The TAMI Project is creating technical, legal, and policy foundations for transparency and accountability in large-scale aggregation and inferencing across heterogeneous information systems. We are outling an information architecture for the Web that can provide transparent access to reasoning steps taken in the course of data mining, and accountability for use of personal information as measured by compliance with rules governing data usage.
TAMI/e2esa face-to-face [agenda][minutes]
Meeting with iARPA. Presentations: Danny's, Lalana's, and Jim's
Submitted paper to IEEE Policy 2008 pdf
Tabulator extension release includes justification UI
Download and install extension
Begun work on Reciprocal Privacy for Social Networks
Scenario 9: MA Disability Discrimination
Detailed workthrough of scenario 9
Scenario 0: MIT Prox Card violation
6.898 Fall course on Accountability architectures for WWW started
Draft specification of AIR (Accountability in RDF) AIR ontology
Decided to move to a more AMORD like language with dependency tracking
First draft of Rei+ ontology
TAMI Architecture (pdf)
Developing Policy Aware Provenance design
Beginning work on scenarios 8, 9, 10
TAMI/e2eSA FTF meeting
Continuing work on Scenario 6
Scenario 4 is starting to take shape, as is the Scheme code that implements one of the reasoning engines.
We've begun work on a Data Purpose Algebra and are continuing to work on expressing Scenario 4.
Work continues on cwm/n3 and TMS-based reasoners, as well as Scenario 4 and the user interface of the project.
We participated in these events:
Integrating Cwm with Inference Web
We've decided to produce more complex scenarios in order to test/modify our design:
the rules of Routine Uses (a federal agency's statement of the terms and conditions for disseminating data from a particular system) and
the rules of a System of Records Notice (a federal agency's statement of terms & conditions for receiving data into a particular system)
We've presented a paper at the AAAI Spring Symposium "The Semantic Web Meets E-Government."
Daniel J. Weitzner, Harold Abelson, Tim Berners-Lee, Chris Hanson, James Hendler, Lalana Kagal, Deborah L. McGuinness, Gerald Jay Sussman, and K. Krasnow Waterman
We've started producing code:
Chris' code - using RDF. Written in a new notation (NS) embedded in MIT/GNU Scheme.
Carlos' code - using RDF and N3
We've agreed that we have enough common understanding to work through a scenario from end to end.
One significant challenge was the range of backgrounds necessitated by the project. In order to ensure a common knowledge base, we provided each other with overviews of relevant topics:
We created a fictional scenario that addresses some of the common public concerns. It involves an airline passenger who is a potential match in the testing of the Transportation Security Administration's Secure Flight program (formerly known as CAPPS); his identity is passed to the FBI's Joint Terrorism Task Force and, ultimately, he is arrested on an outstanding warrant for unpaid child support.
The scenario will allow us to test our ability to build a system that can proofcheck the answers to two important data mining questions:
The scenario was built specifically to require application of rules with three increasing levels of complexity:
Based upon current government efforts, we presume that the historical log of data collection, analysis, and transfer, as well as case activities, will exist in XML. Using our hypothetical, we created
Note 1: Where possible, we used the National Information Exchange Model, the joint Department of Justice and Department of Homeland Security XML for law enforcement.
Note 2: The two versions do not contain identical information. The "cleansed" view contains more of the required information.
We expect that the transactional data will be processed as RDF. A volunteer has produced an RDF version of the transaction data.
Chris Hanson has generated a skeletal RDF Schema for SORN documents and has used that vocabulary to create an example SORN. This uses an updated RDF/XML version of the above transaction data. K is drawing an RDF graph to demonstrate the SORN Schema.
We are operating on the assumption that the rules should be expressed in N3. This required us to build quite a bit of common understanding about how to convert law to rules to N3. This appears to be an iterative process. So far, we have:
Note: These rules will not answer either of our two goal questions, but this was an important first step in determining that we could convert any law into N3.
Note: This will create an opportunity to test a proof for our first goal - was an agency allowed to collect/acquire the data?
We have had quite a bit of discussion about what is the appropriate logic system for this sort of project.
Because of our Semantic Web interests, we are focused on CWM. We have:
We are contemplating using a TMS as the storage mechanism for our proofs and/or as a deductive reasoner. We have:
We are intending to register question answering systems used in TAMI in Inference Web and have those systems generate PML. Then we will use Inference Web to view and manipulate explanations.