I often look to the EDS Fellows' Next Big Thing Blog for ideas about the future of technology. Although a bit more uneven and cursory than some of the quality blogs by mavens at Microsoft and IBM, the EDS Fellows can be counted on to be at or near the cutting edge of the IT world. A recent entry entitled EDS' Next Big Thing Blog : Models with Alternative Vocabularies is a great example, and one with tremendous practical implications for those of us in health IT and clinical research informatics.
The issue addressed in this post is the great technical challenge of enabling intelligent-appearing machine-machine interaction. I say "intelligent-appearing" because I don't personally believe machines need to possess true intelligence in order to do useful things. This allows us to beg the question of whether machines can be said to be truly intelligent. After all, if you want to maintain that Homo sapiens possesses true intelligence, the burden of proof is on you.
The barriers are not so much technical as they are semantic and cultural. Even within one natural language words and phrases have many shades of meaning, just as many different words and phrases can have the same meaning. The difficulties these ambiguities produce are manifold, as eloquently described in Furnas et al., The Vocabulary Problem in Human-System Communication: an Analysis and a Solution. Add in the complication of multiple natural languages, mix with cultural differences in interaction styles and "impedance mismatch" due to different levels of granularity and specificity, and stir in a shot each of "Not Invented Here" syndrome and turf warcraft, and you have a Julia-Child-worthy recipe for siloization.
The challenges are becoming greater as enterprises become global, integrate with global partners, and execute mergers and acquisitions. In addition to having different natural languages in different segments of the enterprise, there are legacies of diverse terminology used to describe shared concepts. We cannot expect to get everyone to use the same terms.
We need modeling tools that separate the vocabulary from the concepts. Such modeling tools would enable a shared model to be expressed in alternative terms that could be for different natural languages, or merely vocabularies of different communities using the same natural language.
I raised this issue at XML conferences back in 2000 and 2001, and suggested then that the only solution worth working toward is a unifying set of tools for federated semantics, tools that would eliminate the requirement that machines all speak the same language in order to talk to each other.
This approach has been incorporated in the Object Management Group (OMG) standard for Semantics of Business Vocabulary and Rules (SBVR). In addition, this specification provides for specification of concept semantics independent of the vocabulary by using the technologies of ontologies and Common Logic. The value of this approach is not yet appreciated in the marketplace.
It is great to know that the OMG is on the case here. To their detriment, they did produce CORBA, but then Ford produced the Edsel and Studebaker the Avanti, and while the Edsel was flop of truly massive proportions and the Avanti did only slightly better, they did portend technologies that did not reach the mainstream of the automotive world until years or in some cases a decade or more had passed. CORBA didn't quite make it in the marketplace - you had to know this was going to happen when Borland's farsighted execs hitched their star to it, as they had to the Pascal language in the '80's - but many great ideas in CORBA are now found in Web Services. Perhaps SVBR will do the same, manifesting in some as-yet-unforeseen protocol that will bridge heterogeneous semantic domains and natural languages seamlessly and without effort.
What does all this portend for health IT? First and foremost, that efforts to produce a Wurlitzer-like Mother-of-all-healthcare-vocabularies are doomed - efforts like HL7v3 and BRIDG, for example. We will learn a lot from them, but it is most unlikely they will solve the interoperability problem, because ultimately they attempt one-size-fits-all solutions.
There, I said it out loud. Now I better hope nobody hears.
What can we do within our own health IT domains? I belieive the most important task is to make sure we each have your own houses in order. Is your organization settled on a suite of standard vocabularies for representing all the various aspects of healthcare? Do you employ implementation-independent messaging protocols? Do all your business units share the job of selecting, implementing, and evangelizing standards, are are there silos that need to be toppled? Let's do what is within reach before we attempt the nearly-impossible.
Or maybe it's actually totally-impossible, in a fragmented world of providers and payers at all strata, and so many of our citizens without health insurance at all. Maybe we must first solve the political and cultural problems that result in young mothers never receiving prenatal care and adults going without diabetes and heart disease screening and counseling, before we tackle anything so abstract as an electronic medical record.
Just another (im)patient point of view...
Comments