Computing Paradigm Not a Branch of Science
Total Page:16
File Type:pdf, Size:1020Kb
letters to the editor DOI:10.1145/1721654.1721657 Computing Paradigm Not a Branch of Science EGINNING WITH THE headline, generically called “computer,” but is key element in all complex computing “Computing’s Paradigm,” a misleading connection, since the domains. Moreover, understanding is The Profession of IT View- two disciplines describe the computer associated with modeling, a key aspect point by Peter J. Denning in different ways—a formal model of of understanding. One cannot deter- and Peter A. Freeman (Dec. computation in Knuth-Dijkstra com- mine whether a system can be built or B2009) reflected some confusion with puting, an actual machine in Denning- represented without the understanding respect to Thomas Kuhn’s notion of Freeman computing. needed to pose hypotheses, theses, or “paradigm” (a set of social and insti- Denning and Freeman proposed formal requirements. Understanding is tutional norms that regulate “normal a “framework” that takes the side of often not addressed very well by begin- science” over a period of time). Para- engineering computing (why I call it ning computing researchers and devel- digms, said Kuhn, are incommensu- Denning-Freeman computing), de- opers, especially as it pertains to infor- rable but determined by the social scribing development of an engineer- mation processes. discourse of the environment in which ing system and leaving no doubt as to The second missing element of con- science develops. the envisioned nature of the discipline. ceptualization is an explicit statement The crux of the matter seems to be All the purportedly different fields they about bounded rationality, per Her- that computing can’t be viewed as a proposed—from robotics to informa- bert Simon (http://en.wikipedia.org/ branch of science since it doesn’t deal tion processing in DNA—are actually wiki/Bounded_rationality), a concept with nature but with an artifact, namely different applications of the same para- based on the fact that the rationality of the computer. For guidance, we reflect digm. To consider them different would individuals is limited by the informa- on at least one scientific antecedent— be like saying quantum physics is differ- tion they possess, the cognitive limi- thermodynamics, which originated ent for nuclear plants and for semicon- tations of their minds, and the finite from the need to understand the steam ductors. The physics is the same; what amount of time they have to make de- engine but is distinguished from steam changes is the engineering process of cisions. Bounded rationality addresses engineering by its search for general its application, as in computing. the tentative nature of design and dis- principles, detached from a specific The abstract problem of symbol covery as an evolving set of decisions machine. The Carnot cycle and entropy manipulation is mathematical and the posed against multiple criteria derived theorem are scientific results, not feats subject of computing science. The in- from understanding and initiation. of engineering. stantiation of the symbol-manipulation The results from conceptualization, or The metatheoretical problem of com- model in useful systems is a problem design, must always be understood as puting seems mainly semiotic. Suppose, for the engineering of computing, a both tentative and knowledge-limited. 200 years ago, somebody had created a discipline that is theoretically, method- Finally, a phase missing from evalu- discipline called, say, Thermozap, that ologically, and conceptually separated ation and action is “technology readi- included the study of the Carnot cycle from the mathematical study of symbol ness” (http://en.wikipedia.org/wiki/ and the building of new steam engines. manipulation. Technology_readiness_level), especially Somebody might have come up with the Simone Santini, Madrid, Spain in deploying real systems. A new tech- insoluble problem of whether the new nology, when first invented or concep- discipline was science or engineering. It I wish to suggest ways to improve Peter tualized is not suitable for immediate was neither but rather a hodgepodge of J. Denning’s and Peter A. Freeman’s application. It is instead usually subject things better left separated. proposed computing paradigm in their to experimentation, refinement, and in- Computing is in a similar situation. Viewpoint “Computing’s Paradigm” creasingly realistic contextual testing. There is an area (call it Knuth-Dijkstra (Dec. 2009). While I accept the tenta- When proven, it can be incorporated computing) that studies scientific tive five phases—initiation, conceptu- into a deployed system or subsystem. All problems posed by the existence of alization, realization, evaluation, and information processes are realized and computing devices. Thermodynamics action—in the proposed paradigm, embedded within the context of existing was part of physics because steam en- they are, in practice, incomplete. deployed systems. Therefore, technol- gines use physical forces. Computing While I agree with initiation (the ex- ogy readiness of a posed information devices are formal machines, so Knuth- istential argument followed by concep- process must stand as a separate phase Dijkstra computing is a mathemati- tualization) as the design argument, between evaluation and action. cal discipline. Then there is the com- three additional phases are missing: 1. Simon, H. Bounded Rationality puting discipline that builds systems The first is a phase 0 I call understand- and Organizational Learning. Organiza- (call it Denning-Freeman computing), ing (or problem understanding). Before tion Science 2, 1 (1991), 125–134. which is definitely part of engineering. one can pose the existential (Denning’s 2. Simon, H. A mechanism for social The error is in thinking they are the and Freeman’s initiation), a phase must selection and successful altruism. Sci- same. Both refer to the same device, address (problem) understanding, a ence 250, 4988 (1990), 1665–1668. 6 COMMUNICATIONS OF THE acM | APRIL 2010 | VOL. 53 | NO. 4 letters to the editor 3. Simon, H. A behavioral model of phase of a MapReduce computation paradigm in testing emphasizes com- rational choice. In Models of Man, So- is essentially a filter and a group-by prehension over memorization. cial and Rational: Mathematical Essays operation in SQL, while the Reduce phase Bela Erdelyi, Lincroft, NJ on Rational Human Behavior in a Social is largely a target-list computation in Setting. John Wiley & Sons, Inc., New SQL. When user-defined functions are York, 1957. included in SQL (as they are in many How to Honor the Heroes of CS David C. Rine, Fairfax, VA commercial implementations), the Communications cover article “Amir functionality provided by parallel SQL Pnueli Ahead of His Time” (Jan. 2010) DBMSs and MapReduce implementations mourned the passing of Amin Pnueli in Authors’ Response: appears to be the same. November 2009. Likewise, Communica- Our argument concerned computing’s “belief The parallel DBMS literature, dating tions mourned (Nov. 2008), along with system.” Kuhn discussed belief systems from the 1980s, includes hundreds of the rest of the computer science com- in science. Whether or not we were true to articles on implementation tactics. munity, the disappearance and passing Kuhn is irrelevant to our argument. Our comment about “standing on the of Jim Gray. Tragic as these events are, Santini says computing is about shoulders…” was meant to suggest that any they are sure to be followed by others, as computers. We disagree. Computing is about new implementation effort should carefully computer science is no longer in its in- information processes, and computers review the prior literature to learn what fancy but well past middle age. I see the are machines that implement information past results are available, then add to the risk that Communications covers (and ar- processes. There are natural, as well as store of total knowledge. ticles) could turn into a gallery of the re- artificial, information processes. Computing The MapReduce team seemed not vered heroes of our science who will be is as much about computers as astronomy is to have done this exercise. Hence the passing away in ever greater numbers. about telescopes. comment. Communications could instead honor its Computing does not separate neatly Michael Stonebraker, Daniel Abadi, icons by, perhaps, adding an obituary into math and engineering, as Santini David J. DeWitt, Sam Madden, column, even as a permanent feature. claims. Computing increasingly employs Erik Paulson, Andrew Pavlo, Panos Louridas, Athens, Greece experimental (scientific) methods to test Alexander Rasin, Cambridge, MA hypotheses about complex information processes. Editor’s Response: Santini’s desire to parse computing into Even in the Classroom, a Communications does indeed publish separate elements will fail, just as all such Click Is Just a Click obituaries to note the passing of prominent previous attempts have failed. Our collective The news item “Web Used for Final computer scientists. In certain cases, concern with information processes keeps Exams in Denmark” (Jan. 2010) gave however, the Editorial Board deems the pulling all the elements together, no matter the impression that such an approach event to be deserving of further recognition. how hard we try to separate them. was never tried before. I have taught Jim Gray was in full vigor when he Peter Denning, Monterey, CA computer- and network-security-re- disappeared without a trace in January Peter Freeman, Atlanta, GA lated classes for the past eight years, 2007, as was Amir Pnueli when he passed incorporating the Internet as a tool away in November 2009. In both cases students use during class, including there was a sense of unusual or unexpected Hold the Accusations That on quizzes and exams. I am sure I am tragedy, which explains the degree of Limit Scientific Innovation not the only instructor in the U.S. al- coverage in Communications. I applaud the debate on MapReduce lowing students to use the Internet for between “MapReduce and Parallel research and comprehension in the Corrections DBMSs: Friends or Foes?” by Michael classroom. In the article “Amir Pnueli: Ahead of Stonebraker et al.