Monday, 21 September 2015

Knowledge management in a digital workplace

Next session is about managing all of the knowledge and data coming out of smart workplaces.

Knowledge Management (KM) is a discipline to enable effective action through access to relevant intellectual assets, including those that are known, but not documented. It's not a technology. Its biggest challenge is going beyond information into knowledge.

Easy bit is capturing the facts, harder is the implicit stuff, stuff that's in people's heads. Contextual information, experience, expertise etc. Can't capture this, so have to connect people through collaboration systems and capture what makes it possible to act effectively in given situations. Very difficult to capture what everyone knows using conventional methods.

Is it time to automate knowledge extraction? Use AI techniques. Machine learning, linguistic analysis.

For example, fraud detection. Conventional way is to get characteristics of a fraudulent transaction from a human expert and code them up manually in policies, rules or in an automated system. Or, get a machine to look at the history of all relevant transactions. Let it discover common patters and automatically build an automated model to detect fraudulent transactions.

Or Helpdesk. Train agents to deal with problems, talk to each other, capture information in a knowledge base. Or, example from a SAP Helpdesk. take a machine learning system and give it data, software manuals, previous queries, also activity monitoring systems, real time data, data from public sources. Built a service that the user could interact with. Learns form all queries that the system can't deal with and gets passed to an agent, can answer question next time round.

Already using this sort of learning eg in Spam filters.

Will become more prevalent. The digital workplace will increase our ability to observe work. What is being done, by whom, with whom, where,how, at what time, in what context, resulting in what. This will all be observable, recorded and analysed. Will result in either decision support systems, automation,

Example of Boston hospital. All patient records and data in this sort of system. Intended to improve diagnosis, treatment. Decisions still doctors, but this acts a smart assistant.

Use data from previous outcomes to have more data driven approach to support decision making.

Technical challenges:

  • Access to all relevant data
  • Ability to distinguish between positive or negative outcome
  • Existence of historical patters
  • Lack of skills in machine learning
  • Computing resource challenges
  • Still need lot of manual intervention and tuning
  • Spurious correlations, not always a casual relationship

Another big issue - the creepiness factor! Big brother factor. Watching people at work to see if you can automate their job. Have to think about how these techniques are applied.

Expect that by 2017 virtual personal assistants will collect data from applications you are using, will contextualise this. Give you advice and recommendations.

By 2020 will be dominan specific knowledge extraction and reuse systems. Eg HR systems will have built into them analytical tools. Will predict turnover etc.

Need to look at our business activities where we have a lot of data and start to think about automating the extraction of knowledge. Look at areas where we can increase our data collection. Architect, procure and deploy new applications aiming to create feedback loops between automated knowledge extraction and decision support or automation. Talk to our vendors to see where they are going in this area. Identify high priority areas with minimal technical challenges.

Need to track industry responses to cultural, legal, security, compliance and ethical questions.

Interesting session, and some thinking to be done about areas where we might implement some of these ideas. After all, we do have a lot of data! How can we turn this into knowledge?

No comments: