Monday, March 14, 2005

Boil the Oceans,Goedel's Theorem and the Medical Knowledge-application Gap

Will Rogers, when asked about the solution to the German U-boat problem during WW I said "Boil the oceans.That'll force all the subs to the surface" When asked, How, he dismissed the issue as "mere details".
Much has been made of the explosion of medical information and the lack of application of same to the trench-dwelling docs.Fontelo and Ackerman posting on the 9/21/04 issue of Medscape in an article entitled "Evidence Based Medicine needs easy-to-use handheld tools for universal application" propose a solution.
A clinician would ask a question ( Dx, Rx, prognosis, whatever) on his PDA and a remote computer would take over. The literature would be reviewed,software would scan the articles,apply algorithms and "rules of EBM",evaluate it all and provide a conclusion back to the doctor.When I read this I thought are we really close to having an "expert system" that would replace the experience and expertise of human expert in various medical disciplines and be able to solve problems in general? ( I realize there are various "limited" (specific application) expert systems in existence but the authors seems to be talking about a general problem solver.) Could we ever have such a thing? It turns out that there is considerable difference of opinion regarding the second question.
In 1928 David Hilbert challenged the mathematical world to devise a procedure to devise the truth or falsity of every arithmetic statement.In 1931, Kurt Goedel demonstrated that this could not be done. A simplified version of Goedel's theorem is "arithmetic is not completely formalizable". The work of Turing and Chaitin enlarged this thought to " There is no systematic way of deciding if a given proposition is decidable or not".
Goedel showed that a system as relatively simple as arithmetic could not be axiomatized or made into a set of rules or algorithms to prove all of the truth within the system.
There is a persuasive body of thought that extrapolates from that. Roger Penrose has said human insight cannot be reduced to an algorithm. William Barrett ( Author of the Illusion of Technique) has argued that if the rules for a relatively simple system such as arithmetic are not adequate to determine the truth of some true statements, how much more inadequate must be rules be for a more complex system. He says we cannot escape the contingencies of life by an all encompassing algorithmic system, (such as a system that could do all the proposed off site computer would do for the inquiring physician's questions.)Maybe the findings of Goedel et al do not preclude what the authors propose and maybe I don't understand what they hope for.But it seems like what they write about leads us to a consideration of what IT can do, ie artificial intelligent systems(AI) or expert systems and we then run head on into the arcane debate over strong and weak AI, an area where I have no credentials.
Already computers have revoluntionized medicine and there is more to come.PDAs are great,having the PDR and drug interactions, and antibiotics guidelines and the ATP risk equations really have made my work easier and better and who can practice without ready access to Google.Wireless systems in hosptials are already giving staff real time access to lab data etc.But the day when you can simply ask your PDA what to do and a computer program will do what the Cochrane group does now (by using computers and humans) and much more is not here yet and some of us believe that even in theory it will never be.It is a quantum leap from computers regurgitating pre-digested guidelines and searching for articles to "reading" articles, analyzing them and answering questions in real time.Knowledgeable, intelligent savy humans are currently needed for that job and they would be not be able to provide the instant answers that they authors imply would be part of their scheme.Will they someday be smart enough to design machines that can replace them?

No comments: