what they are seeing.
Also, I don’t bother to ask for specific requests because increasingly it seems pointless to negotiate with humans. Data analytics of the habits of my peer shoppers are determining what gets stocked on a routine basis. I’m just increasingly alienated from my neighbors. That is no surprise.
My project is to negotiate with the machine to get what I want. The store scenario is a trivial model of the larger economy where the real boss in control is the data analytics. The guy we need to negotiate with is an algorithm that lives on data. The aphorism of “you are what you eat” is especially applicable to this boss: the algorithm’s behavior is determined by the data it ingests. Negotiating with this boss requires carefully adjusting the data it gets to eat.
The bigger problem is dealing with larger the economy, but the store example provides a convenient subject to test some ideas. As an example, I am a big fan of one particular style of the fresh bread that arrives daily. This section is relatively small with just a couple loaves of various types. The type I like most is an unsliced round loaf with a sturdier texture than the more popular long loafs or breads with soft texture. In the past, the store would get 2 loaves a day so that I would have a chance to get one if I arrived early in the morning. Recently I learned through observations that this strategy was no longer working. Despite my regular appetite for this bread, they stopped stocking it. I can see the problem. if I was the only one buying it. A loaf lasts between 3-4 days so my re-appearances to buy this daily bread is unpredictable. What doesn’t get sold that day gets thrown out if not sold by the end of the day (apparently, because I never see day-old dates on the bread).
I self-impose a constraint that my only means of negotiation is my choice of shopping for that day. Thus, when I arrive at the store to find my favorite bread missing, I have to be careful about providing the right data back to let them know of my disappointment. For a while, I made the mistake of buying an alternative variety that was in stock. This was amusing because I noticed quickly that the stock of that particular variety increased after I started buying it. It is as if they noticed that I can be satisfied with a more popular bread style. In exchange for taking away my first choice, they increased the stock of my second choice that happened to be more popular with my neighbors.
In other words, the store’s algorithm is negotiating with me to change my habits to be better in line with my neighbors. So that is how it works. The store’s algorithm is negotiating with my personality to get me to change.
The vibrancy of a free market comes from two-way negotiation. I need to send a message back to the store that their deal is not acceptable. A couple times I would just walk out of the store if my bread was not in stock, even if I had other items I had planned on buying if the bread was in stock. The problem with this is that although the people in the store might (although rarely) notice me leaving empty handed, this information doesn’t get captured as data for the algorithm. I need to make a purchase to show that I didn’t purchase any bread. Normally, my checkout will always include a loaf of bread but I can see this is ambiguous data. Just as my buying my second-choice style is evidence of a change in taste, my not buying any bread is evidence of change in taste away from bread in any form. Clearly, this is not a good strategy.
Eventually, I learned that my favorite style of bread does get stocked very well on Sundays. This makes sense because a round loaf is idea for a Sunday meal gathering (brunch or meal) for close friends and family. I also learned that if I wrap the bread in a plastic bag, it will keep for several days. I could buy two loaves at a time and it would last most of the week with reasonably satisfying bread.
So, short on bread on a Sunday shopping trip, I see that there are 4 loaves of this bread. I imagine the algorithm estimating there will are typically 3-4 Sunday feasts that call for this bread. I’m not one of those 3-4, but I could buy two then it would satisfy me at least until Friday.
Buying two loaves when I only desire one seems to much like hoarding. In addition, I imagined my purchase of two loaves for a week would result in 1-2 disappointments of later arriving customers who had their goals on getting this particular bread. (I’m biased in that I really do like this bread, especially when it is same-day fresh).
However, the modern economy involves negotiating with machines and the way to communicate is sending purchasing signals. Buying an alternative variety of bread convinces it that I’m satisfied with something else. Buying no bread convinces it that I’m no longer a bread buyer. Buying two loaves on Sunday says I like this bread in particular. In fact, if someone complains about the store not having this bread on Sundays, they will order even more for Sundays. In the end, I’ll have to be satisfied with a weekly refresh of my bread stock instead of 3 day restocking. This is a negotiation. The store can’t stock what I want daily. I don’t want what the store can stock daily.
A reasonable compromise is that we’ll both settle on a Sunday hoarding of two loaves. I can live with that. This is a negotiation in a free market, but the negotiation is human-to-machine instead of machine to machine. However, it is also in some ways illicit because the algorithm developers did not intend for someone to be manipulating them this way. I’m using data to “hack” into their algorithm to get what I want. In this case, I think imagine it to be mutually beneficial, but I can see the data scientists disapproving of my deliberate attempts to manipulate them when they are more satisfied with manipulating me.
That’s how negotiation works in a free economy. Despite a mutually favorable deal, there is always a level of resentment on both sides that the other didn’t give them all they wanted. The store’s disappointment is not pushing onto me the daily bread they stocked mid-week. In the end, they may be still throwing out expired bread.
This is just my second week at this message so it is too soon to know if it will work. I wouldn’t be surprised to find a sign next Sunday saying “limit 1 per customer”. Machines don’t negotiate.
One thought on “The meaning of persuasion in the age of data”
Pingback: Discourse between analysis, synthesis, and observation | Hypothesis Discovery