Sunday, March 11, 2012

You Can Lead a Horticulture…

Another year, and once again Apple has managed to dominate the attention of the global media by releasing a new version of the iPad.  Curiously enough, even news outlets that didn’t think it was big news carried it as their top story.  For example, the New York Times article described the improvements in their headline as “modest”.  And yet, it chose to carry that story on the front page of their website.

Well, I’m sure it made sense to them.

Whenever Apple hits the news like this, the world seems to split into two extremely vocal camps.  Camp number one loves Apple and follows every detail.  Camp number two loathes Apple and disparages Apple followers as “sheep” and “lemmings” for snapping up every Apple product.  Curiously enough, they also tend to follow every detail.  (There’s a third camp who doesn’t pay any attention to any of this, but since they tend to be a quiet bunch, we can safely ignore them for this exercise.)  Without stepping too deeply into this debate, I would like to focus on one particular criticism that is sometimes leveled at Apple, which is that its products should not be purchased because they are much more expensive than competitor’s products with the same specifications.

This criticism represents extremely sloppy thinking.  I might even go so far as to call it stupid.

Before the deluge of hate mail begins, let me make clear that I’m not criticizing people who dislike Apple products.  I’m not even criticizing people who choose not to buy Apple products because of their specifications.  However, I am criticizing people who think that the technical specifications are necessarily the only relevant criteria for purchasing a product.

To illustrate what I mean by this, let’s take a look at a branch of artificial intelligence (AI) known as expert systems, and more specifically at automated rule induction.  I find it useful to consider artificial intelligence for subjects like this because AI was designed to replicate, at least superficially, how humans think.  By understanding the limitations of AI, we can sometimes gain insight into the limitations of our own thought processes.

In expert systems, knowledge is represented by a series of rules.  While the concept of an expert system as a form of artificial intelligence is only about forty years old, the idea of capturing knowledge in a simple rule is ancient.  For example consider the following sayings:

Both of these represent knowledge within a particular knowledge domain (meteorology and herpetology respectively) captured in a simple, rhyming rule.  With these types of rules, it becomes much easier to communicate, store and pass this knowledge on.   An expert system uses rules similarly to develop a knowledge base about a knowledge domain.

The creators of the first expert systems collected knowledge manually.  They would interview experts (hence the name) in a particular knowledge domain, and codify the knowledge gathered from these experts in the form of rules.  As the field developed, the concept of automated rule induction was introduced.  In automated rule induction, a computer uses various algorithms to analyze a set of data, and derives various predictive rules from that data set.  For example, perhaps you would prepare a set of data which includes a simplified description of the weather (such as “clear” or “stormy”), the time, and the color of the sky twelve hours previously.  The computer analyzes the data, and discovers a strong correlation between a red sky at night and clear weather the following day.  Similarly, a red sky in the morning is strongly correlated with an upcoming storm.  This example is trivial, but when you increase the number of potential variables to hundreds, with thousands or millions of data points, computers can become very useful at spotting patterns and relationships that a human would miss.

There are two challenges in running an effective automated rule induction program.  The first is to avoid rules that are too narrow.  For example, the computer might produce the following rule: “Whenever the date is August 29th, 2005, there will be a hurricane.”  This rule would be 100% accurate based on the available data.  However, it’s also not particularly useful, because that date will never come around again.

The second problem is a rule that is too wide.  For example, the computer might produce a rule that says “It will snow whenever it is Christmas”.  The computer might choose this rule because it finds that it was true more than 50% of the time in the data set it analyzed.  But it’s so vague that it has extremely limited predictive value.  It is, to use the colloquial term, a “stupid” rule.  If you want a good weather forecasting program, you need a computer with much more intelligent rules, which look at a number of relevant variables and model how they interact.

It’s this second problem that trips up so many people in real life.  Somewhere along the line, they get an overly simplified rule in their head, and for whatever reason, they stick with it.  Maybe it’s too tiring to keep running the automated rule induction program in their brain.  Despite the constant influx of new data to the contrary, they continue to operate with an over-simplified view of the world that ignores most of the key variables.

In fact, this is exactly what racism is.  Whatever else you can say about somebody who is racist, you can also, quite objectively, call them stupid.  This is because the rules in their head which judge people look only at a single variable, and ignore the more complex factors and relationships which are necessary to develop a more sophisticated judgment of a person.  Racists are not intrinsically stupid – they might be quite sophisticated when it comes to other knowledge domains.  They’ve simply picked up a bad set of rules for this domain, and kept them despite evidence to the contrary.

Curiously, this same phenomenon proliferates in the “sophisticated” world of technology as well.  Dave Pogue discussed this at length in a 2007 column, in which he challenged the assumption that more megapixels on digital cameras result in better image quality.  It’s easy to understand where this assumption came from.  In the very early days of digital photography, megapixels were so limited that any time they increased, you did see a noticeable improvement.  But somewhere along the line, the megapixel count stopped being the limiting factor, and other variables, such as the lens and sensor, became much more significant.  But megapixels are simple.  They're clearly labelled on the box.  If you buy based on megapixels, you don’t have to think very hard about your choice.  You can get by using simpler rules.  You can, in short, be stupid.

Which brings us back to Apple.  The differences between an Apple laptop and, for example, a Dell laptop are so numerous that it’s difficult to list them all out.  OS usability.  Form factors (such as size, weight and shape).  Available applications.  Available interface ports.  Susceptibility to malware.  Compatibility with other devices.  Access to friends or colleagues who can provide support if need be.  Any one of these might be important to a particular person.

Computers are, in a word, complicated.  Attempting to define them solely based on a small list of metrics such as gigahertz and gigabytes is simple.  It’s easy.  But it is also, unfortunately, just a bit stupid.

No comments:

Post a Comment