DRAFT UNDER CONSTRUCTION
The IT industry is quickly moving in the 21st century to representing, and doing, business in digital form. An interesting question to ask is "What happens when all business is digital?" There are several interesting aspects to such a question. One revealed by the question of what is "business"? Increasingly, personal business is increasingly complex.
Because of the increase in connectivity afforded by the Internet (web/email protocols), large companies are increasingly outsourcing their functions and individual productivity is increasing - meaning that individuals performing business for large companies are performing increasingly complex tasks, often with other companies, just as we perform our personal business. The upshot of this is that we should rethink what "B2B" means: it means individuals managing complex enterprise tasks, whether for personal business, for consulting, non-profits, small business, or as an employee of large businesses. Our goal is to characterize and support, in fundamental ways, this enterprise management at all scales.
Another trend that affects this is the increasing agreement on what these digital representations of business mean. The meaning of representations, especially in IT, is increasingly referred to as semantics. It is estimated that between a half and three quarters of the $300 Billion spent on IT systems integration is spent on resolving semantics.[McComb] Just managing IT systems is motivation enough for adding meaningful common semantics to digitial representations of data, products, tasks and capabilities. In addition, agreement on common semantics has been developing for over a decade in the business world as a result of increasing requirements to perform business functions automatically. One might ask "What happens when computers understand all of the data? " Because this is the trend.
Much of what has transpired in business semantics has been the conversion of semantics into common syntax. So, for example, the representation of years is now commonly represented as four digits, which are understood by everyone and every system to represent the date in the thousands. This is not explicitly tied to any understanding of the calendar system, nor, so far, has there been any need to do so. At the next level of interoperability, companies are begining to come to at least US agreement on physical address semantics. An older and more complex example are the standards developed by Rosettanet organization for supply chains. But in both cases, interoperability has been achieved through the use of common syntax together with the common use of certain XML tags. Though some might say that industry does not need "semantics", it's already in play, since the common use for apllication interoperability of certain words is semantics.
Clearly the business world is evolving semantics, on a case- by-case basis, using the syntactic methodology, rather than appeal to any deeper representation of the information, such as calendars or geography. However, the development of web services has highlighted the requirement for additional semantics. Companies are increasingly developing Service-Oriented Architectures (SOAs) that support "eco-systems" of service providers, as well as customers. These services are varied enough that it is difficult to develop interoperability on a case-by-case basis. This decreases the potential of a truly dynamic web service-oriented business value network. In addition, semantics, as a general functional approach, has the potential to facilitate not only software integration, but enterprise management in general.
DERI Stanford takes the perspective that a formal concept of what an enterprise must be developed. Attempting to connect the various components of enterprises on an ad-hoc basis, especially as functions are outsourced, will not scale. This means that the job of managing an enterprise, for all of the individuals concerned, will eventually reach a level of complexity beyond human ability.
DERI Stanford's fundamental technology is formal models and computational logic. We view the current intersections of trends as data meets logic, based upon previous experience in data integration, using the Stanford-developed Infomaster[Genesereth et al], in business. (This system was used to solve practical problems in on-line catalogs for commercial e-marketplaces in the late '90s.) We believe that by attacking the enerprise management problem with a set of broad projects, the help of industry, and the assistance of our research partner DERI, we can develop not only new technologies for enterprise mangement, but synthesize a formal model of enterprises and their interactions that will guide future development.
Current DERI Stanford projects can be characterized as four developing technologies:
The Open Workflow Management (OWM) project seeks to develop a formalism that is an encoding of the constraints on the behaviour of distributed agents pursuing individual goals. One of our objectives is to move away from rigid process descriptions that are not likely to well-represent complex extended enterprises. Our model of workflow is open in that there is a degree of freedom in not only how individual objectives are to be accomplished, but also in the process definition at execution time. In particular, we are interested in the formal basis for the management of open workflows, where work is distributed among multiple enterprises and the flow of work is not completely defined prior to work execution.
All of these technologies deal with fundamental models of reasoning over data with semantics. The other projects are address other components of formalizing the management of extended enterprises. The Logical Spreadsheets model, for example, focuses both on the fundamental problem of reasoning with inconsistency, and with a model for allowing individuals to easily satisfy business requirements and policies.
It should be noted that DERI Stanford is not committed to any one particular technology, because what is in fashion today may be obsolete shortly. In particular, we do not follow what is currently called "The Semantic Web", sharing the view that this W3C-led initiative, though good work, is likely not to adapted as practical software by industry. And while we are likewise not committed to exploring only DERI International's Web Service Modeling Ontology (WSMO), but what is valuable about their approach is that 1) they are committed to developing something practical for industry and 2) this technology is far from finished, so that industry can participate in its creation. To summarize this, we say that although we are not using The Semantic Web, we are committed to semantic web technology.
Managing such complex enterprise tasks involves integrating not only data, product descriptions, tasks, advertised capabilities, and processes, but also business and government regulations and policies. That is, it is important not to just match up function providers and consumers, and to orchestrate the resulting processes, but also to do so in such a way that the policies of the consumers and providers are all respected. This will mean the resolution of conflicts and inconsistencies, and will require deep semantic integration of policy and regulation representations.
Computatonal Law is the "flagship" project investigating OWM applications, especially with respect to business policies. An important aspect of extended enteroperise management is that when web services are used in a dynamic fashion, the policies of all of the service providers and consumers should be respected.
Specific applications may include Digital Rights Management, Business Process Mangement, Heath Insurance Assistance, Legal Assistance and Ecommerce terms and conditions resolution. In those cases, the problem can be largely reduced to "simply" the merging of policies and regulations. To give an example from the future of ecommerce, suppose a suppler, B-C, can give a 10% discount if the buyer, A-B, will commit to buy 10,000 pieces over one year and pay within 30 days viaW-F bank electronically.
Suppose the buyer, A-B, using the web service integration search and plan technology of research Category I, is willing to do all of the above but does not have an account at W-F. Furthermore, the buyer has a business policy that any bank used must offer at least .5% interest on any account maintained. W-F offers such an interest rate on an average balance of $25K. Then it must be determined whether the A-B treasurer is willing to place such funds on deposit at W-F for the duration of the transaction with B-C. The A-B treasurer policies determine that this is possible if the transaction last no more than three months and ends before the end of March. All of this must be determined "on the fly" in order to take advantage of the opportunity that new supplier B-C is offering.
This rather convoluted example is unfortunately representative of business transactions and is why businesses depend largely upon relatively inflexible contracts hammered out in advance by legions of expensive lawyers. Such dynamic, autonmatic reasoning, is beyond the capability of current industial IT technology, as it cannot be programmed in advance. But it is possible with the computational logic capabilities being researched by DERI Stanford.
There are three related applications or our technologies we are currently pusuing:
Technically, web services are a breakthrough in that they represent the first time that industry has embraced a machine-readable representation of remote services. In theory, this objective declaration means that connections to services do not have to programmed ahead of time but rather service descriptions can be read at execution time, allowing dynamic consumption of services as needed. The academic community has a large body of techniques built-up over the last 20 years that can be applied to this new development. The results have the potential to be world-changing.
The business case for web services is fairly well-understood, at least for the near-term. We expect that as businessn services become more interoperable, industries and applications will develop that are not yet imagined. One less-well understood objective of businesses should be to increase peer adativity. What we mean by this is that businesses, and people, are eveloving from the client/server and provider/consumer models to peer-to-peer models. Moreover, each peer needs to be increasingly flexible as to which other peers it can deal, and with what transactions.
A novel and winning technical approach is to "leapfrog" over today's business processes approaches that depend upon static process description graphs. Static approaches require service providers to be able to access the process description, which is maintained by the process execution engine provider, and change the web service code to match the process. Or intermediary translators must be written by someone to match the service operations to the process. Similarly, business processes must be matched to each other with this approach based on static processes. We propose instead to provide Adaptive Services And Processes (ASAP) resulting in an advance over the current Service Oriented Architectures. This new target system might be denoted SOA++.
There is a near-term important step in this direction that we intend to develop: Business Service Registries (BSRs). A BSR would differ from UDDI in at least two primary aspects.
The emerging technology of web services facilitates integration maintenance because changes in applications, internal or external, can be hidden behind well-defined and published interfaces. What distinguishes web services from previous software component technologies, and what offers greater future potential, is that the application interfaces can be published in a standard machine-readable format, XML, available on the WWW.
However, the existing potential is limited because XML only defines the syntax of the input and output messages of the services. Without conventions and standards for the semantics of the defined interfaces, programming is still required, and the web services end up being little better than previous Remote Procedure Call technologies.
Fortunately, the near-term prospects for overcoming this problem are very good because of work in semantic web technology, discussed above. With semantics, there is an opportunity for providers of "middleware services" upon which other services can leverage.
DERI Stanford, in partnership with industry and DERI International, will facilitate the interoperability standards for business services. This will include the consensus specifications for fundamental business services that provide a foundation on which other services can build and leverage. These services, in turn, will be founded upon common semantics and business taxonomies. It is necessary to facilitate the commodization of information that will minimize the "friction" of business service exchange. This in turn will allow all companies to increase their global visibility via business service publishing.
DERI Stanford has already looked at the representational requirements for and the potential for establishing a BSR.
The BSR necessarily depends upon dynamic web service integration. This topic may not seem at first sufficiently challenging. Many academic and industrial research groups are working on the goal of allowing web services to be discovered and consumed at execution time, even if such services were not previously anticipated. However, most of these efforts are based on incrementally improving the finite state machine-based approaches, such as BPEL, to make the programming of such service integration-based processes less onerous. Only a few academics have even suggested autononous service integration using planning techniques developed over the last few decades in academia.
DERI Stanford takes the notion of autonomy seriously and is investigating the very hard problem of extending web service representations and finding the right algorithms that will efficiently produce service integration plans that achieve the desired results. The objective here is to allow the user to define high-level goals of effects that ought to occur in the world and then essentially say "make it so".
This is in contrast to the idea of a service directory that would allow one to find a single service to meet current requirements. With this approach, one can define a high-level approach, such as "make this part appear at my door in 3 days", and the system would define a sequence of services that would achieve the goal.
Linear plans that produce such an effect are possible with a standard technique called Situational Calculus. It may be possible to produce plans that incorporate desired preferences with an advanced technique called "Residuation" currently being developed by the Logic Group. Additionaly, we are reserching the use of Redux, developed in the Stanford Mechanical Engineering Department for Concurrent Engineering, as a way of responding to contingencies in service execution, (one source of nonmonotonicity in reasoning) as well as encoding user preferences and parallelism. This research has so far proved non-trivial but promising.
The potential here is to create processes "on the fly" that are able to semi-autonomously respond to new opportunities and contingencies. One example is the bank home mortgage application in which a variety of services are required depending upon current conditions, such as home repair. Were this to be required, the user would like to discover the best repair services for the home conditions and to accept bids for the job. Once the job was completed to the satisfaction of an inspector (another service), the mortgage process would continue on a more standard process involving title insurance and escrow. Standard, less dynamic, service integration techniques based upon state graphs cannot accomodate such flexibility. As with most Internet-based applications, once such flexibility is introduced, un-anticipated applications, and business opportunities, will likely emerge.
We will assume that the problem of semantics is largely solved by our research partner DERI, but we will contribute as well, using the data integration technology of Infomaster and, as much as possible, using practical taxonomies such as "Universal Business Language (UBL) ", a Sun initiative. We do not have to wait, however, for the consensus solutions to the semantics issue before we can develop very advanced solutions to the larger issues of service discovery and composition.
A very advanced but important functionality is service integration planning[Allen et al], [McDermott], [McIlraith&Son]. Suppose a US bank is managing the mortgage process (a web service application we are likely to see soon). It may be that there is a goal of insuring the house for $200K, which may not be possible, but a smart planning system will realize the goal can be decomposed into two subgoals of insuring for $100K each. Compensating transactions are insufficient for such cases.
There is a set of AI planning technologies that can address such problems, including replanning due to contingencies, such as a failure of a supplier to ship on time. Achieving explicit goals is a fundamental part of planning technology, such as in XSRL[Papazoglou et al]. And services will need to advertise pre- and post-conditions in order to use such planning techniques. This is particularly important when services have actions, which affect what next service is needed, including perhaps undoing the action just taken. For instance, a package that arrives too late may have to be returned.
As mentioned in [Petrie-Bussler], if descriptions of pre-conditions and effects were added to WSDL, then static process descriptions, such as BEPEL4WS, would not be necessary: standard AI planning technology can generate sequences of web service calls dynamically given such information. However, there is much that can be done with today's WSDL descriptions. Even before we have pre-conditions and effects for individual web service operations (which are necessary in the long run), the computational logic approach of Infomaster can be used integrate web services in complex and novel ways.
A more sophisticated approach enabled by computational logic and planning is to use knowledge to combine sets of services to achieve the desired result.
DERI Stanford will develop prototype services as well but these are not intended to compete with those of DERI Stanford partners, but rather to support them. We will also develop a research partnership with DERI for joint technical developments. A good example of a first step is the Business Service Registry (BSR).
Stanford will develop a prototype but real and practical BSR that will be tested in the Stanford University administrative environment. DERI will eventually support a core BSR that provides a reference architecture for other BSRs, which cooperate in a federated manner to supply a wide choice of specialized vertical domains, levels of trust, and other features from which BSR users, as well as other BSR providers, can choose.
Other elements of a SOA infrastructure will include those services that provide service-level agreements, secure payment, authentication, document translation, reliable message delivery, and all of the other fundamental devices necessary for the provisioning and consumption of real business services. This SOA infrastructure provides the bottom layer of a service grid on top of which successive layers of business services are built, consuming each other as needed. But the bottom layer provides the commerce "tone" to which all business services can connect in order to be connected to each other throughout the global Internet.
DERI Stanford participants will pay $250,000 in order to become a full research partner with full rights to all of the technologies developed by the consortium. There will also be a council composed of full research partners, which will evaluate the research and develop a consensus on interoperable standards. Technical partners may contribute in-kind technologies and receive rights and privileges based upon a determination of the value of the technologies contributed.
DERI Stanford will initially be a Stanford initiative, working with DERI as a partner. Stanford will have the responsibility to work as a scientific collaborator with DERI and all industrial members, exploring technical options and building prototypes. For example, even though we may build upon the emerging Universal Business Language (UBL) to provide semantics, we will want to take advantage of technologies emerging from the global Semantic Web research, and its follow-on programs. Furthermore, even with basic semantics defined, there will remain the problem of developing a sufficiently expressive but simple machine-readable language to describe business service pre-conditions and effects, such as, the fact that a service can only be used between certain hours and the delivery will occur a certain time for now. Stanford will explore with industry the best way to make these representations practical and efficient.
In order to make the maximum use of the supporting funds, they will be considered to be university gifts. This means that there will be no contracts with deliverables and no restrictions on the Intellectual Property: standard Stanford IP rules will apply. However, the university research intent is to develop a practical set of technologies desired by the sponsors that will translate into commercial services, meeting the requirements that have been set out in the BSR workshops to date.
At some point, we will transfer the initiative to DERI with Stanford remaining as a lead technical advisor. In this role, DERI will contract with Stanford, and possibly other universities such as CMU Stanford, to develop advanced technologies or propose solutions to outstanding hard problems, while DERI develops "hardened" services and solutions for its industrial partners. The IP ownership of these services will be proprietary under contract. We propose an agreement with DERI such that the DERI Stanford sponsors receive free licenses for the commercial services developed there.
Summary
A real Business Service Registry (BSR) does not yet exist. More important, there is no Theory of Extended Enterprise Management that would allow instant keretsus, virtual interprises, dynamic value networks, and host of other possibilities given an eco-system of services that could be dynamically discovered and consumed with a BSR. One is issue that multiple business and government policies need to be merged in a consistent and manner, and the resulting dynamic distributed processes need to managed. There is no underlying theory for doing so now, as there was not for databases prior to the invention of the relational calculus.
This is an opportunity. It represents an important economic niche that will be filled sooner or later, and probably sooner, by a small set of trusted and technically advanced global companies. There will be a federation of BSRs that compete to some extent, similar to "Yellow Pages" in the US, with a similar subscription-based business model. However, like the Underwriters Laboratories in the US, which also has a fee-based business model, these BSRs will provide some level of trust in the services provided. While a federation of peer-to-peer-based federation of BSRs is preferable and technically possible, it is quite likely that some small set of large and trusted BSRs will win out in the long run, analogous to the success of Dun and Bradstreet, Amazon, and Google.
Our approach to the ASAP/SOS++ goal is first to use AI planning techniques in order to dynamically compose web services, and also respond to contingencies and dynamically replan when plans fail. Processes emerge from such dynamic compositions, rather than being constructed and requiring the participation of the web service providers. The novel BSR functionality is that although it can return a single web service or a set of web services that acomplish a simple result, it can also take a desired state of the world and return a planned sequence of web services that would change the world accordingly. Further, we use Computational Law to merge the business policies and processes to ensure the policies of all of the service consumers and providers are respected. Finally, we will develop formalisms for managing the resulting dynamic and open workflows, resulting in a theory of extended enterprise management.
Citations
[Allen et al] Allen, Hendler, and Tate, eds., Readings in Planning, Morgan Kaufmann, 1990.
[Genesereth et al] M. Genesereth, A. Keller, and O. Duschka, "Infomaster: An Information Integration System", Proc. 1997 ACM SIGMOD Conference, May 1997.
[McComb] Business Semantics, David McComb, Morgan Kaufmann 10 September, 2003 ~ ISBN: 1558609172
[McDermott] D. McDermott "Estimated-Regression Planning for Interactions with Web Services", Proceedings of the AI Planning Systems Conference (AIPS'02), June 2002.
[McIlraith&Son]
S. McIlraithand T. Son,
"Adapting Golog for Composition of
Semantic Web Services,"
Proceedings of the Eighth International Conference on
Knowledge Representation and Reasoning (KR2002),
Toulouse, France,
April, 2002.
See also
http://www.daml.org/services/mci-son-kr02.ps
[Papazoglou et al] M. Papazoglou, M. Aiello, M. Pistore, and J. Yang, "XSRL: An XML Web-Services Request Language", Technical Report #DIT-02-0079, University of Trento, Povo, Italy, 2002. See also http://www.ebpml.org/xsrl.zip
[Petrie-Bussler]
C. Petrie and C. Bussler,
"Service Agents and Virtual Enterprises,"
Internet Computing, 7(4) July/August 2003, pp. 2-12.