Friday, December 16, 2011

Artificial Intelligence

No, this is not another story about killer robots....

I just finished "Patches", a science fiction short story by Lesley Choyce.   It is about a dystopian society where all human beings were connected to a central information store called "The Source" through a skin patch that provided a direct interface between the human brain and "The Network".  The Source was sentient, and had taken control of The Network and all humans, issuing pain and information through the patches to maintain order in society.  An android appendage of The Network made a statement that all human beings were equal and they were no more than a collection of their biological parts.  His belief was that because the Network contained all the information known to humans (by tapping directly into their brains and all other data storage devices) it also possessed all the actual knowledge in the world - because knowledge and information were identical.  A small group of humans, struggling to free themselves from the Network, completely disagreed, and said that knowledge was much more than simply a collection of information.

I think this story points out a very likely failure in a machine ever achieving true artificial intelligence.  It seems very plausible to me that a machine would be unable to discern the difference between information and knowledge.  I think this concept may be one of the fundamental barriers that must be overcome for any machine intelligence to achieve sentience.

Unfortunately, many corporations are falling into this same trap - and I believe IT professionals have merrily led the corporate leaders down this false trail.  Since the beginning of the computer age, we have been chanting the power of information - and we have said that without good information, a company cannot make good decisions and is doomed to failure.  Heck, even our own names for what we do: "Information Technology", "Information Services" or even the old "Data Processing" show that we are focused on the collection and management of the information - not the knowledge that information represents.

There is no better example of this than the typical corporate HR form used to apply for a new job.  A couple of years ago, I wrote about the ridiculous amount of information an applicant must provide in order to apply for a job using the typical online recruiting systems.  I would claim that most of this data provides nearly zero additional knowledge to the company - it certainly is not used during the typical hiring process, or if it is, it is used only in the most crude way to filter out candidates and reduce the size of the potential candidate pool.

The same is true for a vast amount of other information gathered from all across the corporation.  There are massive amounts of data being collected, massaged and reported that provide little or no additional knowledge to the company.  This represents a huge expenditure of corporate resources for absolutely no business value.  It also demonstrates a fundamental problem within many corporate boardrooms: the inability of senior business leaders to define the key metrics that are used to run their company.

There are certainly some companies that have defined their Key Performance Indicators (KPIs), and they really do use those metrics to make tactical and strategic decisions.  If you work in one of these companies - you are fortunate - count your blessings.  It was only when I moved to a company that did not have defined KPIs that I learned that it was even possible to run a large corporation without good business metrics.

What is really sad is that it seems these companies do not understand or recognize the issue.  To them, sitting in an Executive Team meeting and suddenly asking:  "How did the flood in Thailand affect our sales to Southeast Asia compared to how our sales were affected in 1980 by the election of Ronald Reagan?" is a critical data point that will drive all future corporate strategy.  The rest of the executives turn to the CIO and say:  "That's a simple report, you have all the data, right?"  And the CIO's response is (of course) "I'm sure we do, we'll have that for you later today."

After delaying 3 critical projects and pulling four straight all-nighters, the report is finally produced and scheduled to run weekly for the executive team meetings.  However the CIO is mad that your data warehouse and business intelligence tools are so poorly implemented that you couldn't meet his expectations.  When the report is sent to the executive team, there is a "thank you" reply from the CEO - and nothing is every heard about that report again.  Six weeks later, you check the logs for that report, and you find that there has been exactly one person who has ever accessed the new report - the IT analyst who was tasked to make sure all the reports run successfully every week...

OK, the report criteria I used for this example was fictional, but everything else was a regular occurrence.  The senior leadership believed it was perfectly acceptable to ask any random "I wonder..." question and the data would be available and the report easily produced.  It never occurred to them how disruptive their seat of the pants management style was to everyone involved in producing those ad-hoc reports.  And even worse, the ad-hoc reports were often not actually used for anything - they were apparently just an idle thought that wasn't actually the basis of any decision making process.

In some companies, this behavior is combated by purchasing more ad-hoc reporting tools in an attempt to provide the executives and their staff the ability to run their own reports.  That may work in larger companies where their is enough staff to cater to the executive whims - but not in companies that run much leaner.  Unfortunately, expecting executives to do their own ad-hoc reporting is a foolish hypothesis - it's simply NOT going to happen, and the result is the worst of all worlds - you have spent the money on the easy to use tools, you do not have the staff to do the ad-hoc analysis, and the executives are still asking the questions and looking at the CIO to provide the answers.

Now, let's say that you are finally fed-up, and you decide to launch a new business intelligence initiative.  The consultants tell you that step 1 is to define the KPIs that are used to drive the business decisions.  So, you go to the executives, and you ask:  "What are the key metrics that drive your decision making processes?"  You get a blank look, followed by the typical executive 2-step shuffle that makes this your fault for not providing the executive with the proper information for them to make *this* decision.  You have no choice but to beat a hasty retreat and rethink your chosen career path.

When you get back to your office and calm down, you realize a startling fact:  the executives do not really know how they make decisions.  They *are* making decisions - they do it every day in all parts of the business - but they simply can't define what information they use to do it.  Even more frightening, you also realize that if they have not defined those metrics, there is absolutely no way the decisions they do make can be consistent, logical or data driven.  Good grief, no wonder the decisions often seem to be somewhat random - they *are* somewhat random!

In an ideal world, the key business decisions are driven by specific data elements and threshold values.  Above the threshold, the decision is X, below the threshold, the decision is Y.  If you have all the key operational metrics defined, your business can almost run on remote control.  At least, that's what the business consultants and best selling authors say.  Of course, in the real world, there are always un-quantifiable factors that influence the decision.  It might be office politics, shareholder sentiment, perceived risk or simply a hunch.    However, should you let those influences drive your decision in the face of opposing data?  Of course, if you do not have the data (opposing or otherwise), then those influences are all you have.  That doesn't seem to me to be a prudent business strategy.

Developing KPIs really is difficult, I'm certainly not trying to say it is easy, because it is not.  You must force yourself to determine the key operational decisions, and also determine what data you use to make those decisions.  In addition, you *must* also determine the threshold values for each data element.  Without a defined threshold to tell you whether to move the lever left or right, you are simply looking at data without a business context.  You may find that some of your decisions are based on trends - upward, flat or downward - once again, without knowing how you will use the data, seeing the data will not help you make the decision.

I suspect that some executives do not really want to define their decision making processes.  They may believe, or want others to believe, that their decision making is a combination of unique insight, luck and magic that is only found within their well-paid head.  If I was an investor in a company with leadership like that, I would be very, very nervous.  I don't want to invest in unique insight, luck and magic - that might be OK for betting the longshots at Santa Anita horse races, but not my 401k money!  I want leadership that can articulate how they run the business, and I want to know that they aren't just flying by the seat of their pants.

I urge all business leaders to take a critical look at the information they are collecting, processing and reporting.  If you can't directly link the data to a specific business decision - then stop collecting and reporting on that data - it's a waste of your precious resources.  Encourage employees to question all requests for new data and new reports.  "Because the VP wants it" is NOT a good reason.  Every report should have a purpose, and "I've always run that report on the 3rd Tuesday of every month" is NOT a valid business purpose.  Don't just collect information - always strive to create knowledge.                            

No comments:

Post a Comment