The AI Summit London, 22nd Sept. 2021.
https://www.alandix.com/academic/talks/AI-Summit-2021-UI-with-AI/
All AI ultimately affects people, in some cases deeply buried, in others interacting directly with users whether physically, such as autonomous vehicles, or virtually, such as recommender systems. In these interactions AI may be a servant, such as Alexa operating on command; or AI may be the master, such as gig-work platforms telling workers what to do. However, potentially the most productive interactions are a symbiosis, human and AI complimenting one another. Designing human-in-the-loop systems changes the requirements of both AI algorithms and user interfaces. This talk will explore some of the design principles and examples in this exciting area.
Key Takeaways:
* Deterministic ground
– helping users know what may or may not adapt
* Appropriate intelligence
– tuning AI to offer human alternatives and fail well
* Epistemic interaction
– choosing user interactions that are informative for ML
user interface, artificial intelligence, design, machine learning, deterministic ground, appropriate intelligence, alien intelligence, Epistemic interaction
6. touchstone phrases
alien intelligence
not as we know it
appropriate intelligence
fitting within interaction framework
optimise the human–AI system, not just the AI
sufficient reason
explanations we would accept from a human being
critical for AI explainability
7. styles of collaboration
Servant – we tell AI what to do (explicitly)
Master – AI tells us what to do, e.g. Uber
Symbiosis – using complementary abilities
“Man-Computer Symbiosis”
J. C. R. Licklider, March 1960
N.B. boundaries may be blurred
– e.g. maps interfaces, recommendations
8. Designing User Interactions with AI
servant, master or symbiosis
Alan Dix
https://alandix.com/aibook/
@alanjohndix @CompFoundry
9. interacting with AI
from adaptive websites to semi-autonomous cars ...
think about abilities (function allocation)
– rote vs creative
– pattern finding vs interpretation
think about handover
– physical handover
– regaining situational awareness
– not at the last moment!
10. design for interacting with AI
some heuristics ...
deterministic ground
– knowing what may change
appropriate intelligence
– when it all goes wrong
11. deterministic ground ... for menus
David
Mary
Florence
Adrian
Andy
Brian
Charlotte
Colin
David
Devina
. . .
Message
adaptive area
most popular/likely at top
quick when it gets it right
deterministic ground
alphabetic below
still easy to scan when it isn’t
13. rules of standard AI interfaces
1. it should be right as often as possible
2. when it is right it should be good
14. rules of appropriate intelligence
1. it should be right as often as possible
2. when it is right it should be good
3. when it isn’t right ...
it shouldn’t mess you up
what makes
a system
really work!
✗ ✔
15. onCue
• intelligent ‘context sensitive’ toolbar
• sits at side of the screen
• watches clipboard for cut/copy
• suggests useful things to do
with copied date
16. 20 21 22 23
25 24 20 17
7 7 3 7
the dancing histograms very useful a
ing out some of the textile sites yo
x's page at http://www.hiraeth.com/
onCue (dot com days!)
intelligent ‘context sensitive’ toolbar
histograms
user selects text
and copies it to clipboard
icons slowly fade in
17. Designing User Interactions with AI
servant, master or symbiosis
Alan Dix
https://alandix.com/aibook/
@alanjohndix @CompFoundry
27. which option? – epistemic interaction …
scroll accordion
A-B user testing
scroll a little better
accordion gives more data
for AI adaptation
28. design for interacting with AI
deterministic ground
– helping users know what may or may not adapt
appropriate intelligence
– tuning AI to offer human alternatives and fail well
epistemic interaction
– choosing user interaction that is informative for ML