Decentralized Decision Making vs ?°one size fits all (一刀切)" approach to governing
Every human being is involved with decision making everyday (ranging from
what to wear, what to eat, to career choice, marriage, and life-and-death
medical decisions). We all make such decisions either quickly or after long
deliberation and live with the consequences.
One of the most fundamental yet not often emphasized concept of
operations research, system and control, and management is that of a
STRATEGY, (or decision rule, or feedback control law, or learning and adaptive
control, if-then table, and host of other technical names). Speaking colloquially,
a strategy is nothing but a recipe (like a cooking recipe) for making certain
decision or taking certain actions under certain circumstances, e.g., if it rains
then I won't go out. Here °rain" is the circumstance and the decision is °not to
go out". More technically, we have input x which are the input information
representing the situation the decision maker (DM) finds himself or herself in.
Then depending on x, the DM choose the output decision or action y, i.e.
mathematically, y=f(x) where the function f represents the strategy. Thus,
the strategy is a recipe of choosing what under which circumstances and
when (mathematically, it is a mapping from the information domain x to the
action domain y). For example, if there are three possible circumstances
x = [ rain, cloudy, shine] and two possible actions y=[ go, don't go], then there
will be y^x (2^3=8) possible strategies. They are exhaustively: f1=always go no
matter what, f2=go only when it is rain or cloudy otherwise don't go, f3=go
when it is rain or shine, otherwise don't go, f4=go only when it rains
otherwise don't, f5=go when it is cloudy or shine but don't go otherwise,
f6=go only when cloudy don't otherwise, f7=go only when it is shine,
otherwise don't f8=don't go no matter what. (note: if 0=go, and 1=don?ˉt
go, we are simply enumerating from 000 thru 111 in binary fashion when
spelling out the eight strategies.)
Note the number of possible strategies explodes exponentially fast,i.e., y
raised to x power. Since for any practical problem the cardinalities of x and y
can be very large, the number y^x is extremely, astronomically, and
combinatorialy large. Often even simple enumerations of all possible
strategies are computationally infeasible. Of course the Holy Grail of decision
making is to make OPTIMAL decisions, i.e., finding the optimal one among the
domain of all admissible strategies. There are three major difficulties: First,
while in practice the decision choices, y, may be limited due to various
constraints to tens or hundreds, the information possibilities, x, can be very
large especially in this information age. We suffer from information overload.
Second, at the same time, in many organizations and bureaucracies, not all
relevant information are passed from lower level up to the highest level.
Information aggregation and friction occurs at all levels of an organization.
Thus, at the top level where the decision is made, the available information
may be corrupted or incomplete. Finally, even when the first two difficulties
disappear, the task of determining the optimal strategy can easily beyond our
reach. Many fundamental limitations well known in the discipline of
operations research and simulation will prevent us from solving
computationally even any moderate size general optimal decision problem in
practice. Consequently, in any centralized decision making approach, the
decision maker must by necessity employ a "one size fit all" approximate and
simplified methodology since for reasons mentioned above s/he is not
possible to know or to utilize detailed local information to tailor his/her
decisions. The result is the familiar phenomena of "一刀切" with its attendant
disadvantages even when the central DM has the best of intentions.
Examples of this are numerous in Chinese science and technology scene. I
don't have to repeat them here.
On the other hand, in western democracy, the emphasis is always to make
decisions at the lowest level practical and possible in order to avoid the
difficulties discussed above. But other difficulties arise. First if we act locally
based only on local information we are not "think globally but act locally". The
well known butterfly effect must be taken into account. We need to
coordinate local decision with its possible global consequences. But this is
easily said than done. How to think globally and act locally utilizing all the
available information is just as difficult as solving the original centralized
decision problem if not harder. The way out of this conundrum is to attempt
"divide and conquer" or solve the problem hierarchically. In the US, the
constitution is the supreme guideline. But it is only guideline in general terms.
It spells out principles but not detail decisions. You rely on lower level to
interpret these general principles and act according to local conditions. Thus,
city can make decisions so long as they do not violate state laws, state can
enact laws so long as they are not in conflict with federal law which
themselves must be judged to be constitutional and hence valid. And the
constitution is a living document subject to interpretations. This solution
approach is by no means perfect but it worked fairly well in the last two
hundred plus years for the US. The theoretical issue for decentralized control
and decision making is to understand what kind of societal structure and
necessary properties makes decentralized decision making possible and
optimal. Furthermore and more importantly, why.
But centralized decision making is not without its merits. Historically, China has
always yearned for a benevolent dictator 仁君. There is no substitute for the
ideal of an all wise and all knowing God, 玉皇大帝. But leaders are only
human. At times they may approach the ideal. I remember in 1999, there was
a large world congress in my field being held for the first time in Beijing. The
site was the International Conference Center on the fourth ring road. At the
conclusion of meetings on the third day of the conference all 2000 some
delegates in 50 buses will need to go to a banquet at the People Hall of
Congress at Tiananmen Square during rush hour. This will be an impossible
task in New York City. But in Beijing, a order came down from the top that due
to importance of the congress and as a courtesy to all the foreign participants,
using police escorts and a rolling blockage of major intersections along the
way, all 50 buses were able to drive nonstop from the fourth ring road
conference site to the people?ˉs congress hall with minimal disruption and
inconvenience to the populace. Every western participant was immensely
impressed by the efficiency. You would not even think about doing
something like this in America. You will be taken to the court and dispute and
lawsuit will last forever.
To sum it up, decentralization, democracy, and government are unsolved
problems in political and mathematical sciences. From Arrow's impossibility
theorem on democracy, to the NP=?=P problem of computer science, to the
optimal control problem for complex systems, they will occupy theoreticians
and practitioners for years to come. This makes employment for many of us
possible and a life of research interesting.
https://blog.sciencenet.cn/blog-1565-7555.html
上一篇:
On red tapes in the administration of scientific grants/contracts下一篇:
Harvard Celebrates the launching of the School of Engineering and Applied Scienc