iSee: Intelligent Sharing of Explanation Experiences
Kyle Martin1,∗ , Anjana Wijekoon1 , Nirmalie Wiratunga1 , Chamath Palihawadana1 ,
Ikechukwu Nkisi-Orji1 , David Corsar1 , Belén Díaz-Agudo2 , Juan A. Recio-García2 ,
Marta Caro-Martínez2 , Derek Bridge3 , Preeja Pradeep3 , Anne Liret4 and
Bruno Fleisch4
1
Robert Gordon University, Aberdeen, Scotland
2
Universidad Complutense de Madrid, Madrid, Spain
3
University College Cork, Cork, Ireland
4
BT France, Paris, France
Abstract
The right to an explanation of the decision reached by a machine learning (ML) model is now an
EU regulation. However, different system stakeholders may have different background knowledge,
competencies and goals, thus requiring different kinds of explanations. There is a growing armoury of XAI
methods, interpreting ML models and explaining their predictions, recommendations and diagnoses. We
refer to these collectively as ”explanation strategies”. As these explanation strategies mature, practitioners
gain experience in understanding which strategies to deploy in different circumstances. What is lacking,
and what the iSee project will address, is the science and technology for capturing, sharing and re-using
explanation strategies based on similar user experiences, along with a much-needed route to explainable
AI (XAI) compliance. Our vision is to improve every user’s experience of AI, by harnessing experiences of
best practice in XAI by providing an interactive environment where personalised explanation experiences
are accessible to everyone.
Video Link: https://youtu.be/81O6-q_yx0s
Keywords
Explainability, Case-Based Reasoning, Project Showcase
1. iSee System Overview
The iSee platform employs a Case-Based Reasoning (CBR) methodology to capture, store and
recommend explanation experiences. When queried, the system draws on a case-base of histor-
ical explanation experiences to suggest an appropriate explanation strategy (i.e. combination of
explainer algorithms designed to holistically satisfy a set of personas and corresponding intents).
Cases are formed of knowledge of the AI model and its user group (problem component), the
explanation strategy recommended (solution component), and feedback from the user group to
describe whether the provisioned explanations were satisfactory (outcome component). In this
manner, cases represent a comprehensive record of explanation experience.
ICCBR CBR Demos’22: Workshop on CBR Demos and Showcases. ICCBR-2022, September, 2022, Nancy, France
∗
Corresponding author.
Envelope-Open k.martin3@rgu.ac.uk (K. Martin)
Orcid 0000-0003-0941-3111 (K. Martin)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
1
Kyle Martin et al. ICCBR’22 Workshop Proceedings
USER MODELS Interactive Retrieval Evaluation Cockpit Revision
INTERACTION
and Adaptation Precise?
KNOWLEDGE EVAL. COCKPIT More FEEDBACK Revised XE
ML Model? Adapted XE Asdfdsafdsf sdfsd a
details
EVALUATION A sdfasdfasdf a
Asdf asdf asd fads
XM
CRITERIA Asdf adsf ads ads ad
Asdfasdfa dsdas
XM
Useful?
Dsadsf dsfasdfadsf a
XM Asd asd sad dsa df
Asdfasdfdfadsaadf
Asdlkfjdslkfjlkjf as
XM
Data? Yes EM
Asdfsadfds asdf sdf adsfdasf adf asdfasd
fdsf f asdf fsdf asfdsf a adsf asdf asdf as
EM asdf adsf asdf asdf ads asdf a asdf asdf XM XM
ONTOLOGY sd asdf s asdf asdf asdf adsf adf
END-USER
Clear?
DESIGNER XM XM
Goal? Satisfied XAI Certification
WP1 WP3 WP4 WP4&5
eXplanation eXplanation Interaction Evaluation
Experience
Methods API Engine Methods API
X eXplanation
M Experience
eXplanation INTERACTIVE EM
EVALUATION
X
M Experience EXPLANATION
METHODS
XM XM XM
ESTRATEGIES
X
M
X
M
X
M
eXplanation
Experience CBR ML MODEL
EM
EM
X
M
X
M Engine XM XM XM EXECUTION
DATA STORAGE
ANALYTICS
X
M
USER
X X XM XM XM HCI
M M
REPRODUCIBILITY
CASEBASE REASONING PLATFORM WP2 USE CASES WP5
Retain
Figure 1: iSee System Diagram
We have developed novel knowledge bases and algorithms to fill the CBR knowledge con-
tainers. The case base knowledge container is filled with explanation experiences, where the
above case composition has been applied to novel application of XAI in literature and real-life
use cases. We have created iSeeOnto, an ontology for the description of XAI systems which
fills the vocabulary container and describe AI models, how they are explained and how these
explanations are evaluated. The similarity knowledge container is currently under development.
There we are building novel similarity methods based on Many-Are-Called, Few-Are-Chosen
(MAC/FAC) and edit distance based methods. Finally, the adaptation container will empower
users to adapt their personalised user strategy, in the first instance through constructive adapta-
tion during reuse, and finally by helping to evaluate their explanation strategy with end users
to and modification to meet needs by revisiting earlier retrieval stages of the process.
2. Community Support
The iSee project aims to become a platform which facilitates capture of explanation experi-
ences and empowers reuse of explanation strategies. To achieve this, iSee would benefit from
community support in two areas:
1. Provision of explainer algorithms to increase available explanation strategies.
2. Description of known use cases to expand explanation experience coverage.
As a benefit to the community, by providing these artefacts the XAI research community can
expect wider impact and more convenient reusability of their contributions.
Funding and Acknowledgements
This research is funded by the iSee project (https://isee4xai.com) which received funding from
EPSRC under the grant number EP/V061755/1. The iSee project is supported by it’s use-case
partners: BT, Bosch, Jiva, Total Energies and Automix. Special thanks to Malavika Suresh and
Craig Pirie as Dr Armstrong and Dr McOlon respectively.
2