Interview
Jeff Jonas, IBM Fellow: Talks at GS Session Highlights
- Jeff Jonas serves as a Chief Scientist at IBM and is one of only 92 IBM Fellows globally, leading work in context computing.
- His early career was opportunistic, driven by specific client problems rather than a fixed roadmap, notably a request from casino executives to prevent transactions with individuals on exclusionary lists.
- Jonas delivered the NORA (Non-Obvious Relationship Awareness) system to the casino industry within a 90-day deadline, treating operational systems like job applications, credit enrollments, and hotel reservations as sensors.
- The NORA system successfully identified 24 professional cheaters utilizing different identities and dates of birth, a capability that attracted the attention of In-Q-Tel.
- This initial success led to the development of systems capable of processing tens of billions of records across thousands of data sources.
- Approximately six to seven years ago, Jonas initiated a "skunk works" project code-named G2, operating secretly for two and a half years with a small team.
- The G2 platform is designed to be agnostic to object type (people, cars, asteroids) and capable of simultaneous processing in multiple languages including Chinese, Thai, Korean, English, and Spanish.
- The technology has been deployed in "sea trials" to modernize voter registration in the US, protect the Malacca Strait for Singapore, and reduce false positives in Anti-Money Laundering (AML) systems.
- Jonas distinguishes his NORA-class technology from the systems used in the film 21, noting they are related but distinct methodologies.
- His awareness of privacy issues evolved during a DARPA program on Total Information Awareness with John Poindexter, shifting his focus from purely detection to sustainable, privacy-conscious system design.
- Jonas defines his core privacy strategy as "avoiding consumer surprise," arguing that systems should not create scenarios likely to become negative headlines.
- He advocates for "Privacy by Design," insisting that every record must have a full pedigree and that no two records should merge without traceable attribution.
- This concept of "data tethering" ensures that adding, changing, or deleting a record propagates accurately through the ecosystem to prevent harm to innocent individuals who have been incorrectly flagged.
- A key tension identified by Jonas exists between transparency and security; while both governments and privacy advocates hate false positives, overly transparent data availability can hinder the detection of sophisticated adversaries.
- The G2 technology was adapted for space exploration by University of Hawaii astronomers to solve an $n$-squared problem regarding asteroid collision prediction.
- Astronomers previously required 10 million computer hours to forecast asteroid interactions over 25 years, whereas Jonas's "space-time box" method reduced this to 2,800 compute hours.
- The "space-time box" methodology groups celestial objects by proximity and time, isolating pairs to within 0.05 AU and reducing the final computational load from 10 million hours to 15 minutes.
- In June, astronomers validated the method by observing two asteroids passing within the diameter of Earth's orbit, a first for which the model provided a forecast.
- Jonas asserts the principle that "data beats math," positing that once mathematical precision plateaus, the solution lies in widening the observation space with additional data rather than increasing algorithmic complexity.
- He argues that reducing error rates by half is often impossible with current math unless new data sources are integrated into the transactional or observational ecosystem.
- Jonas describes his primary contribution as "embarrassingly simple" heuristics that allow heavy mathematical lifting to be applied only to high-probability events.