Oracle Digital Assistant Highlights from Oracle Cloud World 2022

November 3, 2022 | 11 minute read
Salman Sheikh
Director Product Management, Oracle Digital Assistant
Barry Hiern
Director, Product Management, Digital Assistant & AI
Text Size 100%:

Introducing the next generation of Digital Assistant

Disclaimer: This blog outlines upcoming features of the Oracle Digital Assistant Platform as well as our overall direction and is intended for informational purposes only.  It does not imply a guarantee of the functionality being delivered for a given release, and hence should not be used as the basis of a purchasing decision.

Conversational interfaces have come a long way since the initial chatbots were introduced in the mid 2010s. From single domain bots, driven by the end user and through their chat client of choice, they have evolved into a true Digital Assistants able to address the needs of users across multiple business applications and domains.    

While Digital Assistants have enhanced the ability for users to interact with an organisation, they are still dependent on the user being available to do so.  To help with this, the Oracle Digital Assistant team is proud to introduce the next stage in the evolution of the Digital Assistant. One that both simplifies the development of the Digital Assistant itself and enhances the user’s access to it. 

Simplified Conversation Design

The development of a successful conversational interface ultimately comes down to three distinct areas:

  • Understanding the users input and goals.
  • Constructing a dialog flow that is flexible enough to follow the meandering nature of human conversation.
  • Making sure that the user gets the correct answer, or at least one that is appropriate to their question.

Moving from intent to Intent-less NLP development

Building out the various intents to be covered by a Digital Assistant, along with the supporting AI training corpus can, for certain projects, be a challenging exercise.  This is particularly the case when the number of possible questions may be large, (for-example a company’s dynamic employee FAQ).  Alternatively, challenges may occur with a decision support application where the required questions (and hence intents) may not be fully understood prior to going live.  That is, the required intent is simply not known.

Auto generated Intents

To simplify the task of creating “Question” driven conversations, the Oracle Digital Assistant has introduced the ability to build out a dynamic FAQ, directly from your corporate Frequently Asked Questions documents.  This capability, known as “Knowledge Documents”, removes the need for the developer to proactively “think up” the intents.  Rather, the Digital Assistant can ingest your document at design time, extract the Q&A pairs, build an initial training corpus, and auto-generate the appropriate intents for the skill.  This capability greatly speeds up the development process of a conversational FAQ. 

Knowledge Documents example
Figure 1: Using Knowledge Documents to rapidly generate Q&A intents

Intent-less Q&A

While the auto-generation of intents greatly simplifies the creation of an FAQ style skill, it still requires an understanding of the nature of the questions that would be asked, around a given business domain.   Ideally the skill should be able to generalise over the available data and determine the most appropriate answer to the user’s question, without the need to first define the appropriate intents & slots required classify the input, and to respond appropriately.

The Oracle Digital Assistant leapfrogs the traditional ‘intent-entity-slotting’ NLP techniques through the introduction of a new state-of-the-art “Prompt-based” NLP model.  Rather than relying on a pre-defined set of intents and utterances, to represent every possible question against the available data, in this model, the NLP engine is instead given prompts (via meta data) as to the nature of the query output.   In this way, it is not necessary to identify all potential queries ahead of time.

The first example of this intent-less conversation model can be seen with the introduction of SQL-Dialogs in the Digital Assistant.

SQL-Dialogs allows users to speak to their databases in natural language, from which the NLP engine can generate the corresponding SQL query statement to access the data.

For example:

User Input “Tell me the average renumeration for office staff stationed in the Big Apple”
Output from SQL Dialogs NLP model
Copied to Clipboard
Error: Could not Copy
Copied to Clipboard
Error: Could not Copy
SELECT Avg(sal)
FROM   Emp T1, Dept T2
WHERE  T1.deptno = T2.deptno
AND    T2.loc    = ‘NEW YORK’
AND    T1.job    = ‘CLERK’

 

Here, in the case of accessing a database, the Database Schema acts as the metadata prompt indicating what the SQL to get the appropriate answer would be.  The developer did not need to supply multiple example utterances mapped to physical database queries.

As the SQL-Dialogs NLP engine works along with the intent-slot approach, the user will be able to switch back and forth between pre-defined, goal-based, interactions and those that are ad-hoc in nature.  All within the same chat channel. 

SQL Dialog example
Figure 2: Running SQL queries using natural language in multiple channels

 

The SQL-Dialogs feature is only the first example of this “next-gen” Natural Language Processing model to be available within the Digital Assistant.  Access to other data-stores and knowledge bases will be introduced soon, further expanding its ability understand the various ‘adhoc’ queries raised by the end user; all without the need to pre-define the intents.

Building The Right Conversation Flow.

The development of a robust set of Intents and Entities (the NLP model) is half the challenge when building a successful conversational interface.  That said, the ability to have a dialog flow that supports the user going off-topic (aka not following “the Happy Path”) is the other.

Through use of the “Visual Flow Designer” (VFD) the creation of complex conversation flows is greatly simplified by associating intents with fragments of a conversation (flows).  In this manner the requirements of specific goals or even “off topic” processing can be visualised as individual flows and called as required during the conversation. 

Visual Flow Designer example
Figure 3: Using Visual Flow Designer to model out dialog flows

 

To bring the two halves of conversation design together into a single seamless development task, the VFD has been extended to allow for:

  • The dialog flow to automatically reflect the nature of the underlying NLP model components, such as the use of a Composite Bag entity
  • The direct creation of elements of the NLP model from within the visual flow.
  • Easily access the contextual data of the conversation via the Apache FreeMarker expression builder
Visual Flow Designer - Composite Bag example
Figure 4: Composite bag example using Visual Flow Designer

But is it the right conversation for our users?

When designing a conversational AI, it is always important to ensure that the dialog flow, and even its persona is relevant for the users for whom it is targeted.  As such, there is inevitably a degree of “back & forth” between those developing the skill, and the business stakeholders.  To help ensure that the “right” conversation is built from the get-go, the Digital Assistant introduces the Dialog First & Skill Planning tool.

Based on an “User Said <–> Digital Assistant Replies” back & forth interaction, the Skill Planning tool allows for the creation of a functional wireframe mock-up of the conversation.  By validating the “by example” conversation, the business stakeholders and development team are able subsequently agree on the nature of the dialog flow and the path it should take.

Furthermore, via the use of the integrated NLP engine, the conversational nature of the wireframe can be mapped to specific dialog flows and intents, to generate out a first pass of a functional digital assistant. 

Visual Flow Designer - Skill Outline
Figure 5: Skill outline in Visual Flow Designer

The best Interface is no interface…

Along with the ability to expose complex applications in a simple intuitive way, the introduction of Digital Assistants has had a significant impact on the area of Customer Service.  Through deflection of repetitive requests to the chat interface, call centre wait times have been dramatically decreased, with agents freed up to spend additional time with more complex enquiries.  

However, due to the inability to access the chat client, or simply personal preference, there is always a large percentage of users who will still reach out to the ‘Help Desk’ via a more traditional approach.  That is, via a telephone and the call centre’s direct phone number.

To address this community Oracle Digital Assistant introduces the Contact Centre AI which brings the ability to have incoming voice calls answered automatically by the Digital Assistant via integration with your organisations IVR and telephony gateway.

The Contact Centre AI brings

  • Seamless integration to your IVR, such that calls are answered by the Digital Assistant.
  • Next Gen Speech recognition, optimised for the lower transmission rates of Telephone voice channels (8Khz)
  • Advanced “Human like” neural voice text to speech engine to respond to the caller (with a choice of voice persona).
  • The ability to design your Contact Center IVR flows via a visual paradigm using the Digital Assistant’s Visual Flow Designer.
The four pillars of Conversational IVR
Figure 6: The four pillars of Conversational IVR

 

By allowing the Digital Assistant to respond to inbound calls, customers can speak naturally instead of navigating complex IVR menus.  They will get help faster and can also resolve multiple issues in a single call.  Also, as with the text-based chat, the Digital Assistant will be able to transfer the call to a live agent if it is not able to address the needs of the caller.  The Digital Assistant will tackle the large volume of routine customer requests and will free up the human agent to focus on the more complex customer interactions.

Introducing the Digital Workforce

While the use of conversational AI has greatly simplified how users can perform business tasks, it still fundamentally is dependent on the user to be present and actively involved.  True automation however needs to enable users to hand-off their work and responsibilities, particularly for mundane and time-consuming tasks, to AI. 

Enter the Digital Employee; an autonomous team member that can assist human employees to optimise their time to more efficiently get job-tasks done.   With the ability to process the mundane tasks behind the scenes, Digital Employees will free up their human colleagues to focus on the tasks that require the innovation, empathy, and decision-making capability of a real person (i.e., the interesting stuff!).  Furthermore, by automating repetitive and tedious tasks the likelihood of mistakes and breaches of policy are minimised.

We are sharing our vision for the next evolution of the Oracle Digital Assistant Platform to power the digital workforce of the future. This digital workforce platform will transform Digital Assistants into highly capable Digital Employees. A Digital Employee will automate end-to-end business processes by autonomously interacting with multiple team members and personalizing each interaction with a deep understanding of personal workflows, dialog context, and process execution. 

This new platform will introduce several key capabilities that ensure a productive interaction with your human employees, including:

Multi-Modal Workspaces

  • Employees will interact with Digital Employees in the same manner as with their human counterparts, via collaborative tools such as MS-Teams & Slack.
  • But when the task is more involved, they will step in to a shared workspace where they continue to converse but also collaborate with the Digital Employee to automate business processes and access business data
Working with a Digital Employee in Slack to query candidate data using natural language
Figure 7: Working with a Digital Employee in Slack to query candidate data using natural language

Multi-user Process Automation

  • Digital employees will be trained to automate end to end business workflows, not just one-off transactions,  
  • They will be deployed within teams, to automate the tedious, manual work of getting multiple, disparate business systems to work together 
  • With an autonomous engagement model, digital employees will reach out to work with multiple users to get the job done
Shared workspace
Figure 8: Collaborating with a Digital Employee in a shared workspace to automate an end to end process

Self-Service Authoring and Training of the Digital Employee

Fundamental to powering the digital workforce, is empowering non-technical business users to train the Digital Employees to learn new automations.

  • Visual designers to model both the conversational interaction and the underlying business processes of the tasks assigned to the Digital Employee. 
  • Specify business tasks conversationally by instructing the Digital Employee on what to look out for, and subsequently what to do with the information.
  • Point your Digital Employee at your database and/or document repository/knowledge base to automatically generate information centric skills – without the need to predefine your queries and Intents.
Self-service Authoring
Figure 9: Empowering non-technical business users to train Digital Employees in a variety of ways

A Roadmap for the Future of Work

While chatbots have evolved to expand their capabilities across multiple business domains, they are still fundamentally reactive in nature.  That is, they are designed to respond to a user’s enquiry and process that on-demand.  With the vision of Digital Employees, Oracle takes Digital Assistants to the next level by introducing autonomous digital teammates able to proactively address the needs of the user independent of the user’s direct involvement.  With the use of advanced Natural Language Processing, your users can instruct their digital companion as to task in hand and let them loose to do the job.  Hence freeing up time to allow your users to work more effectively on higher value work.

 

 

Salman Sheikh

Director Product Management, Oracle Digital Assistant

Salman Sheikh is Director of Product Management for Oracle Digital Assistant. In this role, he is responsible for go to market strategies as well as leveraging market insights to formulate the long term product direction. Salman has a personal passion and vision to bring business automation into the hands of average users through conversational interfaces to transform the very nature of work!

Show more

Barry Hiern

Director, Product Management, Digital Assistant & AI

Oracle Chatbot
Disconnected