Originally Posted by
esscee
Any computer is only as good as the input information to it plus its own "100%" solid safe operating software. You only have to consider the situation of Windows 10 and the many problems of nearly every "Update".
Great point. Even Machine Learning, a subset of AI, where the system/computer/machine learns over time to improve performance with experience, errors are made--potentially fatal. In supervised ML all instructions are explicitly programmed in, and then the machine can use preset parameter weights to learn, in semi-supervised learning, there is a balance between pre-programmed instructions and flexible learning parameters, and then there is unsupervised ML; this was very useful in uncovering genes we did not know existed during the Human Genome Project. Theoretically, yes these ML algorithms can save lives or land with greater ease under some circumstances, but no one knows in advance, and at least with Boeing's recent failures (787. 737 Max) we should all be more skeptical of so-called capabilities of new smart-autonomous systems. In research with my team: biometrics, AI financial/stock market analysis, autonomous car research, and cybersecurity--so much goes wrong in the testing, a lot we did not even anticipate. Speaking with my pilot friends here on PPRUNE and elsewhere, split-second decisions on an aircraft are not only complex but often life and death.