We are constantly reminded that we live in an information society - more than a society: an information world. We are told that we are going to drive on a superhighway in this world; or perhaps we will stay at home while the information that is on the superhighway, races toward us at the speed of light. Whichever does the moving, ourselves or the information, we are going to be immersed, we are told, in a great bath of up-to-the-moment knowledge about the world in which we make our management decisions. We will have accessible data bases, frequent reports, E mail, video conference calls, to say nothing of the more traditional telephone and 

All of this is heady stuff, suggesting that we are already on the very threshold of the 21st Century. My purpose this morning is to utter a few cautions and a few suggestions as to how we may deal with the abundance of information that is promised - or threatened. Superhighways, after all, frequently offer superabundance of traffic as well as superscarcity, of parking spaces. 

Identifying the Scarce Factors. The second task in designing a complex system is to identify the scarce factors: the bottlenecks that are going to place effective limits on the system's power to accomplish its goals. In the design of complex things, we cannot optimize over everything; our bounded rationality does not permit it. We must focus on those particular variables that are going to affect system performance most strongly and critically. These are the scarce factors. 

For the Easter Holidays some parents buy a rabbit as a pet for their children. As the rabbit also can be helpful in keeping the lawn trimmed, a superabundance of grass can be reduced by rabbits. Cautious parents acquire one rabbit, incautious parents sometimes acquire a pair. We all know what happens then: soon, we have, not a pair of rabbits but a lot of rabbits, and not a superabundance of grass, but very little grass at all. The factor that was originally scarce - rabbits - has become overabundant, and the factor that was originally overabundant - grass - has become rare. 

Information is not the Scarce Factor. Many of us have not noticed that a number of years ago we had already entered an era in which information was no longer scarce: in which information was thrust at us in a steadily increasing flow. By this, I don't mean that we always had the information that would be relevant to our decisions (e.g., whether the stock market would go up or down the next day, or whether next months' orders would increase or decrease). What I mean is that we were provided with large quantities of information which we could sift and filter to find the relevant nuggets, but which we could surely not absorb in its totality. 

Human Time and Attention is the Scarce Factor. Now if one factor in a system (in this case, information) has become abundant, what has become scarce? What has become scarce is the time of human beings, your time and my time, to process that information. While the means of acquiring information have been expanding, there has been no expansion in the number of waking hours in the day available for attending to that information. Time and attention are scarce, and becoming scarcer every day. In particular, we have no time to process the new information that is being disgorged on us. 

Filtering Uses a Scarce Resource. But we, and our assistants and secretaries for that matter, employ the same scarce resource - our time and attention - to filter information as we and they employ to make use of the information that is allowed to flow through the filter. Who is going to filter the increased volume of information that faces us'? Sixty or seventy years ago, the telephone company extrapolated the growth in use of telephone services, and from that extrapolation estimated the number of women they would have to employ as operators in the future. The estimated number turned out, in a very few years, to exceed the entire adult female population. AT&T then turned to automation of the telephone switching system, with the results we are all familiar with today. 

Filtering Must be Automated. The task of filtering information must follow the same route of automation. But the task of filtering all the kinds of information that flow through a communication system is far more difficult than the task of connecting a phone with a dialed number. In the dialing case, the filtering is really done by the sender of the communication; only the actual operation of the switching circuits was automated, not the decisions about what was to be switched where. In the task before us, the automatic system must make the decisions about what information is to be transmitted to whom. 

Filtering Requires Intelligence. Filtering information requires understanding both the information and the needs of the user sufficiently well to be able to determine what is relevant and what irrelevant, what is important and what unimportant. In other words, it requires intelligence - either human intelligence or artificial intelligence. And as the human intelligence is already fully engaged - that is why there is a problem - artificial intelligence is what is required. 

Intelligence is Mainly Non-Numerical. The information that people use in making decisions, especially the really important decisions like those we often call "policy," is predominately non-numerical. When I speak of artificial intelligence, I am not referring to the well-known ability of computers to crunch vast quantities of numbers - however important that ability may be for engineering design, scheduling, or some kinds of modeling and fiscal planning. Computers have no special affinity to numbers; they process symbols, that is to say patterns, of any kind, including verbal, diagrammatic and pictorial as well as numerical. 

Much of the information that is needed for management decision takes the form of words (e.g., in trade magazines, in correspondence, in reference manuals), or diagrams and pictures (product designs, architectural drawings, or whatnot) and is not numerical. In designing a communication systems, including its data bases and its artificially intelligent components, these non-numerical forms of information will undoubtedly play a much larger role than the numerical. 

The Important Information is Not All Internal. Much, usually most, of the information that is important for management decisions originates outside the company. The information system must be designed to include this external information, and decisions have to be made about when to generate it as a corporate effort and when to buy it. 

Hence, in all of my discussion of the corporate information and communication system, I am assuming that all the various kinds of relevant information, verbal, diagrammatic/pictorial and numerical are included, and that the sources of this information will be both external and internal to the corporation. 

Let me cite some examples, not all of them dealing with top management decisions, but all of them relevant to how we will be able to structure information for top management decisions. It is a commonplace today that the information needed to determine the credit-worthiness of a prospective borrower (at least for consumer credit) can be gathered, filtered and evaluated quite adequately by a computer program, and such decisions are being made as I speak here - probably every few seconds. There is a commercial product on the market today that makes medical diagnoses (filtering information about symptoms) as skillfully as most internists. There are computer programs today that can create a competent abstract of a technical paper from the text of the paper, and other programs that, upon reading such an abstract, can decide quite competently whether the article should be called to the attention of an engineer who has described (in terms of key words, say) his or her information needs. 

Notice that the programs or sets of programs I am describing actually do more than filter information. In many cases, they not only decide what information, selected from what is accessible, should be used, but they also actually use the information to make decisions. The role of artificial intelligence need not be limited to the filtering job, it can also participate in the decision-making job. 

Suppose we were to develop a scheme for evaluating our own information gathering and information filtering methods and habits? How well would we score on it? For example, it is rumored that a great many people read a daily newspaper- the New York Times, the Wall Street Journal, or-the Lord forbid-U. S. News. How many of the items they scan tell them things they did not already know (or that weren't also in yesterday's newspaper)? Of the new pieces of information, how many couldn't be reasonably predicted? And on which of these pieces of information did they act? And if they acted at all, did it have to be in 24 hours? Would a week have done as well, or a month, or a year? If you apply these tests to your own reading habits, you might well conclude that you should cancel your newspaper subscription, sell your TV and radio and purchase the World Almanac annually. 

But in criticizing our present habits of information intake, I have not addressed the question of error. The issue is not whether an automated filtering system would make errors; of course it would. The issue is whether it would make more and more serious errors of omission and commission than the present human system does. And the answer to that depends on the level of intelligence we are able to build into it - a level that increases with each year of our experience in building such systems. 

Years ago, when I took an airplane to New York and it was about to land at La Guardia airfield in grey, foggy weather, I would pray that it was being guided by a human pilot. Today, I pray that it is guided by an automatic landing system. Is this a matter of fickle affections - a switch from an allegiance to humans to an allegiance to machines? Not at all. It has nothing to do with my affections; it has to do with my assessment of the state of the art - my judgment of which system is the more reliable. Perhaps 20 years ago the human system was more reliable; now the automatic system is. It is by this same criterion that we will judge information filtering systems and automated management decision systems. 

Reduction in Communication Costs. The multinational corporation is probably far more a product of reductions in communication costs than of reductions in transportation costs, for it more and more decentralizes its actual manufacturing operations. In so doing, it places a new load on its communication system to preserve the reality or the fiction (we can discuss which of these it is) that the multinational firm is still a single organization. the danger is that, in order to maintain this reality or fiction, we will press toward centralization of decision making at the expense of creating and exacerbating the very information overloads that I have been warning against. 

Tendencies toward Centralization. Perhaps mountains have to be climbed "because they are there," as Mallory said of the reason for his ascent of Everest. It is not equally obvious that communications have to be sent because they can be. In past times, momentous questions of centralization or decentralization of organizational decision-making were often settled by the fact of physical distance. If the plant was 200 miles from the head office, the President could not wander casually into the plant manager's office to tell him or her what to do. I have met, in my career, more than one plant manager who resisted (or refused) promotion to a larger plant in the headquarters city for just this reason. 

Some years ago, the wireless and the long-distance telephone began to change things in this respect. (It has often been noted that the functions of Ambassadors were affected in fundamental ways.) But these were devices of narrow bandwidth and (until quite recently) considerable noisiness. A resourceful, physically remote manager could retain considerable autonomy if they provided the only link to the home office. Now the information superhighway, in its multitudinous forms, threatens to destroy that autonomy almost entirely. 

Is Centralization a Bad Thing? I have been discussing the cheapening of communications, and the centralization of decision making that is its probable consequence, as though they were bad things. That is obviously too simplistic a view, but it is not a view to be ignored. We should avoid concluding that, just because it is cheap to transmit messages, communication is therefore nearly costless. To draw this conclusion is to commit precisely the fallacy against which I am warning: the fallacy of treating communications, rather than the time required to deal with them, as the scarce factor in organizational decision making and organizational life generally . 

Designing an efficient and effective communication system under the circumstances I am describing requires determining which parts of the communication and decision-making processes (including the filtering) are most economically and effectively done by people, and which by machines. Having determined that, we can see where the bottlenecks are located and what the consequences are, in terms of the loads imposed on those bottlenecks, for centralization or decentralization. 

If we have information source I at location A, source 11 at location B and decisions K to be carried out at A and L at B, we can begin to ask such questions as: What is the relevance of the information in I to K and to L? And similarly, what is the relevance of the information in 11 to K and to L? What means do we have for filtering and compacting the information in I, and carrying out partial or full analyses of it, before shipping any of it to B, and ditto for the information in II? 

To carry out such analysis, we would do well to start with an analysis of the decisions themselves: the information that could and should be input into them, and the kinds of analyses that need to be performed. We need to discover also what parts of the information is in human heads and what part stored in data bases of various other kinds. 

It rapidly becomes apparent that creating an effective communication system is quite equivalent to solving a whole system of simultaneous equations. After a few possible basic plans have been designed, they have to be evaluated by seeing to whom (or what) they assign the decision making responsibilities, the analytic responsibilities, the data filtering responsibilities, and so on. At they same time, it must be determined where these components of the system are best located physically. And the analysis is likely to be valid to the extent that priority is given to the scarce resource: the human ability to use the information that is directed to it. 

This paper was presented at the Carnegie Bosch Institute's "International Conference High Performance Global Corporations", April 21, 1995, in Boca Raton, Florida.