8/2/2019 AI Project

1/11

Medicine:-

Medicine is the art and science of healing. It encompasses a range of health care practicesevolved to maintain and restore health by the prevention and treatment of illness.

Contemporary medicine applies health science, biomedical research, and medical technology todiagnose and treat injury and disease, typically through medication, surgery, or some other formof therapy. The word medicine is derived from the Latin ars medicina, meaning the art of healing .

Though medical technology and clinical expertise are pivotal to contemporary medicine,successful face-to-face relief of actual suffering continues to require the application of ordinaryhuman feeling and compassion, known in English as bedsidemanner.

Artificial Intelligence and Medicine

Artificial intelligence in medicine (AIM) has reached a periodof adolescence in which interactions with the outside world arenot only natural but mandatory. Although the basic researchtopics in AIM may be those of artificial intelligence, theapplied issues touch more generally on the broad field of medical informatics. To the extent that AIM research is drivenby performance goals for biomedicine, AIM is simply one component within a wide range of research and development activities. Furthermore, an adequate appraisal of AIM researchrequires an understanding of the research motivations, the complexity of the problems, and asuitable definition of the criteria for judging the field's success. Effective fielding of AIM

systems will be dependent on the development of integrated environments for communicationand computing that allow merging of knowledge-based tools with other patient data-managementand information-retrieval applications. The creation of this kind of infrastructure will requirevision and resources from leaders who realize that the practice of medicine is inherently aninformation-management task and that biomedicine must make the same kind of coordinatedcommitment to computing technologies as have other segments of our society in which theimportance of information management is well understood. "

Man strives to augment his abilities by building tools. From the invention of the club to lengthenhis reach and strengthen his blow to the refinement of the electron microscope to sharpen hisvision, tools have extended his ability to sense and to manipulate the world about him. Today we

stand on the threshold of new technical developments which will augment man's reasoning, thecomputer and the programming methods being devised for it are the new tools to effect thischange.

Medicine is a field in which such help is critically needed. Our increasing expectations of thehighest quality health care and the rapid growth of ever more detailed medical knowledge leavethe physician without adequate time to devote to each case and struggling to keep up with thenewest developments in his field. For lack of time, most medical decisions must be based on

8/2/2019 AI Project

2/11

rapid judgments of the case relying on the physician's unaided memory. Only in rare situationscan a literature search or other extended investigation be undertaken to assure the doctor (and thepatient) that the latest knowledge is brought to bear on any particular case. Continued trainingand recertification procedures encourage the physician to keep more of the relevant informationconstantly in mind, but fundamental limitations of human memory and recall coupled with the

growth of knowledge assure that most of what is known cannot be known by most individuals.This is the opportunity for new computer tools: to help organize, store, and retrieve appropriatemedical knowledge needed by the practitioner in dealing with each difficult case, and to suggestappropriate diagnostic, prognostic and therapeutic decisions and decision making techniques.

In a 1970 review article, Schwartz speaks of the possibility that the computer as an intellectualtool can reshape the present system of health care, fundamentally alter the role of the physician,and profoundly change the nature of medical manpower recruitment and medical education--inshort, the possibility that the health-care system by the year 2000 will be basically different fromwhat it is today.

The key technical developments leading to this reshaping will almost certainly involveexploitation of the computer as an 'intellectual,' 'deductive' instrument--a consultant that is builtinto the very structure of the medical-care system and that augments or replaces many traditionalactivities of the physician. Indeed, it seems probable that in the not too distant future thephysician and the computer will engage in frequent dialogue, the computer continuously takingnote of history, physical findings, laboratory data, and the like, alerting the physician to the mostprobable diagnoses and suggesting the appropriate, safest course of action.

This vision is only slowly coming to reality. The techniques needed to implement computerprograms to achieve these goals are still elusive, and many other factors influence theacceptability of the programs.

What is "Artificial intelligence in Medicine?"

Artificial Intelligence is the study of ideas which enable computers to do the things that makepeople seem intelligent ... The central goals of Artificial Intelligence are to make computersmore useful and to understand the principles which make intelligence possible

This is a rather straightforward definition, but it embodies certain assumptions about the idea of intelligence and the relationship between human reasoning and computation which are, in somecircles, quite controversial. The coupling of the study of how to make computers useful with thestudy of the principles which underlie human intelligence clearly implies that the researcherexpects the two to be related. Indeed, in the newly-developing field of cognitive science,computer models of thought are explicitly used to describe human capabilities.

From the very earliest moments in the modern history of the computer, scientists have dreamedof creating an 'electronic brain'. Of all the modern technological quests, this search to createartificially intelligent (AI) computer systems has been one of the most ambitious and, notsurprisingly, controversial.

8/2/2019 AI Project

3/11

It also seems that very early on, scientists and doctors alike were captivated by the potential sucha technology might have in medicine (e.g. Ledley and Lusted, 1959). With intelligent computersable to store and process vast stores of knowledge, the hope was that they would become perfect'doctors in a box', assisting or surpassing clinicians with tasks like diagnosis.

With such motivations, a small but talented community of computer scientists and healthcareprofessionals set about shaping a research program for a new discipline called ArtificialIntelligence in Medicine (AIM). These researchers had a bold vision of the way AIM wouldrevolutionize medicine, and push forward the frontiers of technology.

AI in medicine at that time was a largely US-based research community. Work originated out of a number of campuses, including MIT-Tufts, Pittsburgh, Stanford and Rutgers (e.g. Szolovits,1982; Clancy and Shortliffe, 1984; Miller, 1988). The field attracted many of the best computerscientists and, by any measure, their output in the first decade of the field remains a remarkableachievement.

In reviewing this new field in 1984, Clancy and Shortliffe provided the following definition:

'Medical artificial intelligence is primarily concerned with the construction of AI programs thatperform diagnosis and make therapy recommendations. Unlike medical applications based onother programming methods, such as purely statistical and probabilistic methods, medical AIprograms are based on symbolic models of disease entities and their relationship to patientfactors and clinical manifestations.'

Much has changed since then, and today this definition would be considered narrow in scope andvision. Today, the importance of diagnosis as a task requiring computer support in routineclinical situations receives much less emphasis (J. Durinck, E. Coiera, R. Baud, et al., "The Role

of Knowledge Based Systems in Clinical Practice," in: eds Barahona and Christenen, Knowledgeand Decisions in Health Telematics - The Next Decade, IOS Press, Amsterdam, pp. 199- 203,1994), So, despite the focus of much early research on understanding and supporting the clinicalencounter, expert systems today are more likely to be found used in clinical laboratories andeducational settings, for clinical surveillance, or in data-rich areas like the intensive care setting.For its day, however, the vision captured in this definition of AIM was revolutionary.

After the first euphoria surrounding the promise of artificially intelligent diagnostic programs,the last decade has seen increasing disillusion amongst many with the potential for such systems.Yet, while there certainly have been ongoing challenges in developing such systems, theyactually have proven their reliability and accuracy on repeated occasions (Shortliffe, 1987).

Much of the difficulty has been the poor way in which they have fitted into clinical practice,either solving problems that were not perceived to be an issue, or imposing changes in the wayclinicians worked. What is now being realized is that when they fill an appropriately role,intelligent programs does indeed offer significant benefits. One of the most important tasks nowfacing developers of AI-based systems is to characterize accurately those aspects of medicalpractice that are best suited to the introduction of artificial intelligence systems.

8/2/2019 AI Project

4/11

Is it Possible for Computing Machines to Think?

No--if one defines thinking as an activity peculiarly and exclusively

human. Any such behavior in machines, therefore, would have to becalled thinking-like behavior.

No--if one postulates that there is something in the essence of thinking which is inscrutable, mysterious, and mystical.

Yes--if one admits that the question is to be answered by experiment and observation, comparingthe behavior of the computer with that behavior of human beings to which the term "thinking" isgenerally applied.

AI can support both the creation and the use of

medical knowledge

Human cognition is a complex set of phenomena, and AIsystems can relate to it in two very different ways.Proponents of so-called 'strong' AI are interested increating computer systems whose behaviour is at somelevel indistinguishable from humans (see Box 1). Successin strong AI would result in computer minds that mightreside in autonomous physical beings like robots, orperhaps live in 'virtual' worlds like the information spacecreated by something like the Internet.

An alternative approach to strong AI is to look at human cognition and decide how it can besupported in complex or difficult situations. For example, a fighter pilot may need the help of intelligent systems to assist in flying an aircraft that is too complex for a human to operate ontheir own. These 'weak' AI systems are not intended to have an independent existence, but are aform of 'cognitive prosthesis' that supports a human in a variety of tasks.

AIM systems are by and large intended to support healthcare workers in the normal course of their duties, assisting with tasks that rely on the manipulation of data and knowledge. An AIsystem could be running within an electronic medical record system, for example, and alert aclinician when it detects a contraindication to a planned treatment. It could also alert the clinician

when it detected patterns in clinical data that suggested significant changes in a patient'scondition.

Along with tasks that require reasoning with medical knowledge, AI systems also have a verydifferent role to play in the process of scientific research. In particular, AI systems have thecapacity to learn, leading to the discovery of new phenomena and the creation of medicalknowledge. For example, a computer system can be used to analyse large amounts of data,looking for complex patterns within it that suggest previously unexpected associations. Equally,

8/2/2019 AI Project

5/11

with enough of a model of existing medical knowledge, an AI system can be used to show how anew set of experimental observations conflict with the existing theories. We shall now examinesuch capabilities in more detail.

Reasoning with medical knowledge

Expert or knowledge-based systems are the commonest type of AIM system in routine clinicaluse. They contain medical knowledge, usually about a very specifically defined task, and areable to reason with data from individual patients to come up with reasoned conclusions.Although there are many variations, the knowledge within an expert system is typicallyrepresented in the form of a set of rules.

There are many different types of clinical task to whichexpert systems can be applied.

Generating alerts and reminders.

In so-called real-time situations, an expert system attachedto a monitor can warn of changes in a patient's condition. Inless acute circumstances, it might scan laboratory testresults or drug orders and send reminders or warningsthrough an e-mail system.

Diagnostic assistance. When a patient's case is complex, rare or the person making the diagnosisis simply inexperienced, an expert system can help come up with likely diagnoses based onpatient data.

Therapy critiquing and planning. Systems can either look for inconsistencies, errors andomissions in an existing treatment plan, or can be used to formulate a treatment based upon apatient's specific condition and accepted treatment guidelines.

Agents for information retrieval. Software 'agents' can be sent to search for and retrieveinformation, for example on the Internet that is considered relevant to a particular problem. Theagent contains knowledge about its user's preferences and needs, and may also need to havemedical knowledge to be able to assess the importance and utility of what it finds.

Image recognition and interpretation.

Many medical images can now be automaticallyinterpreted, from plane X-rays through to morecomplex images like angiograms, CT and MRIscans. This is of value in mass-screenings, forexample, when the system can flag potentiallyabnormal images for detailed human attention.

8/2/2019 AI Project

6/11

There are numerous reasons why more expert systems are not in routine use. Some require theexistence of an electronic medical record system to supply their data, and most institutions andpractices do not yet have all their working data available electronically. Others suffer from poorhuman interface design and so do not get used even if they are of benefit.

Much of the reluctance to use systems simply arose because expert systems did not fit naturallyinto the process of care, and as a result using them they required additional effort from alreadybusy individuals. It is also true, but perhaps dangerous, to ascribe some of the reluctance to useearly systems upon the technophobia or computer illiteracy of healthcare workers. If a system isperceived by those using it to be beneficial, then it will be used. If not, independent of its truevalue, it will probably be rejected.

Happily, there are today very many systems that have made it into clinical use. Many of these aresmall, but nevertheless make positive contributions to care.

Diagnostic and educational systems

In the first decade of AIM, most research systems were developed to assist clinicians in theprocess of diagnosis, typically with the intention that it would be used during a clinical encounterwith a patient. Most of these early systems did not develop further than the research laboratory,partly because they did not gain sufficient support from clinicians to permit their routineintroduction.

It is clear that some of the psychological basis for developing this type of support is nowconsidered less compelling, given that situation assessment seems to be a bigger issue thandiagnostic formulation. Some of these systems have continued to develop, however, and havetransformed in part into educational systems.

DXplain is an example of one of these clinical decision support systems, developed at theMassachusetts General Hospital (Barnett et al., 1987). It is used to assist in the process of diagnosis, taking a set of clinical findings including signs, symptoms, laboratory data and thenproduces a ranked list of diagnoses. It provides justification for each of differential diagnosis,and suggests further investigations. The system contains a data base of crude probabilities forover 4,500 clinical manifestations that are associated with over 2,000 different diseases.

DXplain is in routine use at a number of hospitals and medical schools, mostly for clinicaleducation purposes, but is also available for clinical consultation. It also has a role as anelectronic medical textbook. It is able to provide a description of over 2,000 different diseases,

emphasizing the signs and symptoms that occur in each disease and provides recent referencesappropriate for each specific disease.

Decision support systems need not be 'stand alone' but can be deeply integrated into an electronicmedical record system. Indeed, such integration reduces the barriers to using such a system, bycrafting them more closely into clinical working processes, rather than expecting workers tocreate new processes to use them.

http://www.openclinical.org/aisinpractice.htmlhttp://www.openclinical.org/aisp_dxplain.htmlhttp://www.openclinical.org/aisp_dxplain.htmlhttp://www.openclinical.org/aisp_dxplain.htmlhttp://www.openclinical.org/aisinpractice.html

8/2/2019 AI Project

7/11

The HELP system is an example of this type of knowledge-based hospital information system,which began operation in 1980 (Kuperman et al., 1990; Kuperman et al., 1991). It not onlysupports the routine applications of a hospital information system (HIS) including managementof admissions and discharges and order entry, but also provides a decision support function. Thedecision support system has been actively incorporated into the functions of the routine HIS

applications. Decision support provides clinicians with alerts and reminders, data interpretationand patient diagnosis facilities, patient management suggestions and clinical protocols.Activation of the decision support is provided within the applications but can also be triggeredautomatically as clinical data is entered into the patient's computerized medical record.

Expert laboratory information systems

One of the most successful areas in which expert systems are applied is in the clinical laboratory.Practitioners may be unaware that while the printed report they receive from a laboratory waschecked by a pathologist, the whole report may now have been generated by a computer systemthat has automatically interpreted the test results. Examples of such systems include the

following.

The PUFF system for automatic interpretation of pulmonary function tests has been soldin its commercial form to hundreds of sites world-wide (Snow et al., 1988). PUFF wentinto production at Pacific Presbyterian Medical Centre in San Francisco in 1977, makingit one of the very earliest medical expert systems in use. Many thousands of cases later, itis still in routine use.

GermWatcher checks for hospital-acquired (nosocomial) infections, which represent asignificant cause of prolonged inpatient days and additional hospital charges (Kahn etal.,1993). Microbiology culture data from the hospital's laboratory system are monitoredby GermWatcher, using a rule-base containing a combination of national criteria and

local hospital infection control policy. A more general example of this type of system is PEIRS (Pathology Expert InterpretativeReporting System) (Edwards et al., 1993). During it period of operation, PEIRSinterpreted about 80-100 reports a day with a diagnostic accuracy of about 95%. Itaccounted for about which 20% of all the reports generated by the hospital's ChemicalPathology Department. PEIRS reported on thyroid function tests, arterial blood gases,urine and plasma catecholamines, hCG (human chorionic gonadotrophin) and AFP (alphafetoprotein), glucose tolerance tests, cortisol, gastrin, cholinesterase phenotypes andparathyroid hormone related peptide (PTH-RP).

Laboratory expert systems usually do not intrude into clinical practice. Rather, they are

embedded within the process of care, and with the exception of laboratory staff, cliniciansworking with patients do not need to interact with them. For the ordering clinician, the systemprints a report with a diagnostic hypothesis for consideration, but does not remove responsibilityfor information gathering, examination, assessment and treatment. For the pathologist, thesystem cuts down the workload of generating reports, without removing the need to check andcorrect reports.

Machine learning systems can create new medical knowledge

http://www.openclinical.org/aisp_help.htmlhttp://www.openclinical.org/aisp_puff.htmlhttp://www.openclinical.org/aisp_germwatcher.htmlhttp://www.openclinical.org/aisp_peirs.htmlhttp://www.openclinical.org/aisp_peirs.htmlhttp://www.openclinical.org/aisp_germwatcher.htmlhttp://www.openclinical.org/aisp_puff.htmlhttp://www.openclinical.org/aisp_help.html

8/2/2019 AI Project

8/11

Learning is seen to be the quintessential characteristic of an intelligent being. Consequently, oneof the driving ambitions of AI has been to develop computers that can learn from experience.The resulting developments in the AI sub-field of machine learning have resulted in a set of techniques which have the potential to alter the way in which knowledge is created.

All scientists are familiar with the statistical approach to data analysis. Given a particularhypothesis, statistical tests are applied to data to see if any relationships can be found betweendifferent parameters. Machine learning systems can go much further. They look at raw data andthen attempt to hypothesize relationships within the data, and newer learning systems are able toproduce quite complex characterizations of those relationships. In other words they attempt todiscover humanly understandable concepts.

Learning techniques include neural networks, but encompass a large variety of other methods aswell, each with their own particular characteristic benefits and difficulties. For example, somesystems are able to learn decision trees from examples taken from data (Quinlan, 1986). Thesetrees look much like the classification hierarchies discussed in Chapter 10, and can be used tohelp in diagnosis.

Medicine has formed a rich test-bed for machine learning experiments in the past, allowingscientists to develop complex and powerful learning systems. While there has been muchpractical use of expert systems in routine clinical settings, at present machine learning systemsstill seem to be used in a more experimental way. There are, however, many situations in whichthey can make a significant contribution.

Machine learning systems can be used to develop the knowledge bases used by expertsystems. Given a set of clinical cases that act as examples, a machine learning system canproduce a systematic description of those clinical features that uniquely characterise theclinical conditions. This knowledge can be expressed in the form of simple rules, or oftenas a decision tree. A classic example of this type of system is KARDIO, which wasdeveloped to interpret ECGs (Bratko et al., 1989).

This approach can be extended to explore poorly understood areas of medicine, andpeople now talk of the process of 'data mining' and of 'knowledge discovery' systems. Forexample, it is possible, using patient data, to automatically construct pathophysiologicalmodels that describe the functional relationships between the various measurements. Forexample, Hau and Coiera (1997) describe a learning system that takes real-time patientdata obtained during cardiac bypass surgery, and then creates models of normal andabnormal cardiac physiology. These models might be used to look for changes in apatient's condition if used at the time they are created. Alternatively, if used in a researchsetting, these models can serve as initial hypotheses that can drive furtherexperimentation.

One particularly exciting development has been the use of learning systems to discovernew drugs. The learning system is given examples of one or more drugs that weaklyexhibit a particular activity, and based upon a description of the chemical structure of those compounds, the learning system suggests which of the chemical attributes arenecessary for that pharmacological activity. Based upon the new characterization of chemical structure produced by the learning system, drug designers can try to design a

http://www.coiera.com/papers/HauCoiera.pdfhttp://www.coiera.com/papers/HauCoiera.pdf

8/2/2019 AI Project

9/11

new compound that has those characteristics. Currently, drug designers synthesis anumber of analogues of the drug they wish to improve upon, and experiment with theseto determine which exhibits the desired activity. By boot-strapping the process using themachine learning approach, the development of new drugs can be speeded up, and thecosts significantly reduced. At present statistical analyses of activity are used to assist

with analogue development, and machine learning techniques have been shown to at leastequal if not outperform them, as well as having the benefit of generating knowledge in aform that is more easily understood by chemists (King et al., 1992). Since such learningexperiments are still in their infancy, significant developments can be expected here inthe next few years.

Machine learning has a potential role to play in the development of clinical guidelines. Itis often the case that there are several alternate treatments for a given condition, withslightly different outcomes. It may not be clear however, what features of one particulartreatment method are responsible for the better results. If databases are kept of theoutcomes of competing treatments, then machine learning systems can be used to identifyfeatures that are responsible for different outcomes.

Problems and Technologies

Artificial intelligence is a broad collection of technologies and goals. Its researchers work both toextend their understanding of the ways in which intelligent systems can be constructed and toapply that knowledge in the real world. Medicine, on the other hand, is a much older enterpriseand is much clearer about its goals. Artificial intelligence in medicine is a hybrid field, formedout of the union of these two enterprises. United through AIM, these two communities caninteract in three quite distinct ways.

First, the relationship can be technology driven. Medicine can provide AI researchers with acomplex set of real-world problems with which to evolve their techniques. The outcome of goodtechnology-driven research is the development of general purpose technologies that can beapplied in many different domains.Those AIM researchers working in this way would judge their research a success if it werepublished in an AI journal. For example, much of the early work in computer-based diagnosisfocused on medicine because it was such a good testing ground.Today however, AI researchers working on advance diagnosis technologies have moved awayfrom medical problems, and are more likely to be working on diagnosing faults in digitalcircuits, computer networks, or photocopiers.Second, AIM can be problem driven. In this case, there are pressing medical problems needing

solutions, and AI competes with other alternatives to provide those solutions. Success here ismeasured by the ability to solve real-world problems, and one would expect a researcher to beable to publish this type of work in mainstream clinical journals. Success would be measured byreductions in patient morbidity or mortality or by improvements in the efficiency of health caredelivery. It should follow that otherwise technologically disinterested clinicians would beinterested in the results of this work because it would be clinically relevant to them.The third approach of AIM does not extend the boundaries of AI or solve the problems of medicine; instead, it is inward looking. Driven initially by the need to solve real medical

8/2/2019 AI Project

10/11

problems, one may nevertheless need to solve technical issues that are not familiar to clinicians.At best, these problems can appear esoteric to clinicians; at worst, they can seem irrelevant.Equally, these technical problems might be inspired by AI but not be of major interest to it.Indeed, some of the most difficult, work in AIM does belong right here.For example, clinical outcomes need to be measured to improve medical practice. Consequently,

ways of creating an electronic patient record need to be found so that clinical data can be pooledand analyzed.However, to achieve this, methods are needed to extract meaning from an often complex medicalrecord.Consequently, there is a large effort, both in medical informatics and in AIM, to developterminology coding schemes for electronic patient records. Separated by such a long-chain of reasoning from the original problem, this type of research usually appears in specialist journalsdevoted to medical informatics, and not routinely in medical or AI journals.It would be fair to say that the dominant mode of AIM (and perhaps to a lesser extent Informatics as a whole) over the last 10 years has been inthis third category. There are plenty of journals in which work is published for and by the

informatics community.However, both AIM and informatics at present have a minimal impact on the fields that gavethem birth. Have medical informatics and AI in medicine become so engrossed in their ownproblems that they have lost sight of the needs of medicine and AI? Or are they now validresearch fields that can stand on their own without justifying their work within these other fields?The answer has to be both. AIM and Informatics each has a unique set of technical problems tosolve. They also, however, must judge what their labors achieve against their impact on thewider world. While medicine will always be there for researchers in AI to test their technologies,AIM is first and foremost a sub discipline of medicine. Its achievements will thus ultimately onlyreach medical significance when they can demonstrate a positive impact on health care. The finalarbiters of success must rest in phrases like clinical outcomes and cost -effectiveness and notin measures li ke computational complexity.

Challenges

With an acceptance that AIM must adopt health cares problems as its own comes a redefinitionof goals. For example, it is now generally recognized that, beforeAIM research can produce systems that have a significant impact; a substantial informationinfrastructure will have to be in place. Consequently, some researchers have now moved fromAIM to assist with the enormous task of developing an informatics infrastructure for health care.At the same time that AIM is redefining its priorities, medicine itself is undergoing a quietrevolution known as evidence - based medicine, and this may well transform the nature of

clinical practice. The mountain of research produced by medicine each week is now so greatthat the time lag between a treatment being proven effective and actually coming into routinepractice is often measured in years. As a consequence, even the most diligent practitioners areunable to deliver care that represents the best available practice to their patients.The problem lies in the fact that the mechanisms for transferring evidence into clinical practiceare unable to keep up with the ever growing mountain of clinical trial data. For example, thefirst trial to show that streptokinase was useful in treating myocardial infarction was published in1959. Convincing evidence mounted in the early 197Os, and the first multi trial meta-analysis

8/2/2019 AI Project

11/11

proving the drugs value was published in the early 1980s. However, formal advice thatstreptokinase was useful in the routine treatment of myocardial infarction only appeared in thelate 1980s. This was a full 13 years after a close examination of the published literature wouldhave indicated the treatments value. Many people in medicine see a move to universally available and codified guidelines for clinical

practice as the solution to this problem. Not surprisingly, this poses both cultural and technicalproblems for health care. It should be enormously exciting to AIM researchers that medicinestechnical problems are ones that AI may be able to help solve.Developing large database of practice guidelines requires knowledge-based technologies tocreate and maintain them. Therefore, AIMs skills in knowledge acquisition and representationare needed to help develop methods that allow doctors to compare newly published data withexisting guidelines and to update the knowledge base as appropriate. It may also be desirable tocustomize these knowledge bases to reflect local needs and conditions in different regions andcountries. Work has already been underway in these areas for some time, but much remains to bedone.13-15. Ultimately, what is required is a way for practicing clinicians to access suchguidelines quickly, incorporate them into their clinical practices, and then submit their own

experiences back to the knowledge base to help improve it.This raises one final challenge whose impact has yet perhaps to be fully appreciated-the rise of the Internet.The Internet is important for two reasons. First, it seems to be custom made to solve some of theinherent communications problems that are at the heart of creating a truly evidence-basedmedical practice. Its second lesson is perhaps more subtle but may in the long run prove to be asimportant for AIM. Computer science and AI in particular, is based on the power of formalizingknowledge representation and reasoning. Artificial intelligence in medicine is focused ondeveloping ways of formalizing everything from diagnosis to the terminology medicine uses.Yet, the Internets growth has probably been possible only because it formalizes so little. Whenit comes to defining how one should publish on the Internet, less is definitely more. Is there anequivalent level of semi formality for protocols and clinical guidelines? If so, this would allowsome ability to manage and update guidelines while still making them flexible enough to use indifferent and sometimes unexpected ways.In summary, through the rise of evidence-based medicine and the enormous challenges it poses,AIM and medical informatics have a new and important role to play. It may well be that in 25years the question will not be, What has artificial intelligence in medicine ever achieved? butrather, How could medicine have ever got he re without it?