Persistent context

October 11, 2007

If there are two people from the tech world whose blogs I would choose to read if I was given a choice of two, they might just be Jon Udell and Jeff Jonas, and they will forever be joined in my mind, cause I found Jeff via Jon. Unfortunately, I have never had a chance to meet Jon Udell, but I have had the chance to talk to Jeff.

Previously I talked about the data path. I would like to extend the thought on persistent context a little further. I have long felt that one of the flaws in most software design is a lack of context. If I am in one view where certain actions are not possible, why do I need to know about 10 things that don’t make sense. Similarly with data, maintaining context is critical, since the data only make sense in the appropriate context. “Andy is naked” and “Andy is naked in the middle of Sibera in December” are not equivalent statements. The first one just tells you that Andy forgot to wear his clothes when he woke up in the morning. The second one tells you that Andy is going to die very soon.

Tasteless humor aside, it is somewhat simple to put data at point A in a certain context, and data at point E in a certain context, but what if in going from A –> B –> C –> D –> E, the context gets lost. How is the context at point A related to the context at point E? Those to me are some of the fundamental reasons why the concept of persistent context is so critical. Now think of a pharmaceutical company, distributed globally. The company has many departments, some working closely together, others not so much, even if they are part of the same drug development pipeline. Does someone in the oncology franchise know what the person in the diabetes franchise is up to? Perhaps they can search some internal database and find work done previously by someone in a different department which might be relevant, but do they know the context that the data represents? What if some new information is found by a third department. How does that change the context?

Jeff’s words are resonate mightily with me.

Treat queries like data to avoid asking every question every day

My interpretation of this, especially in a life science context is that the relationships across studies and experiments need to be captures and it is absolutely essential to capture this meta-data. In a post that I is lying in my drafts folder (I will update when I am done with it), I argue that the semantic web probably means more for the “life science” web than for the World Wide Web. If you are storing relationships, i.e. your queries, and treating them as pieces of data, you are essentially capturing relationships, and the semantic web provides an elegant framework to do so.

Another principle of Jeff’s work (note that his work is on fraud, so the language might not be completely topical).

Enterprise awareness is computationally most efficient when performed at the moment the observation is perceived

Bringing it into the context of the globally distributed pharma company; we need to have systems in place when the mining is not done after the fact, but the information is flagged when data or a new query is added to the system. This goes back to the original post of data following the data. We shouldn’t have to find an adverse event when someone goes hunting, in a classic federated approach. We should know as the data are collected and meta-analyses are performed.

Will such intelligent systems make their way into the biosciences? I see that happening at some point, but until they are given enough value from the powers that be, and believe me they are not, adoption is going to be slow. I have seen companies like ProSanos do some very creative work for adverse event detection. But those are the exceptions rather than the rule.

In future posts, I will look into some of these matters more deeply, as my understanding of some of these concepts gets further enhanced.

Further reading
Introducing Perpetual Analytics
Federated discovery vs. persistent context

Picture courtesy Esther Dyson via a Creative Commons license

Technorati Tags: Jeff Jonas, Perpetual Analytics, Persistent Context