Skip to main content

Digital Experience

IBM Think: The Future of Watson Conversation

IBM Think 2018 speakers Brian Loveys and Vivian Lee outlined the upcoming roadmap for Watson Conversation. It’s all part of the Watson Conversation Service, which you can embed in a wide range of sites and apps.

Highlights from the session: 


Have had Watson Conversation and Watson Virtual agent. They are being merged into one offering: Watson Assistant.

No need to worry about migration. It will be OK.

Watson assistant is:

  • A natural language assistant on a machine
  • Has machine learning capabilities
  • Use Cases
    • Customer Service (by far the #1 use case)
    • HR Bots
    • Gaming and virtual reality for command and control
      • recently announced an SDK with Unity

What’s New

Have a new home page with lots of orientation videos, intro to Assistant, and login. Will be opening up the beta in April. Can request access in your workspaces after login.

A lot of the new features are mostly built around making it easier to manage the intent, dialog, and administration. This makes sense when you think about it. Setting up your conversations will allow you to have a successful conversation assistant. They better they do at this, the more the AI / chatbot will do the job.

Note: Don’t request beta for your production instance.

  1. Digression: dynamically answer questions outside of the normal flo
    1. Includes handoff to live agents
  2. Prebuilt content: Common bot conrol. They have a wide range of customer care and industry
  3. Dialog Folders: Stay organized as you scale your dialog
  4. Set context in the dialog UI: no need to do it in json ediitor
  5. User Analytics: Continue to bolster the capabilities. Can now track by user id
  6. Dialog Tracing: great feature for trouble shooting
  7. Search: available in intents and entities
  8. Integrate with IBM Cloud functions: can call cloud functions directly from dialog
  9. Separate log files: separate workspaces from your log file, allowing you to improve a bot while in production.

What’s Next / Direction

There are two key challenges:

  1. How do I reduce the effort to build and maintain my virtual assistants?
  2. How do I build an engaging Assistant experience?

That leads to three investment areas:

  1. Modularize and scale your Assistant
  2. Enhance Build Experience
  3. Improve Recommendations

Scaling Your Assistant

This is about scaling your assistant with skills. An assistant is a composition of skills. (Think Alexa and skills). Users can converse over a variety of channels like Facebook, slack, ServiceNow, etc.

A skill is essentially a reusable component that can answer questions of a certain type

Each conversation has a kind which represent reasoning technology being used to power the skill.

Demo of how this might work

  • This is accessed via the beta.
  • Homepage changes. Couple new tabs now include assistants and skills.
    • Learning material still on the home page
  • Assistants
    • Can edit or create assistants. This will become your workspace
    • He edited the stock bot to see the normal conversation admin UI
    • It would be different with conversation vs discovery skills
    • Created a new Assistant
      • First question: add a skill
      • Added US Open
      • Added facebook as a channel
        • with credentials
      • Will allow pre and post callouts to customize this
      • Session and state management come out of the box

Enhanced Build Experience

  • “Disambiguation” think of did you mean. This allows the AI to learn.
  • Enable user experiences beyond simple text
    • add buttons, images, videos, pause delays
    • Support the channel (Facebook vs Slack for example)
  • Allow for open ended entities
    • Classic scenario: can’t find the entity you want to extract
    • Remind me to … (open ended and use the context to extract the entity)
  • Improved DevOps
    • Versioning via snapshot
    • Github integration
    • Makes it easier to promote to dev, test, prod

Improve and Recommend

  • Augment current dashboard with additional metrics
    • Find conflicts in intents. (These look really similar. You should probably look at them.)
  • Recommendations of additional entity synonyms to consider
    • Suggest additional synonyms
  • Conflict detection identification and overlapping intents
    • Find those infinite loops you may create
  • Identification of problematic conversations

Bottom Line

IBM is investing in a number of core areas to improve the overall Watson Assistant.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Michael Porter

Mike Porter leads the Strategic Advisors team for Perficient. He has more than 21 years of experience helping organizations with technology and digital transformation, specifically around solving business problems related to CRM and data.

More from this Author

Follow Us