I am an old engineer from the age of Fortran, etc. When I worked at GM, I was well past the point of writing programs but knew enough about how to specify them. Fundamentals still matter! Rule one of specifying: create a model of the design first. In my case it was a logic flow chart. The model would include number of devices and failure modes of those devices. From this a Critical path can be determined and FEMA(failure modes and effects analysis) can be performed. In the case of AI it may be necessary to check to see if the system is learning the “wrong” things as it develops, and how it could affect the critical path. All this could have been done using statistical methods before deployment. All I saw was failure analysis after the fact, which makes the customer the tester.
Thank you @neural foundry. Really the research teams doing some really great work. Standing on shoulders of giants. The 2 main research papers referenced in the write up🙏.
Hahaha.... I would 100% fit in the same boat. The autopsy/audit stories I have, which would have easily been mitigated by meticulous design analysis... Thank you for the FMEA reminder. I'd forgotten that. We would loose our jobs "breaking things", then.
I am an old engineer from the age of Fortran, etc. When I worked at GM, I was well past the point of writing programs but knew enough about how to specify them. Fundamentals still matter! Rule one of specifying: create a model of the design first. In my case it was a logic flow chart. The model would include number of devices and failure modes of those devices. From this a Critical path can be determined and FEMA(failure modes and effects analysis) can be performed. In the case of AI it may be necessary to check to see if the system is learning the “wrong” things as it develops, and how it could affect the critical path. All this could have been done using statistical methods before deployment. All I saw was failure analysis after the fact, which makes the customer the tester.
Sorry FMEA, not FEMA.
Thank you @neural foundry. Really the research teams doing some really great work. Standing on shoulders of giants. The 2 main research papers referenced in the write up🙏.
https://youtu.be/SpPhm7S9vsQ?si=tMemnfxO_XCkuaPH
Hahaha.... I would 100% fit in the same boat. The autopsy/audit stories I have, which would have easily been mitigated by meticulous design analysis... Thank you for the FMEA reminder. I'd forgotten that. We would loose our jobs "breaking things", then.
https://substack.com/@interestingengineering/note/c-190744984?r=223m94