Download to read offline
![Types of coherence
training decision and
explanation
new decision
model
M
M′
X, dX, eX Y, dY, eY
updated training and model
X, d′X, e
′X
potential
incoherence
dx d′
≁ X
eX e
≁ Y
dY e
≁ X[Y]
ex e′
≁ X](https://image.slidesharecdn.com/bath-2025-explaining-251114084807-623ef0e4/75/Explaining-ourselves-people-computers-and-AI-26-2048.jpg)
![Types of coherence
training decision and
explanation
new decision
model
M
M′
X, dX, eX Y, dY, eY
updated training and model
X, d′X, e
′X
potential
incoherence
dx d′
≁ X
eX e
≁ Y
dY e
≁ X[Y]
ex e′
≁ X](https://crownmelresort.com/image.slidesharecdn.com/bath-2025-explaining-251114084807-623ef0e4/75/Explaining-ourselves-people-computers-and-AI-26-2048.jpg)
Uploaded byAlan Dix
Explaining ourselves – people, computers and AI
Talk at University of Bath, UK, 12th November 2025 https://alandix.com/academic/talks/Bath-2025-explaining-ourselves/ As humans we constantly act and speak so that implicitly or explicitly we make our intentions more understandable by others. This talk explores a number of ways in which this can also be possible in our interactions with computers in general and AI in particular. This includes coherence in explainable AI; epistemic interaction designed to give AI more information about users; human explanations in synergistic interactions with AI; and explainable user interfaces for non-AI human–computer interaction. Several of these topics have been developed as part of the TANGO EU Horizon project on hybrid human–AI decision making. Some are illustrated in prototypes; others are at conceptual state; all pose challenges and opportunities for future research.
More Related Content
Similar to Explaining ourselves – people, computers and AI
Explaining ourselves – people, computers and AI
- 1. Alan Dix Explaining Ourselves- people, computers and AI https://alandix.com/academic/talks/Bath-2025-explaining- ourselves/
- 2. today I amnot talking about … • qualitative–quantitative reasoning • deep digitality and digital thinking • next generation UX tools • long tail of small data • physicality • now • digital light • walking round Wales • virtual crackers and slow time • digital humanities and community heritage • modeling dreams, regret and the emergence of self
- 3. plugs 2nd edition out now! 2 nd edition 2026 plus … AIfor HCI HCI the Basics AI for Social Justice
- 4. This project hasreceived funding from the European Union’s Horizon Europe research and innovation programme under Grant Agreement No. 101120763. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or HaDEA. Neither the European Union nor the granting authority can be held responsible for them. it takes two to tango a synergistic approach to human-machine decision making https://tango-horizon.eu/
- 5. AI and UserInteraction H CI for AI Rich System s Intelligent Interfaces Ethics S o c i a l J u s t i c e G o v e r n a n c e Human like Computing I n t e r f a c e s f o r A I d e v e l o p e r s Big D ata and Evaluation
- 6.
- 7.
- 8.
- 9. human–{X} interactions • human–human •common ground, collaborative principle , accountability • human–world–human • feedthrough, onomatopoeic action • human–object • ecological psychology, affordance, epistemic action • human–technology • technology evolution, designed affordances, Norman’s loop
- 10.
- 11. AI design algorithm choice rules,policies data AI designer & developer
- 12.
- 13. good AI+HCI not aboutdeveloping the most accurate AI but creating the most effective and enjoyable overall human-technical system
- 14. synergy cooperation and adaptation adaptingAI for human interaction adapting user interfaces for AI AI
- 15. synergy cooperation and adaptation adaptingAI for human interaction adapting user interfaces for AI AI
- 16. adapting interaction forAI ‘best’ UI – short term gain epistemic interaction – long term gains
- 17.
- 18. epistemic interaction redesign userinteractions to make more information available for machine adaptation
- 19. document results –option 1 – scroll scroll down to see more results
- 20. document results –option 2 – accordion press ‘+’ to expand interesting items
- 21. which option? –epistemic interaction … scroll accordion A-B user testing scroll a little better accordion gives more data for AI adaptation
- 23.
- 24. human–AI communication training data advice/ prediction explanation (XAI) gender and ethnic bias in black-box ML Query by Browsing
- 25. coherence humans expect human explanationsto be consistent through time AI explanations should be too training data advice / prediction explanation (XAI) S. Myers, N. Chater, Interactive explainability: Black boxes, mutual understanding and what it would really mean for AI systems to be as explainable as people, 2024. doi:10.31234/osf.io/ha37x … but how?
- 26. Types of coherence trainingdecision and explanation new decision model M M′ X, dX, eX Y, dY, eY updated training and model X, d′X, e ′X potential incoherence dx d′ ≁ X eX e ≁ Y dY e ≁ X[Y] ex e′ ≁ X
- 27. Partial patch model setof all possible inputs L2 R ⇢ 2 L3 R ⇢ 3 L4 R ⇢ 4 L1 R ⇢ 1 L5 R ⇢ 5 size and shape of patches dealing with overlap areas of greater or lower patch density
- 29.
- 30. communication training data advice /prediction explanation (XAI) critique explanation
- 31. QbB with userexplanation – global features
- 32. QbB with userexplanation – local feature
- 34.
- 35.
- 36. the problem what justhappened? what did I do? how do I do this again? all of us, but especially elderly, less IT-literate, poor motor control, after interruptions we need explainable user interfaces
- 37. why needed –locus of control early models – user in control CLI & GUI, Norman’s execution–evaluation cycle never just the user other people, physical sensors, timing now notifications, AI who is in control?
- 38. why needed –complexity and scale ever smaller screens every larger data and computatio
- 39. why needed –shrinking gaps theories of language – clarity => distinctions binary opposition (Saussure), information transmission (Goodman) UI – reducing boundaries spatial fat finger, two/three finger swipe temporal click vs double-click, changing meanings, appearing targets semantic spelling correction, autocomplete
- 40.
- 41. can explanations help? understanding,control, empowerment when things go wrong undo and repair when things worked replay/redo memory in a post ‘recognition rather than recall’ world providing mutual help third party helpers – what happened here? sharing knowledge – here’s what I did
- 42. what XUI mightbe like?
- 43. what XUI mightbe like? commands and scripts what happened when – scripts and undo factoring applications
- 44. what XUI mightbe like? commands and scripts what happened when – scripts and undo what happened here? focus on object e.g. WS2
- 45.
- 46. how to doit • within application • UI architecture crosses boundaries: • monolithic (e.g. Seeheim) – lexical/syntactic/semantic • components (e.g. MVC) – maybe more difficult? • cross-application • OS standards and protocols? • model-based user interfaces • in theory • and practice e.g. React lexical level Apple Events (Numbers)
- 49. take aways … learningfrom human communication … but different! • epistemic interaction • implicitly revealing intentions • coherence • can AI be as (more?) consistent as people :-/ • talking back • explicitly explaining ourselves to AI • explainable UI • busting the myth of ease of use
- 50. Explaining Ourselves people, computersand AI https://alandix.com/academic/talks/Bath-2025-explaining-ourselves/
Editor's Notes
- #16 Lawrence Alma-Tadema's water-colour of an ambivalent Pandora, 1881 https://en.wikipedia.org/wiki/Pandora%27s_box#/media/File:Lawrence_Alma-Tadema_10.jpeg https://www.publicdomainpictures.net/en/view-image.php?image=82495&picture=cat-peeking-around-corner
- #17 Lawrence Alma-Tadema's water-colour of an ambivalent Pandora, 1881 https://en.wikipedia.org/wiki/Pandora%27s_box#/media/File:Lawrence_Alma-Tadema_10.jpeg https://www.publicdomainpictures.net/en/view-image.php?image=82495&picture=cat-peeking-around-corner
- #43 Qurren, CC BY-SA 3.0 <http://creativecommons.org/licenses/by-sa/3.0/>, via Wikimedia Commons https://commons.wikimedia.org/wiki/File:5.25-inch_floppy_disk.jpg Eric Gaba, Wikimedia Commons user Sting, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons https://commons.wikimedia.org/wiki/File:Seagate_ST33232A_hard_disk_inner_view.jpg BalticServers.com, CC BY-SA 3.0 <https://creativecommons.org/licenses/by-sa/3.0>, via Wikimedia Commons https://commons.wikimedia.org/wiki/File:BalticServers_data_center.jpg Arielinson, CC BY-SA 4.0 <https://creativecommons.org/licenses/by-sa/4.0>, via Wikimedia Commons https://commons.wikimedia.org/wiki/File:Control_Room_DSC0028.jpg Raralu4440, CC BY-SA 4.0 <https://creativecommons.org/licenses/by-sa/4.0>, via Wikimedia Commons https://commons.wikimedia.org/wiki/File:Apple_Watch_5_40mm_on_my_desk.jpg