1 of 13

Actual causality for everything in life�thanks to Joe Halpern�

Hana Chockler

Department of Informatics

King’s College, London

IBM Research

© 2010 IBM Corporation

© 2010 IBM Corporation

1

2 of 13

Actual Causality

A theoretical concept from AI

Extends causal counterfactual reasoning

Enables us to reason causally

about a specific situation

that occurred in the past

©Halpern & Pearl, 2001

+

@Joe Halpern – many definitions and results

Then Joe spent a sabbatical at the Hebrew University –

a key moment in my academic journey (I was a PhD student)

  • Hana Chockler, Joseph Y. Halpern:�Responsibility and Blame: A Structural-Model Approach. IJCAI 2003

  • Hana Chockler, Joseph Y. Halpern, Orna Kupferman:

What causes a system to satisfy a specification? ACM TOCL 2008

© 2010 IBM Corporation

2

3 of 13

Hana Chockler, Joseph Y. Halpern:�Responsibility and Blame: A Structural-Model Approach.

IJCAI 2003 and JAIR 2004

Quantification of causality,

allowing to rank causes by importance

To this day my most cited paper (I just checked again)

… also Joe’s 16th most cited paper (which is not saying much)

Allows to focus on the most influential causes – very important for large practical applications

© 2010 IBM Corporation

3

4 of 13

Hana Chockler, Joseph Y. Halpern, Orna Kupferman:

What causes a system to satisfy a specification? ACM TOCL 2008

Took a very long time to get published…

© 2010 IBM Corporation

4

5 of 13

Formal Verification

A huge and difficult

to understand system M:

A correctness specification φ

Does M satisfy φ?

no

counterexample

yes

the system is correct!

© 2010 IBM Corporation

5

6 of 13

Does M satisfy φ?

no

counterexample

yes

the system is correct!

Do we understand the counterexample?

A correctness specification φ

A huge and difficult

to understand system M:

Do we really believe it?

Formal Verification

© 2010 IBM Corporation

6

7 of 13

Does M satisfy φ?

no

counterexample

yes

Do we understand the counterexample?

A correctness specification φ

A huge and difficult

to understand system M:

Do we really believe it?

Everything is actual causality!

© 2010 IBM Corporation

7

8 of 13

A huge and difficult

to understand system M:

A correctness specification φ

Does M satisfy φ?

no

counterexample

yes

the system is correct!

Hana Chockler, Joseph Y. Halpern, Orna Kupferman:

What causes a system to satisfy a specification? ACM TOCL 2008

Everything is actual causality!

© 2010 IBM Corporation

8

9 of 13

A huge and difficult

to understand system M:

A correctness specification φ

Does M satisfy φ?

no

counterexample

yes

the system is correct!

Ilan Beer, Shoham Ben-David, Hana Chockler, Avigail Orni, Richard J. Trefler:

Explaining Counterexamples Using Causality. CAV 2009 (and FMSD)

My second most cited paper

© 2010 IBM Corporation

9

10 of 13

Explaining counterexamples using causality�(Red Dots)�part of IBM tool

A timing diagram of a buggy hardware execution

φ = always ((!START and !STATUS_VALID and END) ->

next(!START Until (STATUS_VALID and READY))

causes marked as red dots

© 2010 IBM Corporation

10

11 of 13

Black-box reasoning about AI models’ decisions

Intervene on inputs

inputs

outputs

Input transformation

Observe the outputs

Reason about the way AI makes its decisions

One-node causal model

  • Hana Chockler, Joseph Y. Halpern:�Explaining Image Classifiers. KR 2024
  • Hana Chockler, David A. Kelly, Daniel Kroening, Youcheng Sun:�Causal Explanations for Image ClassifiersJAIR 2026

© 2010 IBM Corporation

11

12 of 13

Ongoing theoretical work with Joe: harm, fairness, neural networks

  • Hana Chockler, Joseph Y. Halpern:�On Testing for Discrimination Using Causal Models. AAAI 2022

  • Sander Beckers, Hana Chockler, Joseph Y. Halpern:�A Causal Analysis of Harm. NeurIPS 2022 and  Minds and Machines 2024

  • Sander Beckers, Hana Chockler, Joseph Y. Halpern:�Quantifying Harm. IJCAI 2023,

… journal version under submission.

  • We last saw Joe on the 31st of October 2025 in London and planned to start working on a new paper

  • A lot of ideas and future directions…

© 2010 IBM Corporation

12

13 of 13

Writing papers with Joe

  • Very ordered, meticulous about keeping all versions
  • Never just erased anything – always respectful to his co-authors
  • Some papers have %joe25 and up – that’s how many passes Joe made on the paper
  • Makes Latex impossible to read for anybody but Joe

© 2010 IBM Corporation

13