Saturday, March 2, 2013

What humans do best - part 1

I've been thinking on something I heard today in a economics podcast where the speaker was talking about what is likely to happen to our jobs in the coming years. He made the case that most of them will suffer the same fate as farm jobs and for a similar reason. It isn't that they will all be sent overseas. It is that they will simply cease to exist because no human will do the work anymore. This wasn't news to me since I see it at work. In fact, it is my job to make it happen. The blunt truth is that I work to eliminate tasks from the lists worked by people by automating the work. The jobs I impact don't usually go away, but they certainly do change.

What caught my attention was the speaker's focus on the kinds of things we like to think that we like doing. What can humans do well that sets us apart from the tools we fashion? It used to be that no computer could beat the best human chess player. That is no longer true. It used to be that no computer could safely drive a car in a complex urban environment. There are very bright people demolishing that belief right now and doing a good job of it. Offer up a physical task and there are many who want to tackle it and automate it so the take away lesson is to be cautious of claiming any reserved turf at all.

There was a class of problems, though, that might be immune. It isn't that we can't automate solutions to these problems, it is just that we probably won't. These are the problems we LIKE to solve. They don't have to have much else in common except the simple fact that we like them, thus the hurdle to automating them is higher in the sense of the costs involved. Someone automating drudge work like I do might be met with concern from those suffering the changes we bring upon them, but after it is over and their job descriptions adapt to the new reality (assuming they still have the job), we usually meet with smiles and appreciation and a desire for us to go away so it doesn't happen again any time soon. However, if we try to take on a task they seriously enjoy, they will fight before and after and impose extra costs upon any who would pay us to do the deed.

A psychologist might examine why we like certain problems even to the point of resisting giving up the work when others can do it better, cheaper or faster. They might examine motivations and rewards. A biologist might examine the same behavior and study survival odds of those who keep the tasks compared to those who don't. Each of those might be interesting paths, but the one that caused my ears to perk up was the biological one as it relates to the problem of caring for one's family. All animals face this kind of problem and different solutions arise with different species, but among humans there is evidence that our oldest solutions were overlain by newer ones specific to certain settings and when that happened modern humans became what they are today. We became a rapidly growing species that displaced all other hominids and is in the process of displacing much more distant relatives to such an extent we should refer to modern times as a mass extinction.

In the ideal case, the problem one faces with the simplest task of feeding oneself and procreating in a paleolithic era with associated tools is still complicated, but it isn't much more complicated than other mammals have to solve. Our task was a touch more complicated because our primary competitors were other hominids and ourselves. When the others were gone, we faced a positive feedback loop by competing with each other.

Ideally, though, the problem of meeting the basic needs of a human family is all about providing food, water, shelter, and security. How we solve for all the preferences of our family members varies considerably depending on the means we have available and the relevant information we need. If we have it ALL, though, the problem resolves to one of optimization. If you are the head of a family it isn't hard to imagine that the depth of your knowledge and ability to think through all the options associated will all available resources is how you find the optimal solution. If your family members are also talented, you might delegate or share the planning tasks. How many physical brains are involved in the planning effort doesn't really matter. They key measure is how the preferences are communicated and resources assigned. If several people are involved and closely coordinating they function effectively as one person. If they are willing to take food from a baby that has more than it needs to give it to another baby that doesn't have enough, they are optimizing in this fashion.

From a biological perspective, it shouldn't surprise anyone why we like to solve this kind of problem ourselves and only surrender it grudgingly. We are the children of past generations who solved this problem well enough to procreate. We should have an evolutionary inclination toward wanting to solve these problems for the simple fact that those who don't want to slog through the hard work are less likely to have and raise kids.

From a biological perspective, it shouldn't surprise anyone why some of us get annoyed with parents who do NOT like to solve these problems. If you are the kind of person capable of caring about the welfare of children who are not your own, you might be tempted to expand your optimization effort to include them for moral reasons. Basically, the other parents are freeloading on your good will. While the problems are easy to describe they are hard to solve. Even if we find solutions that don't deprive our own families too much to cast our wider net of concern, it is still hard work we don't accept easily.

The optimization problem in a paleolithic setting is simpler than in a modern setting. There were vastly fewer people then and our knowledge of how things worked was much more limited. There is no doubt our ancestors knew things we have since lost, but they didn't need to know as much as we do and absent many of the resources we now have available it is easy to argue the knowledge wouldn't have done them any good anyway. What need did someone have for trying to accurately predict the position of Mars in the sky to within one arc-minute if they lived 20,000 generations ago? Their problems were still very complex, though, even if one could know all the details needed as inputs. Some did it well and we are their great-to-the-nth-grandchildren.

There are two twists to this type problem that make them even more difficult. The most obvious one is that the planners don't have ALL the information. F. A. Hayek pointed out that they can't. In fact, they can't EVER. There is the annoying little possibility that approximate information might not be good enough too because the optimal solution might be very non-linear. In other words, small changes to the inputs might lead to wildly different solutions. Think about the weather on Earth and you'll have an example of that. Set that aside for now, though. The information the planners need can't EVER be completed even in tiny family settings, but it might be possible to get close enough if the solution space isn't too non-linear. The problem with completion is that it is quite like a person's preference is unknown to them until they are faced with a decision and options from which to pick. They learn OF their preference by the result of their decision. Without a time machine, there is no way to get that preference back to a planner. 

How can a planner know the preferences of their family members when no one CAN know until the preferences are discovered? What we usually do is choose for them in advance and hope for the best. What do you want to eat for breakfast ten years from the day you first read this? Do you know? Why would you bother figuring it out, let alone recording it for someone in your family to help plan for it? You might be willing to plan a week ahead or do simple budgeting to help, but that is potentially useful. Why would someone in a paleolithic setting bother doing that? They couldn't be sure they would survive the next winter, let alone eight or nine more.

The second problem is  a mathematical one and is FAR from obvious. It is related to the fact that multivariate optimization problems become hideously complicated when the number of variables grows large. It gets so bad that the odds of finding a solution drop to near zero even if the number of possible solutions is finite or constrained as we expect they might be. I'll save this for next time, though.  In a nutshell, though, you need a time machine AND omniscience to pull it off yet mortal humanity does moderately well in finding solutions. If you write algorithms to automate problem solutions as tasks, be prepared to face your limits.




Post a Comment