Here is an interesting idea that has been well articulated in a new book, The Learning Layer by Steven Flinn.  I recently had a chance to speak with Steve about this marriage of aspects of Web 2.0 and artificial intelligence (aka adaptive systems) that can have useful applications within the enterprise.

Steve is the CEO of ManyWorlds, a firm that conducts R&D in the area of next generation systems and business processes and provides practical applications of this work to organizations. He was an executive at Royal Dutch Shell where he held  a variety of positions including Chief Information Officer and Vice President of Strategy and Strategic Alliances. Steve has a background in economics, mathematics and computer science.

Steve noted that use of personalized responses based on user behavior has been pervasive on the consumer Web through such things as Amazon’s recommendations.  However, this technology has been applied much less within the enterprise.  He feels that this is ironic since behavioral information available within the enterprise  can generally be much richer than out on the Web. You have a more clearly defined set of users and many more interactions to data mine, along with more related teams who collaborate and generate more behavioral data.  I would certainly agree with the difference in the quality of information on user behavior and also add that many Web 2.0 applications such as wikis work better within the trusted environment of the enterprise. This seems to be another case.

The Learning Layer approach takes this personalization several steps further. Not only are personalized recommendations provided to individual users based on their behavior and the behavior of others, but the system feeds these recommendations back to itself to continuously adapt on an automated basis. Steve said that the technology is currently available to do this, it just needs to be properly applied.

For example, a system managing content might make recommendations for related content based on a user’s profile and actions.  Using the Learning Layer approach, it would also keep track of all user behavior and feed this back into the system on a regular basis. The relationship between two sets of content may become stronger or weaker depending how it is currently being used.  The same logic can be applied to the connections between people to see the ebb and flow of connections.

The approach can be applied to work flow and here it gets even more interesting in my opinion. Just as old school knowledge management created more direct business value when aligned to business processes, I see the same thing happening here.  Let’s take the example of a property casualty insurance underwriter. After the system takes in enough actions to be able to differentiate the skill level of users, it is ready to go.

Now if an underwriter with no experience in underwriting laundry mats, for example, starts to work on one the system recognizes this. It also knows the steps that an inexperienced underwriter should take when working with laundry mats and provides these process steps. It can also recommend a person who is slightly more advanced than the user who can offer guidance. If the user has middle level experience, then the process steps can be tailored to that level. In the meanwhile the system is observing the ongoing user behavior on an aggregated basis and making adjustments in the proper process steps for everyone at all levels.

The technology is around to create this type of system. I can see the value and wish I had this capability when I designed knowledge management systems for underwriters in the early 90s.  Call centers that deal with complex topics would be another great target area. You need to have enough complexity to warrant this type of intervention and then enough users to generate useful data for the system to apply.

We also discussed the concept of learning value that Steve raises in the book. He took the concept of value of information from decision analysis and applied it to learning. In decision analysis people calculate the value of having certain information to help with decisions. The same concept can be applied to learning.  When undertaking an activity there is the direct value and the value of the learning derived from the undertaking. This often translates into the amount of uncertainty that can be eliminated by the new knowledge and its effect on actions. Steve noted that learning only has real value if it changes behavior (i.e., decisions).  If people will still do the same thing regardless then nothing is gained. That sounds simple but it is often overlooked.

I like this approach. I think it does extend the possibilities of enterprise 2.0. If we can create data rich environments through the transparent interactions within enterprise 2.0 then we have expanded the learning opportunities. Then if we can use this expanded learning to better guide individual behavior we have taken it a notch further. Now if we can turn this learning back on the system to auto-generate changes within the system itself, we have taken things another step further.  I think the data gained from the transparency of enterprise 2.0 is a large piece of the value. Here is an approach to make better use of this transparency.