Designing Systems for Health-Care Provider-Patient Partnerships – “Building systems smart enough to work with us” – 2017 Harvard Business School Symposium: Engineering & Entrepreneurship: Innovation in Healthcare

Barbara J. Grosz, Higgins Professor of Natural Sciences, SEAS

When Barbara started working in AI, she was asked whether she was creating intelligence for the CIA!

Today’s systems include vacuum robots, recommender systems are ubiquitous, and financial systems – and they are beginning to enter healthcare.

There is a great deal of enthusiasm and, by contrast, speculation and fear. Barbara feels we have to build systems that augment human intelligence, not replace it – making good team members.

Barbara asked what question might Turing ask now? Barbara thinks a good test would be whether we can build a system that would be a good team member – and specifically she wants to talk about Engineering and systems design.

If you don’t think about the system working as part of a team with the human being, you can lead to silly design problems – eg asking Siri for gas stations, getting a list of 16, then asking it which ones are open – and Siri offering to google “which ones are open”! This leads to several ethical problems.

Teamwork is not simply the sum of individual plans

If you’re going to design a collaborative system, you need to build teamwork in from the start

There are various SharedPlans Theories about teamwork.

Barbara studies a paediatric team looking after patients with complex problems – 15 team members and lots of data flow. A very challenging team to manage!  And there is nobody in charge, and nobody knows everything that is going on – so if you’re going to build a system that joins this team, you need a very different system.

There are various ways of using this SharedPlans theory to intervene here.

  • Consensus on recipe
    • support for providers establishing agreement on high-level approach, establishing mutual belief
  • Recipes may be partial and evolve over time
    • build systems that support these dynamically evolving plans
  • Team members commit to performance of group activity and to each other’s success
    • support communication and coordination at appropriate levels and times – who needs what information and when
  • Locality of information about delegated tasks
    • information sharing without information overload

New Algorithms for Information Sharing


  1. Loose-coupling – detailed recipes are not available for computing the value of information
  2. Extended duration – allows learning about interaction among tea members to use as a signal of needing to know.

Barbara’s team developed a Mutual Influence Potential Network to learn collaboration patterns over time to work out who might need to know things. So the system was reasoning about who might have information that someone needed – the MIP-DOI algorithm.

Did a test using this algorithm in a collaborative document environment – the personalised way produced better documents.

Ethics and design for AI-Technology working with people

Collaboration has to be in the design from the start – as does ethics.

Design with ethics in mind. What are the ethical consequences of the systems we are building, especially in healthcare?

Barbara’s concluded by describing how at Harvard they are trying to bring ethics in earlier in processes and courses in computer science.

“A computer system should make us feel smarter, not dumber and work seamlessly with us, like a human partner.” – Grosz





Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s