Tomas Petricek
email: t.petricek@kent.ac.uk
twitter: @tomaspetricek
office: S129A
ELIZA (1964)
Written by Joseph Weizenbaum at MIT
To demonstrate that the communication between man and machine was superficial
1. Looks for simple patterns in text
2. Replace words to form a question
Surprisingly effective
Weizenbaum's own secretary asked Weizenbaum to leave the room so that she and ELIZA could have a real conversation.
Weizenbaum's commentary
I had not realized (..) that short exposures to a simple
computer program could induce powerful
delusional thinking in quite normal people."
Unintended consequences of Artificial Intelligence
We know how it works technically...
But no idea how it works in the world!
Deep Blue vs Gary Kasparov (1997)
First computer to beat a world champion.
Advanced Chess
(since 1998)
Can a computer program complement a human?
Human thinking
State space search
Automation or symbiosis?
How users think about AI systems
Should we expect new wave of HCI soon?
Perhaps accidentally, AI winters are HCI summers!
Data science tasks
Human and computer
Trifacta Wrangler
Combining UX and AI techniques for generating data extraction scripts
Human computer symbiosis
User provides examples to refine answer
AI attempts to fit model to samples
Model is readable source code
Visipedia
Human in the loop for image recognition
AI asks for help, using questions easy for human
Human computer symbiosis
Humans are good at recognizing key features
Computers can efficiently search
Dialog metaphor for the interaction
Racist Google Photos
Example of biased AI
What is the reason why the app does this?
What was the training data used by Google?
How to avoid this?
Right to explanation
Explainable AI and user interaction
How Britain Voted
The Times
Using decision trees to map the structure of the Brexit referendum
Statistical models
Explicit logical model
Agency and training data sets
Programming by example such as Wrangler
The nature of human and AI reasoning styles
Legal status of non-symbolic intellectual property
What does it mean to be creative?
Does output look like a result of creative action?
Is computer following creative processes?
How do users interact with the system?
Deep style transfer
Using machine learning to change image style
Looks pretty, but is there any creativity behind the system?
The Drawing Fool
Software that explains
its artistic decisions
Uses newspaper to decide mood, AI for drawing and
AI for reflection
Rethink human-centric notion of creativity
Computers and creativity
Why explanation matters
The problem with AI
Shneiderman (1989) has argued that AI in interfaces reduces predictability, which is essential for usability.
AI in user interfaces
Phones changes what we can fit on the screen
Motion sensing introduced new kinds of games
Chatbots (try to) make computers easier to use
Voice recognition allows hands-free interactions
Mobile text-entry methods
T9 predictive text
One key for three letters
Graffiti for Palm OS
Letter recognition
Predictive keyboards
Guess word from a stroke
Principles of AI for text entry
User experience questions
Human in the loop data science
Ways of efficient collaboration
Dialog and programming by example
The problem of AI explainability
Avoiding bias in AI systems
Statistical and logical models
The problem of AI creativity
Creativity as a user-defined criteria
Human reasoning as an inspiration
What you should remember from this lecture
Tomas Petricek
t.petricek@kent.ac.uk | @tomaspetricek
Papers and links