Uploaded July 2022 | Updated September 2026, 9 hours ago
Abstract: The modern “toolkit” enabling AI and Robotics has been growing at an accelerated rate over the last decade, in sync with the explosive growth of compute, storage, data and applications. While our industry continue to get more and more productive leveraging and expanding this toolkit, our ability to contain and limit how these systems fail, misbehave, or are manipulated - due to unforeseen circumstances (say, domain drift) or attack vectors (say, adversarial actors) has not grown at the same explosive rate.
As the gap between our functional capabilities and closed-form side-effect predictability continues to expand, the damage the systems we build could cause continues to grow. In this talk, I’ll discuss the Swiss Cheese model, a mental model and a set of mechanisms that helped make Aerospace the safest mode of transportation today - especially challenging given Aerospace is also the most complex system-of-systems ever productized, with the most exposure to mode confusion compared to any other commercialized product. The success of Aerospace (from a Safety POV) should be an inspiration to other domains that have similar potential to do harm.
I often refer to today’s AI and Robotics toolkit as full of “sharp knives” - a sharp knife in the hands of a great chef will allow them to create amazing food, but in the hands of an enthusiastic amatour or an adversarial party can cause unintended (or intentional) harm; in my opinion, for one to be an “expert practitioner” one must not only deeply understand how a tool is to be used, but also how it could fail, get misused or abused.
Bio: After a career spanning 5 startups and senior executive/technology positions at Microsoft and Amazon, Gur co-founded the Amazon Prime Air program, where he and his team designed an autonomous, independently-safe UAS (Unmanned Aircraft System) that enables certifiable, safe, fast, scalable and NextGen-integrated delivery services; on Aug 31st 2020 the Prime Air technology & service was granted an FAA Part 135 Air-Carrier certificate, enabling commercial BVLOS (beyond visual line of sight) autonomous operations in the US. More about the technology behind Prime Air can be found at youtube.com/watch?v=yHY-ZWpC8T4
Gur has a broad set of research and engineering interest and experiences – spanning formal software methods, semantic, probabilistic and hybrid AI systems, model-based engineering & systems theory, machine vision (both classic e.g. photogrammetry and ML-based), high-availability & -scale services, audio processing (coding, beam forming, reconstruction, text to speech etc), custom hardware an(esp. Non von Neumann architectures), Safety engineering in the realm of AI, and technical standards.
If you have a modern GSM phone, you are using an architecture & technical standards that Gur helped develop. Prior to Prime Air, Gur helped built technology that was used to carry ~10% of the world’s international phone calls (“toll bypass”,, 1997), designed and operationalized one of the highest-scale web-services (then) in existence at 1.5 million transactions per second (2004), architected and operationalized the first automated capture-to-textured-metric-3D model geoprocessing system (including the world’s highest-end metric aerial camera system) in 2006, and designed the Federated Airspace Management Model that is emerging as the foundational technology to enable autonomous operations within NextGen (2015). Along the way, Gur filed hundreds of patents, at an average rate of ~20 per year. Gur started his engineering career at a young age – selling his first program (for a Sinclair Spectrum) at age 12.
Gur caught the aerospace bug early on from his father, who was a commercial (707, 747, 777) and military pilot, is a current Israel DoT Safety Board investigator, and homebuilder of Experimental designs (BD-5, Rutan LongEZ, and others). While piloting was fun, designing, understanding the physics of, building and automating airplanes proved to be much more interesting, starting a career arc that resulted in Amazon Prime Air.
In early Sept 2020, Gur retired from Amazon and is pursuing a range of interests across multiple organizations, such as new propulsion technologies, novel aircraft configurations, novel AI Hardware, and Systems-Architectures and Algorithms that drive forward the state of the art in deterministic and certifiable AI; related to this last effort, Gur was selected as an SME to assist NASA Aeronautics in developing the agency’s long-term AI strategy YouTube Video Presentation on "Moving Forward Safely: Amazon's Approach to Drone Delivery: mail.google.com/mail/u/0/?tab=rm#sent/QgrcJHrnsbWJSjVCZWljXWqRzCmwzrHZVWg?projector=1
Abstract: The modern “toolkit” enabling AI and Robotics has been growing at an accelerated rate over the last decade, in sync with the explosive growth of compute, storage, data and applications. While our industry continue to get more and more productive leveraging and expanding this toolkit, our ability to contain and limit how these systems fail, misbehave, or are manipulated - due to unforeseen circumstances (say, domain drift) or attack vectors (say, adversarial actors) has not grown at the same explosive rate.
As the gap between our functional capabilities and closed-form side-effect predictability continues to expand, the damage the systems we build could cause continues to grow. In this talk, I’ll discuss the Swiss Cheese model, a mental model and a set of mechanisms that helped make Aerospace the safest mode of transportation today - especially challenging given Aerospace is also the most complex system-of-systems ever productized, with the most exposure to mode confusion compared to any other commercialized product. The success of Aerospace (from a Safety POV) should be an inspiration to other domains that have similar potential to do harm.
I often refer to today’s AI and Robotics toolkit as full of “sharp knives” - a sharp knife in the hands of a great chef will allow them to create amazing food, but in the hands of an enthusiastic amatour or an adversarial party can cause unintended (or intentional) harm; in my opinion, for one to be an “expert practitioner” one must not only deeply understand how a tool is to be used, but also how it could fail, get misused or abused.
Bio: After a career spanning 5 startups and senior executive/technology positions at Microsoft and Amazon, Gur co-founded the Amazon Prime Air program, where he and his team designed an autonomous, independently-safe UAS (Unmanned Aircraft System) that enables certifiable, safe, fast, scalable and NextGen-integrated delivery services; on Aug 31st 2020 the Prime Air technology & service was granted an FAA Part 135 Air-Carrier certificate, enabling commercial BVLOS (beyond visual line of sight) autonomous operations in the US. More about the technology behind Prime Air can be found at youtube.com/watch?v=yHY-ZWpC8T4
Gur has a broad set of research and engineering interest and experiences – spanning formal software methods, semantic, probabilistic and hybrid AI systems, model-based engineering & systems theory, machine vision (both classic e.g. photogrammetry and ML-based), high-availability & -scale services, audio processing (coding, beam forming, reconstruction, text to speech etc), custom hardware an(esp. Non von Neumann architectures), Safety engineering in the realm of AI, and technical standards.
If you have a modern GSM phone, you are using an architecture & technical standards that Gur helped develop. Prior to Prime Air, Gur helped built technology that was used to carry ~10% of the world’s international phone calls (“toll bypass”,, 1997), designed and operationalized one of the highest-scale web-services (then) in existence at 1.5 million transactions per second (2004), architected and operationalized the first automated capture-to-textured-metric-3D model geoprocessing system (including the world’s highest-end metric aerial camera system) in 2006, and designed the Federated Airspace Management Model that is emerging as the foundational technology to enable autonomous operations within NextGen (2015). Along the way, Gur filed hundreds of patents, at an average rate of ~20 per year. Gur started his engineering career at a young age – selling his first program (for a Sinclair Spectrum) at age 12.
Gur caught the aerospace bug early on from his father, who was a commercial (707, 747, 777) and military pilot, is a current Israel DoT Safety Board investigator, and homebuilder of Experimental designs (BD-5, Rutan LongEZ, and others). While piloting was fun, designing, understanding the physics of, building and automating airplanes proved to be much more interesting, starting a career arc that resulted in Amazon Prime Air.
In early Sept 2020, Gur retired from Amazon and is pursuing a range of interests across multiple organizations, such as new propulsion technologies, novel aircraft configurations, novel AI Hardware, and Systems-Architectures and Algorithms that drive forward the state of the art in deterministic and certifiable AI; related to this last effort, Gur was selected as an SME to assist NASA Aeronautics in developing the agency’s long-term AI strategy YouTube Video Presentation on "Moving Forward Safely: Amazon's Approach to Drone Delivery: mail.google.com/mail/u/0/?tab=rm#sent/QgrcJHrnsbWJSjVCZWljXWqRzCmwzrHZVWg?projector=1
