A Computational Model of Moral and Legal Responsibility via Simplicity Theory
1 like194 views
The document discusses using cognitive models and simplicity theory to understand how humans attribute responsibility. It suggests that responsibility attribution is a fundamental cognitive mechanism that aims to identify unexpected situations based on the complexity of causation versus description. Simplicity theory defines unexpectedness based on the length of the shortest program needed to generate a situation versus determine it. This theory can help predict factors that influence blame, like outcome severity, and could provide an alternative to traditional logical and statistical approaches to computing responsibility.
![● Traditional research track in AI & Law:
– structural (logical) approaches
● focus on reasoning constructs: Ontologies [Lehmann et al., 2004],
Inferences [Prakken, 2002] or Stories [Bex et al., 2000]
How to compute responsibility?](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-4-320.jpg)
![● Traditional research track in AI & Law:
– structural (logical) approaches
● focus on reasoning constructs: Ontologies [Lehmann et al., 2004],
Inferences [Prakken, 2002] or Stories [Bex et al., 2000]
– quantitative approaches
● focus on relative support of evidence: Bayesian inference [Fenton et
al., 2012], Causal Bayesian Networks [Halpern, 2015]
How to compute responsibility?](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-5-320.jpg)
![● Traditional research track in AI & Law:
– structural (logical) approaches
● focus on reasoning constructs: Ontologies [Lehmann et al., 2004],
Inferences [Prakken, 2002] or Stories [Bex et al., 2000]
– quantitative approaches
● focus on relative support of evidence: Bayesian inference [Fenton et
al., 2012], Causal Bayesian Networks [Halpern, 2015]
– hybrid methods [Vlek et al., 2014], [Verheij, 2014]
How to compute responsibility?](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-6-320.jpg)
![● Traditional research track in AI & Law:
– structural (logical) approaches
● focus on reasoning constructs: Ontologies [Lehmann et al., 2004],
Inferences [Prakken, 2002] or Stories [Bex et al., 2000]
– quantitative approaches
● focus on relative support of evidence: Bayesian inference [Fenton et
al., 2012], Causal Bayesian Networks [Halpern, 2015]
– hybrid methods [Vlek et al., 2014], [Verheij, 2014]
● Here we introduce an alternative research direction,
building upon cognitive models.
How to compute responsibility?](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-7-320.jpg)
![● Experiments show that people are more prone to blame an agent
for an action:
flooded mine dilemma (trolley problem variation)
[A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012, pages 947–952]](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-12-320.jpg)
![● Experiments show that people are more prone to blame an agent
for an action:
– the more the outcome is severe,
– the more they are closer to the victims,
– the more the outcome follows the action.
flooded mine dilemma (trolley problem variation)
[A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012, pages 947–952]](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-13-320.jpg)
![● Experiments show that people are more prone to blame an agent
for an action:
– the more the outcome is severe,
– the more they are closer to the victims,
– the more the outcome follows the action.
● The cognitive model of Simplicity Theory predicts these results.
flooded mine dilemma (trolley problem variation)
[A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012]](https://image.slidesharecdn.com/sileno-171215151233/85/A-Computational-Model-of-Moral-and-Legal-Responsibility-via-Simplicity-Theory-14-320.jpg)
A Computational Model of Moral and Legal Responsibility via Simplicity Theory
- 1. 15 December 2017, JURIX @ Luxembourg
- 2. with the (supposedly) near advent of autonomous artificial entities, or similar forms of distributed automatic decision making, to define operationally the notion of responsibility becomes of primary importance.
- 3. ● Traditional research track in AI & Law: How to compute responsibility?
- 4. ● Traditional research track in AI & Law: – structural (logical) approaches ● focus on reasoning constructs: Ontologies [Lehmann et al., 2004], Inferences [Prakken, 2002] or Stories [Bex et al., 2000] How to compute responsibility?
- 5. ● Traditional research track in AI & Law: – structural (logical) approaches ● focus on reasoning constructs: Ontologies [Lehmann et al., 2004], Inferences [Prakken, 2002] or Stories [Bex et al., 2000] – quantitative approaches ● focus on relative support of evidence: Bayesian inference [Fenton et al., 2012], Causal Bayesian Networks [Halpern, 2015] How to compute responsibility?
- 6. ● Traditional research track in AI & Law: – structural (logical) approaches ● focus on reasoning constructs: Ontologies [Lehmann et al., 2004], Inferences [Prakken, 2002] or Stories [Bex et al., 2000] – quantitative approaches ● focus on relative support of evidence: Bayesian inference [Fenton et al., 2012], Causal Bayesian Networks [Halpern, 2015] – hybrid methods [Vlek et al., 2014], [Verheij, 2014] How to compute responsibility?
- 7. ● Traditional research track in AI & Law: – structural (logical) approaches ● focus on reasoning constructs: Ontologies [Lehmann et al., 2004], Inferences [Prakken, 2002] or Stories [Bex et al., 2000] – quantitative approaches ● focus on relative support of evidence: Bayesian inference [Fenton et al., 2012], Causal Bayesian Networks [Halpern, 2015] – hybrid methods [Vlek et al., 2014], [Verheij, 2014] ● Here we introduce an alternative research direction, building upon cognitive models. How to compute responsibility?
- 8. ● In human societies, responsibility attribution is a spontaneous and seemingly universal behaviour. Responsibility attribution for humans 12 Angry Men, 1956Rashomon, 1950
- 9. ● In human societies, responsibility attribution is a spontaneous and seemingly universal behaviour. ● Non-related ancient legal systems bear much resemblance to modern law and seem perfectly sensible nowadays. Responsibility attribution for humans Rashomon, 1950 12 Angry Men, 1956
- 10. ● In human societies, responsibility attribution is a spontaneous and seemingly universal behaviour. ● Non-related ancient legal systems bear much resemblance to modern law and seem perfectly sensible nowadays. → responsibility attribution may be controlled by fundamental cognitive mechanisms. Responsibility attribution for humans 12 Angry Men, 1956Rashomon, 1950
- 11. ● In human societies, responsibility attribution is a spontaneous and seemingly universal behaviour. ● Non-related ancient legal systems bear much resemblance to modern law and seem perfectly sensible nowadays. → responsibility attribution may be controlled by fundamental cognitive mechanisms. Responsibility attribution for humans Working hypothesis: attributions of moral and legal responsibility share a similar cognitive architecture 12 Angry Men, 1956Rashomon, 1950
- 12. ● Experiments show that people are more prone to blame an agent for an action: flooded mine dilemma (trolley problem variation) [A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012, pages 947–952]
- 13. ● Experiments show that people are more prone to blame an agent for an action: – the more the outcome is severe, – the more they are closer to the victims, – the more the outcome follows the action. flooded mine dilemma (trolley problem variation) [A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012, pages 947–952]
- 14. ● Experiments show that people are more prone to blame an agent for an action: – the more the outcome is severe, – the more they are closer to the victims, – the more the outcome follows the action. ● The cognitive model of Simplicity Theory predicts these results. flooded mine dilemma (trolley problem variation) [A. Saillenfest and J.-L. Dessalles. Role of Kolmogorov Complexity on Interest in Moral Dilemma Stories. CogSCI 2012]
- 15. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain.
- 16. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness
- 17. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity concerning how the world generates the situation
- 18. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity concerning how the world generates the situation description complexity concerning how to identify the situation
- 19. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity concerning how the world generates the situation description complexity concerning how to identify the situation The two complexities are defined following Kolmogorov complexity.
- 20. Kolmogorov complexity length in bits of the shortest program generating a string description of an object
- 21. Kolmogorov complexity length in bits of the shortest program generating a string description of an object string equivalent programs “2222222222222222222222222” = “2” + “2” + … + “2” = “2” * 25 = “2” * 5^2
- 22. Kolmogorov complexity length in bits of the shortest program generating a string description of an object depends on the available operators!! string equivalent programs “2222222222222222222222222” = “2” + “2” + … + “2” = “2” * 25 = “2” * 5^2
- 23. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity about how the world generates the situation description complexity about how to identify the situation length of shortest program creating the situation length of shortest program determining the situation
- 24. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity about how the world generates the situation description complexity about how to identify the situation length of shortest program creating the situation instructions = causal operators length of shortest program determining the situation instructions = mental operators
- 25. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity about how the world generates the situation description complexity about how to identify the situation length of shortest program creating the situation instructions = causal operators length of shortest program determining the situation instructions = mental operators SIMULATION REPRESENTATION SIMULATION REPRESENTATION
- 26. Simplicity theory ● Human individuals are highly sensitive to complexity drops: i.e. to situations that are simpler to describe than to explain. ● Core notion: Unexpectedness causal complexity about how the world generates the situation description complexity about how to identify the situation length of shortest program creating the situation instructions = causal operators length of shortest program determining the situation instructions = mental operators SIMULATION REPRESENTATION SIMULATION REPRESENTATION for the agent!!!
- 27. Examples 22222222222222 is more unexpected than 21658367193445 (in a fair extraction)
- 28. Examples 22222222222222 is more unexpected than 21658367193445 meeting Obama is more unexpected than meeting Dupont (in a fair extraction) Unexpectedness captures plausibility (or any other famous person) (or any other unknown person) meeting an old of friend of mine (or any other known person)
- 29. ● Focusing on intensity, we can capture anticipation as: emotion what the situation induces to the agent reward inverse model unexpectedness Simplicity Theory: Intention
- 30. ● Focusing on intensity, we can capture anticipation as: ● If the agent A expects that the best way to bring about s is via a: emotion what the situation induces to the agent reward inverse model unexpectedness Simplicity Theory: Intention
- 31. ● Focusing on intensity, we can capture anticipation as: ● If the agent A expects that the best way to bring about s is via a: emotion what the situation induces to the agent reward inverse model unexpectedness Simplicity Theory: Intention intention as driven by anticipated emotional effects
- 32. Simplicity Theory: Intention ● Focusing on intensity, we can capture anticipation as: ● If the agent A expects that the best way to bring about s is via a: emotion what the situation induces to the agent reward inverse model unexpectedness intention as driven by anticipated emotional effects inadvertence
- 33. Simplicity Theory: Moral responsibility ● Difference between intention and moral responsibility is one of point of views. computed by A
- 34. Simplicity Theory: Moral responsibility ● Difference between intention and moral responsibility is one of point of views. computed by A computed by a model of A computed by an observer O
- 35. Simplicity Theory: Moral responsibility ● Difference between intention and moral responsibility is one of point of views. computed by A computed by a model of A computed by an observer O prescribed role, reasonable standard reward inverse model
- 36. Simplicity Theory: Moral responsibility ● Difference between intention and moral responsibility is one of point of views. ● Introducing causal responsibility computed by A computed by a model of A computed by an observer O prescribed role, reasonable standard reward inverse model
- 37. Simplicity Theory: Moral responsibility actualized emotion causal responsibility conceptual remoteness inadvertence + + – – for observer O attributed to A attributed to Afor observer O
- 38. Simplicity Theory: Moral responsibility actualized emotion causal responsibility conceptual remoteness inadvertence + + – – for observer O attributed to A attributed to Afor observer O ● From moral to legal responsibility: – equity before the law (e.g. the “death of a star” case)
- 39. Simplicity Theory: Moral responsibility actualized emotion causal responsibility conceptual remoteness inadvertence + + – – for observer O attributed to A attributed to Afor observer O ● From moral to legal responsibility: – equity before the law (e.g. the “death of a star” case) – law, as a reward system, defines emotion
- 40. Example 1: Negligent hunters Summers v. Tice (1948), 33 Cal.2d 80, 199 P.2d 1 Two hunters shot at the same time harming their guide.
- 41. Example 1: Negligent hunters Summers v. Tice (1948), 33 Cal.2d 80, 199 P.2d 1 they thought the harm was impossible Two hunters shot at the same time harming their guide.
- 42. Example 1: Negligent hunters Summers v. Tice (1948), 33 Cal.2d 80, 199 P.2d 1 they thought the harm was impossible but it was reasonable to consider the danger Two hunters shot at the same time harming their guide.
- 43. Example 1: Negligent hunters Summers v. Tice (1948), 33 Cal.2d 80, 199 P.2d 1 they thought the harm was impossible but it was reasonable to consider the danger therefore they're (morally) equally responsible. Two hunters shot at the same time harming their guide.
- 44. Example 1: Negligent hunters Summers v. Tice (1948), 33 Cal.2d 80, 199 P.2d 1 they thought the harm was impossible but it was reasonable to consider the danger therefore they're (morally) equally responsible. negligence Two hunters shot at the same time harming their guide.
- 45. Example 2: Navigating oil Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea.
- 46. Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. Example 2: Navigating oil
- 47. Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. with poor maintenance, sea contamination by oil leakage predictable Example 2: Navigating oil
- 48. Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. fire after oil leakage in sea difficult to occur Example 2: Navigating oil with poor maintenance, sea contamination by oil leakage predictable
- 49. Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. fire after oil leakage in sea difficult to occur therefore, defendant is not responsible Example 2: Navigating oil with poor maintenance, sea contamination by oil leakage predictable
- 50. Overseas Tankship (UK) Ltd v. Morts Dock and Eng. Co Ltd – “Wagon Mound (No. 1)” (1961), UKPC 2. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. fire after oil leakage in sea difficult to occur therefore, defendant is not responsible foreeseability Example 2: Navigating oil with poor maintenance, sea contamination by oil leakage predictable
- 51. Example 3: Navigating oil, continued Overseas Tankship (UK) Ltd v The Miller Steamship Co – “Wagon Mound (No. 2)” (1967), 1 AC 617. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. NEW EVIDENCE: flammable objects in the water.
- 52. Example 3: Navigating oil, continued Overseas Tankship (UK) Ltd v The Miller Steamship Co – “Wagon Mound (No. 2)” (1967), 1 AC 617. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. NEW EVIDENCE: flammable objects in the water. with poor maintenance, sea contamination by oil leakage predictable fire after oil leakage possible, because of flammable objects therefore, defendant is responsible 1st argument: foreseeability
- 53. Overseas Tankship (UK) Ltd v The Miller Steamship Co – “Wagon Mound (No. 2)” (1967), 1 AC 617. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. NEW EVIDENCE: flammable objects in the water. Example 3: Navigating oil, continued 2nd argument: weighting of risks (anticipations)
- 54. Overseas Tankship (UK) Ltd v The Miller Steamship Co – “Wagon Mound (No. 2)” (1967), 1 AC 617. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. NEW EVIDENCE: flammable objects in the water. risk Example 3: Navigating oil, continued 2nd argument: weighting of risks (anticipations)
- 55. Overseas Tankship (UK) Ltd v The Miller Steamship Co – “Wagon Mound (No. 2)” (1967), 1 AC 617. At a landing stage oil was spilled for days in the sea. It was then ignited during works on a ship nearby. NEW EVIDENCE: flammable objects in the water. risk risk as generalization of foreseeability: Hart and Honoré’s view! Example 3: Navigating oil, continued 2nd argument: weighting of risks (anticipations)
- 56. Conclusions ● Our contribution attempts to open an alternative research track for the computation of responsibility in AI & Law.
- 57. Conclusions ● Our contribution attempts to open an alternative research track for the computation of responsibility in AI & Law. ● Underlying model derived from general cognitive functions (SIMULATION, REPRESENTATION, REWARD INVERSE MODEL)
- 58. Conclusions ● Our contribution attempts to open an alternative research track for the computation of responsibility in AI & Law. ● Underlying model derived from general cognitive functions (SIMULATION, REPRESENTATION, REWARD INVERSE MODEL) ● It enables a smoother transition from moral to legal reasoning, and provides grounds to quantify legal concepts.
- 59. Conclusions ● Our contribution attempts to open an alternative research track for the computation of responsibility in AI & Law. ● Underlying model derived from general cognitive functions (SIMULATION, REPRESENTATION, REWARD INVERSE MODEL) ● It enables a smoother transition from moral to legal reasoning, and provides grounds to quantify legal concepts. ● Computation integrates quantitative and structural aspects: potential ground for unifying other approaches, e.g. exploiting explicit knowledge and probabilistic information. – further work is needed for a complete operationalization and for detailed comparisons