Transcript slides - GeniePal

Genie Pal
A Versatile Intelligent Assistant
To Help Both Work And Personal life
Content
Targeted Problems
Our Approach
Elaboration of Our Solution
More on Focus on Web
More on AI Development
Team
The Targeted Problems
•
Using a computer system is not always easy and convenient

You have to learn to use computer UI – it is not like communicating with another human
being. And the learning has to occur for every new software or computer application,
especially those commercial systems.

To do some slightly sophisticated job using a computer system, you often need to be trained,
and you need to remember all the details in the procedure for doing the job. Missing one
step, or doing it incorrectly, you can fail the job. Often, you know what has to be done, but
you just don’t remember how.

There are lots of laborious, time-consuming jobs when we use computer, and many of them
are essentially repetitive routines despite differences in details, but we still need to do them
manually.

Siri-like human-computer interaction is great – you can tell computer what to do just as
telling a human being. But Siri is designed by Apple, and cannot help you unless it has the
functionality you want. This infeasibility is related to the second big issue mentioned below.
The Targeted Problems
•
It is difficult to add functionality to a traditional computer system

Functionality and features need to be designed and implemented as code. Once done and
released, code and features become fixed (until upgrade).

Design and implement an application software is hard, and requires significant cooperation
between computer engineers/developers and domain experts to cover all potential use
cases. Even that cannot guarantee to fully meet every customer’s requirements.

Customization and configuration is often hard, and not for normal users, let alone extending
functionality and integrating features of multiple applications and systems.

Bottom line: you cannot expect a traditional computer system to do beyond what the code
directly supports, and you cannot expect Apple’s Siri to help you if the features you like have
not been designed in Siri.
Our Approach
•
Make computer system to be teachable by users

We as users often know what to do, but computer systems don’t. If not all of us know, often
than not, one of us in the organization would know. And even if none of us average users
knows, domain experts would normally know. In any of these cases, if a system is teachable,
we wouldn’t always need computer experts and engineers to give us new functionality –
users or domain experts can just teach the computer.

Teaching a solution is much easier than teaching how to find a solution in general.
Correspondingly, designing a system with generic problem-solving intelligence is very hard,
but making it teachable is not infeasible, especially when we focus on teaching existing
solutions.

Existing solutions can be taught this way: users can show or demonstrate to computer how
to do a job or solve a type of problems by examples, just as how those more skillful people
teach apprentices or novices in workshops or offices.
Elaboration of Our Solution
•
Enable some basic natural language and semantic analysis capability.
•
Introduce the capability for the system to observe user’s behavior in addition to
other related things.
•
Implement the capability for the system to imitate the user’s behavior it
observes.
•
Attach semantic meanings to observed user’s behavior, with user’s help in
addition to internal semantic analysis.
•
Allow generic domain knowledge such as concepts and their language
expressions to be provided by user, so as to further facilitate semantic analysis in
user-defined domain.
Elaboration of Our Solution
•
Get abstraction of what user does in terms of the attached semantic meanings.
•
Use learned abstract behavior models to apply to tasks with similar intentions
and purposes, and adjust/adapt its actions according to different requirements
and context.
•
Verify, adjust and improve learned behavior models according to system’s own
execution experience.
Elaboration of Our Solution
•
Focus on Web application and services
•
Prototype designed and implemented in Android mobile platform
•
Targeted for both personal assistance usage and productivity/work assistance
usage
Elaboration of Our Solution
• GeniePal – An Integration Of Artificial Intelligence,
Web Technology, Mobile and Cloud Computing.
Elaboration of Our Solution
• GeniePal – System Architecture (1).
GeniePal Mobile Client
GeniePal Cloud Service
Elaboration of Our Solution
• GeniePal – System Architecture (2).
GeniePal Mobile Client
GeniePal Cloud Service
( Genie Mastermind )
User Interaction Engine
Language &
Semantic
Engine
GeniePal Client
Knowledge
Engine
GeniePal Client
Knowledge
Database
Execution
Engine
Genie
Mastermind
Knowledge
Engine
Genie Mastermind
Knowledge
Database
Learning
Engine
Genie Mastermind Training
System
Learning
Engine
Language &
Semantic
Engine
Why Focus on Web
•
The number of public web portals exceeds one billion on the Internet. The web
remains the most popular Internet application platform.
•
Web is also universally supported as an “Intranet” application platform – it has
been a standard. The migration to web-based technology is largely done for
commercial systems.
•
Web is not only easily accessible, but also simple and cheap to support, so it is
expected to maintain its unmatched popularity in foreseeable future.
•
The above means that we can potentially bring Siri-like intelligence to the largest
set of applications used by billions of ordinary users and commercial customers.
•
Web capability is also very important for generic AI development.
•
Implementing our idea is easier on Web partly because it is a public standard not
controlled by any companies.
More on AI Development
•
Our work puts more emphasis on enabling learning conditions and knowledge
transfer compared with mainstream AI R&D.
•
We deem observation and imitation capability as two essential enabling
conditions for learning. The two, combined with other AI elements, could be used
to construct very effective learning methods.
•
The ability to learn existing knowledge is as valuable as (if not more important
than), the ability to explore new knowledge – as human beings, most of our
abstract knowledge is learned from schools and books, and they are all existing
knowledge. In fact, not much new knowledge is discovered by us as individuals.
More on AI Development
•
From a practical perspective, even if we do not aim at creating systems that can
be as smart as we human beings in finding new knowledge, but only aim at
systems that can learn existing solutions, it can still solve a lot of practical
problems (most jobs we do everyday are not about cracking hard problems and
finding new solutions).
•
Understanding more about how existing knowledge can be transferred and
learned may benefit other AI learning development, e.g. AlphaGo. Combined
with other AI technology, learning can be more effective, and it won’t be limited
to obtaining existing knowledge.
Who are the Developers of Genie Pal
•
We are currently a small group of veteran computer engineers. We are not only
interested in generic computer science research, such as artificial intelligence, but
also eager to make real solutions and systems to solve practical problems. Our
members are in Canada and US, and we work together through virtual
collaboration working environment. We can be reached at
[email protected].