Embed Size (px)
Citation preview
EECS 498Conversational Artificial Intelligence
Principles and Practice
Dr. Kevin Leach; Professor Jason Mars; Brian Yang; Oliver Strong1
For Today
• What is this course?• What isn’t this course?• What is Conversational AI?• How does it work?• How do you build conversational AI?• What are we doing in this course?
5
Course Administration
• Course Meetings:• MW 1:30-3 PM in DOW1010
• No required textbook
• Office Hours• Kevin: Wednesday 3-4pm (BBB 4705), or by appointment• Brian/Oliver: by appointment (Google Calendar signup)• Nikhil and Sahil: TBA
6
Course Administration
• Course Website: https://dijkstra.eecs.umich.edu/eecs498
• Piazza: https://piazza.com/umich/winter2020/eecs498• Sign up if you are not already• Please use Piazza before emailing instructors• Discussion is crucial!
7
Piazza
8
9
Conversational Artificial Intelligence
• “OK Google, teach me computer science.”
• Conversational AI is the use of software to empower computers with the ability to complete tasks or hold natural language conversations with humans.
10
Conversational Artificial Intelligence
11
Conversational Artificial Intelligence
12
Major Design Experience and Special Topics
• This course counts as an MDE elective• You may be required to take EECS 498 and TCHNCLCM 497 concurrently• Consult undergraduate advising office for more information (BBB 2808)
• This course is almost exclusively project-based• An opportunity for high impact!
• This is a 400-level EECS Special Topics course• More independence required
13
Major Design Experience (MDE)
• Decide on teams early!• Up to 8 people per group (at least 5)
• Grading is almost exclusively based on project output• Design• Presentations• Demonstrations• Robust code and engineering
• You will be working with a real company! (Clinc)14
Grading
• Approximate breakdown• Project pitch (5%)• Scoping Review (5%)• Sprint Reviews (40%)• Cooperative Testing (10%)• Final Presentation / Demonstration (20%)• Demonstration Video (10%)• Participation (10%)• Extra Credit (up to 5%)
• TL;DR: 100% project15
Attendance
• We’ll record lectures and put them up, but you are expected to attend sprint reviews in person.
• Receiving and incorporating feedback from others is crucial• Participation is 10% of your grade
16Crede et al. “Class Attendance in College: A Meta-Analytic Review of the Relationship of Class Attendance With Grades and Student Characteristics.” Review of Educational Research, 2010. Vol. 80. DOI: 10.3102/0034654310362998
Structure and Style of the Course
• Special Topics electives are inherently less structured
• Learn by doing and sharing• Reviewing and analyzing what we build, together• Ambition is required
• Tools: Clinc Conversational AI Platform17
What this class is• A unique opportunity to
• Build a high-impact conversational AI project
• Learn to use a robust, enterprise-grade conversational AI platform (Clinc)
• Work with a team of students and conversational AI experts on a large, complex software system
• Gain feedback, coaching, and advice that will positively impact your future career
18
What this class is not
• This is not an AI Foundations course• This is not a Machine Learning
course• This is not a math course• This is not an introductory
programming course• This is not a course with spoon-fed
projects
19
Our commitment to you
• We will give you advising and coaching based on real-world experience that will help develop your career
• We will give you feedback to help make your project the best resume booster it can be!
• We will give you opportunities to pitch ideas, develop them, demonstrate them, and integrate them in a real, customer-facing platform
• We will expect your ambition and excitement during this semester
20
Artificial Intelligence
21
Conversational Artificial Intelligence
• Artificial Intelligence –“intelligence exhibited by machines or software” (Wikipedia)
Broadly, AI seeks to emulate cognitive processes that humans have.
- Game AI- Optical Character Recognition (OCR)- Machine learning- Natural language understanding (this course)
22
Conversational Artificial Intelligence
• Artificial Intelligence –“intelligence exhibited by machines or software” (Wikipedia)
• Conversational –“interactive communication between two or more people” (Wikipedia)
• Conversational Artificial Intelligence –Enabling a machine to have natural conversations
23
So how does it actually work?
• We use an AI engine to help understand natural language that is provided by the user.
• Once we understand what the user wants to do, we use business logic to help complete the user’s request
• Once the business logic completes the request, we construct a response that is given to the user
24
Conversational AI: Example workflow
25
Automated Speech Recognition
(ASR)
Spoken language (i.e., sound)
utterance
“Please pay Dr. Leach $1000.”
Natural language understanding
Intent Classification:transfer_money
Slot mapping:recipient: “Dr. Leach”amount: “$1000”
Business Logic:
Deduct $1000 from accountAdd $1000 to recipient account
Response Generation
Template Responses
“OK, I gave Dr. Leach $1000.”
“Sorry fam, you don’t have
enough cash”
Text-to-speech (TTS)
Conversational AI: Example workflow
26
Automated Speech Recognition
(ASR)
Spoken language (i.e., sound)
utterance
“Please pay Dr. Leach $1000.”
Natural language understanding
Intent Classification:transfer_money
Slot mapping:recipient: “Dr. Leach”amount: “$1000”
Business Logic:
Deduct $1000 from accountAdd $1000 to recipient account
Response Generation
Template Responses
“OK, I gave Dr. Leach $1000.”
“Sorry fam, you don’t have
enough cash”
Text-to-speech (TTS)
Applications of Conversational AI
27
Applications of Conversational AI
28
Conversational Artificial Intelligence
• Goal: empower computers to have meaningful, natural, and actionable conversations with humans
29
Conversational Artificial Intelligence: Reality
30
Applications of Conversational AI: Task-based
• “Turn on my ceiling fan.”• “Accelerate to 9001 miles per hours.”• “Open the pod bay doors, HAL.”
31
Applications of Conversational AI: Chatbots(Dialog Systems)• User: Hello.• Bot: How are you?• User: I’m good, how about you?• Bot: Good! What are you up to?• User: Enjoying EECS498.• Bot: Awesome possum.
32
Applications of Conversational AI: Chatbots
• User: Hello.• Bot: How are you?• User: I’m good, how about you?• Bot: Good! What are you up to?• User: Enjoying EECS498.• Bot: Awesome possum.
33
Applications of Conversational AI: QA Systems
• “How do I turn on cruise control?”
• “Who is Kebert Xela?”
34
Applications of Conversational AI: Summary
• Task-oriented (intent-based)• Chatbot / Dialog systems• QA
• Can you think of other broad areas?
35
Let’s brainstorm
• In small groups, come up with a few ideas for conversational AI applications
• Doesn’t matter if they’re task-oriented, chatbot, or QA style
• In a perfect world, what would you want to be able to talk to a computer about? What would it be capable of doing?
36
So what are we actually doing?
• In this class, you will• Work on a project in a manner that mimics industrial development
• Identifying problems• Designing solutions• Communicating ideas• Incorporating feedback• Delivering demonstrations and prototypes
• We will adopt an agile development methodology• Roughly every 2 weeks, we will have a sprint where you complete some deliverable
• Note: Some lecture slots will be used for group presentations/demos
37
So what are we actually doing?
• Aside from the project, you will:• Learn the basics of ML, NLP, and conversational AI
• Learn how to use an enterprise-grade conversational AI platform, Clinc
• Learn about cutting-edge NLP and AI research
• Gain coaching from industrial practitioners
• Get career advice
38
So what are we actually doing?
• Action items:• Form teams by September 16!
• Identify teammates through class or piazza• Email Jason and me a list of email IDs of your team
• Turn in project proposal by September 23• Slide deck that describes a problem that conversational AI can solve• Describe how conversational AI could solve it
• Present project proposals during 9/23 and 9/25• Get feedback from instructors and classmates
39
Project Ideas: Last semesters
• Groceries: Find ingredients in a grocery store• Recipes: find and walk through steps of a recipe• Housing: search home listings• Travel: create travel itineraries
• Spotify Playlist curation: create playlists meeting certain criteria• Mental health curation: chatbot to help users with panic attacks• Code summarization: ask GitHub to explain code to you• Accessible computing: Summarize images for blind users• Homework help: Explain complicated math equations
• (hint: don’t try an EECS advisor)
40
41
42
