A few random thoughts to finish the week.
In between bad movies on the plane I was thinking about how content management needs to evolve. The irony well described in the The Innovator’s Dilemma is that the very innovations we used to drive change and create our business become mired in habit and dogma. We use past success as a means of justifying a lack of creativity and call it “experience” and risk aversion. I hate that.
The most interesting development that affects this evolution is in the area of cognitive systems. You may have seen this article from Stephen Hawking where he predicts that the economic impact will be no less than disaster for the middle class as computers begin to assume tasks formerly in the knowledge worker domain.
This made me wonder, how is this change going to start? What exactly are some of the incremental movements that might directly affect the ECM business in the near term. What does that mean for those of us comfortably seated in the middle class.
The technical developments are already happening in various ways but for the purposes of a conversational post I did not want to scour the internet for examples. Here are a couple of my favorite ideas for discussion.
AI Assisted Authoring
Who doesn’t love spell check when writing a blog post. Who doesn’t hate autocorrect when texting. Small examples of how the computer tries to help humans better communicate with other humans. These are fairly limited in scope though. Recently I was going over a demo for LEAP Concert – a new collaborative editing app. I won’t go into details but one feature is that you can assign sections of a document to other team members and one of the scenarios called for asking the colleague to insert a relevant quote.
It occurred to me – why does that collaborative author have to be a person at all? There are many scenarios where AI will soon, if not already, be able to analyze the document, understand the authoring task and assemble and insert the appropriate text for review by the coordinator. Imagine contract assembly and review completely carried out by a virtual attorney. Skynet & HAL , Attorneys at Law. That’s not scary at all, is it?
OCR Exception Handling
I hate OCR. Not because it isn’t useful and important to the business. I hate it because for years the idea has been creating unrealistic expectations in the buying public. Vendor marketing just fuels the fire. Recognition rates in the nineties are thrown around and people have a “Hollywood” view of its accuracy. OCR never claims to be perfect. What marketing doesn’t tell you is that the errors that do occur are never concentrated. If you have a million documents whatever errors you do have are spread across thousands. Five percent error doesn’t mean five percent of the documents need to be reviewed. It means that out of the total number of errors they could be ANY WHERE. Sure we tweak confidence levels, mix in different engines and take other steps but a process to handle exceptions will always be needed.
I believe we are at the limits of what the current algorithms and strategies can do. Some vendors are calling their recognition AI but I am suspect. New labels on existing techniques. I would hope to see more advanced AI not limited to Natural Language Processing applied not to the initial recognition but rather to the error handling workflow. Our own brains don’t actually see everything. Most of your field of vision right now is a construct of inferences made by your brain on recent inputs,experience and context. What a human QA does when they process index errors is use their experience and the context of the error to inform choices in making a correction. It is this sort of muli-contextual analysis that AI’s will soon if not already excel at,making it possible to correct for gaps in OCR/ICR processes.
It would make more sense from a computational perspective to apply AI to the exception handling rather than using this mechanism for everything. This implies that at some point, a hierarchy of computing develops where rudimentary processing is not necessarily replaced by the higher order. The AI doesn’t replace the entire system. Only the human.
What now?
I am obviously not an expert in this field but it seems to me the value of the knowledge worker has traditionally been rooted in decision making. At some point though it will be possible to automate making all decisions that can be directly derived from data. What is left for us? Inference and assumption. Creativity and imagination. Quite frankly things that the corporate world is better at suppressing rather than fostering. We need to reimagine how work gets done when inference and insight are more valuable than data and trends.
In the new economy data may be the new currency but meaning is what we buy with it.
Lee Dallas
Marko Sillanpää
You make some great points. I’d like to point out that automated authoring already occurs today.
Many of the Contract Life Cycle Management (CLM) vendors use clause libraries with their contract drafting. Tasks that had been performed by a paralegal to ensure all of the appropriate clause are in a contract are now done by the CLM system, often this occurs directly from the ERP tool. Some solutions will even have these clauses create new obligation items that will track the contract performance.
In these scenarios a sales person will say create my client contract and based on the Salesforce data and rules in the CLM, the first thing anyone in the legal team sees is a fully drafted contract. There are even cases where the legal team doesn’t get involved until the first pass of that contract has been returned from the customer.
From an ECM platform perspective, we would say that this is all possible if we customize compound document assemblies. For contract management this is happening in products today.