Master thesis: Formalization of Attribution of Knowledge to Artificial Agent and their Principals
Attribution of knowledge to artificial agents and their principals has been presented by Samir Chopra and Lawrence White in 2005. They define the principal of an agent as its employer or any other legal person engaging the agent to carry on transactions on its behalf. The principal question is to decide when we can say that an artificial agent knows a proposition and that its principal can be attributed this knowledge. So, this implies some legal and philosophical view of the subject.
We want to formalize this theory and apply it to the two cases presented in section VI.
- Attribution of knowledge to artificial agent and their principals [Samir Chopra and Lawrence White]. This paper introduces a pragmatic analysis of knowledge attribution. Its application is done in the legal theory of artificial agents and their principals.
- Privacy and Artificial Agents, or, Is Google Reading My Email? [Samir Chopra and Lawrence White] This paper illustrates how the concepts of legal agency and attribution of knowledge gained by agents to their principals are crucial in understanding whether a violation of privacy has occurred when artificial agents access user’s personal data.
- Cognitive automata and the law [Giovanni Sartor]. This paper discusses the possibility to attribute mental states to artificial agent with implications in various areas of the law.
- Papers visited in the course “Selected topics in Artificial Intelligence”:
Cohen and Levesque in their paper “Intention Is Choice with Commitment” present the logic of belief, goals and intention.
Nicoletta Fornara and Macro Colombetti give a “Commitment Based Approach to Agent Communication”.
Eduardo Alonzo proposes a language to describe rights in his paper “Right and argumentation in open Multi-agent Systems”.
How to formalize Attribution of knowledge to artificial agents and their principals?
Here is a list of concepts (elaborated after reading Chopra and White papers) that might be incorporated in the formalization: Agent, Knowledge, Proposition, Belief, Intention, Obligation, Permission, Accessibility, Authorization, Duty, Communication, Ability, Right, Privacy, Violation, Agency, Power, Reading, Autonomy, Liability.
Sub questions:
-How to represent agents and principals?
-How to model the state of agents/principals at a certain time t?
-How to model the behavior of agents/principals (How they change their states)?
-How to model the point of view of lawyer in attribution of knowledge? (For example right, duty, privacy…)
People are frequently involved in online transactions with software applications like Amazon, eBay and others. It is important to know what can happen in case of litigation between a customer and Amazon for example. The resulting formalization could be helpful:
-To make knowledge attribution more precise (make some distinction in knowledge attribution in other to decide when a principal can be attributed the knowledge of its agents)
-As a first step for agents to reason about knowledge attribution.
-As a requirements to agent software engineering.
-Define an ontology that will be necessary to represent information subject to our work (agents/principals in their states and concepts presented in III).
-Decide which of “epistemic logic”, “doxastic Logic”, “temporal logic” and “communication based on commitments” will be appropriate the world of agents/principals as presented background papers.
Concepts |
Alternatives |
Why? |
|
|
|
Agents |
logic |
To model and analyze systems |
Proposition |
Propositional logic |
Uses "proposition" |
Knowledge |
Epistemic logic |
Uses "Knowledge" |
Belief |
Doxastic logic |
Uses "Belief" |
Intention |
Doxastic logic |
Uses "Intends" |
Obligation |
Deontic logic |
Uses "Obligation (O)" |
Accessibility |
Epistemic logic |
Uses of Kripke Model |
Authorization |
Deontic logic |
- |
Communication |
Epistemic logic |
Uses of Epistemic Updates |
Alternating-time Temporal Epistemic Logic |
Uses a semantic that contains epistemic accessibility relation for each agent. Could also be apply as Knowledge Precondition |
|
Commitments based approach |
States with respect to a state diagram clearly define |
|
Agency |
Alternating-time Temporal Epistemic Logic |
Integrates Agency |
Liability |
Deontic logic |
- |
Tab1: Concepts and Alternatives
Requirements |
Formalism |
|
|
State of Agents |
Epistemic logic (what agent knows) |
Doxastic logic (what agent beliefs or Intends) |
|
Deontic logic (What agent is obliged or authorized to do) |
|
Communication between agents |
Epistemic Updates |
Alternating-time Temporal Epistemic Logic (Common knowledge, distributed knowledge, corporations modalities) |
|
Commitments approach (assertives, directives, commissives, declarations, proposals |
|
Temporal aspects |
Alternating-time Temporal Epistemic Logic (System transition function) |
Commitments approach (Temporal propositions) |
Tab2: Requirements and formalism
-Use the previous to build formalism.
-Apply the formalism to some cases.
Problem 1: Amazon case from Samir Chopra and Lawrence White, 2005
A friend wants to buy me a book as a gift. He asks for my shipping address so that he can send me the book. I direct him to my wish-list at Amazon.com saying “Amazon knows my shipping address”. When my friend has decided which book he want to buy, he pays and picks the shipping option. Amazon generates a shipping invoice complete with shipping address. Is Amazon liable if the book is not ship to my address?
Problem 2: Google case from Samir Chopra and Lawrence White, 2007
Users of Google Gmail received frequent advertisements concerning words contained in their emails.
The collection and use of information is done by programs such as Google’s AdSense scanning technology, which when applied to Google’s Gmail system leads to the generation of advertisements that are relevant to identified keywords in message bodies.
Google argues that their artificial agents know what you are talking about in your email, but there is no violation of privacy because humans are not involved in that process. Here is one case where knowledge gained by agents is not attributed to their principal.
We want to define the formalism and apply it to the two examples presented in section VI.
The following concepts will not be considered in the formalism: Autonomy, Privacy, Violation, and Reading.
These others will not be taken into account at the first step of the formalization: Duty, Ability, Right, Power, and Permission. The reason is that they need deep information about law that cannot be gained during this thesis.
-The formalism should include concepts that have been listed in the scope of the thesis.
-Model the two examples of section IV with the formalism.
-The theory is ambiguous
-Trying to merge different logics may give problems
-The formalism may be abstract.
First month (18/02-19/03): Read background papers concerning Attribution of Knowledge and possible formal languages (BDI, ATL)
Second and third months (20/03-21/05): Decide which formal logic will be used to represent our requirements. Define the ontology and model our examples.
Fourth month (22/05-23/06) : Start writing (first draft).
Fifth month (24/06-25/07) : Final sessions (Final version of the thesis)