AI and safeguarding critical infrastructure

Team 3 looks at AI for monitoring and protecting various infrastructure. I see that there are three groups of applications.
The first one is the use of Computer vision to augment video collect and labelling. Doing this on both waste trucks to detect roadside problems and tracking people on surveillance cameras seems dangerously dystopian (even having "optional" privacy features). Team 3 discuss what I see as the biggest problem which is having all of the data with personal information (faces with times and place). As both papers suggested having edge on device inference would mean that only "safe" data would ever leave the device (for example "broken stop sign at 11 Brent road"). It is much easier to manage a dangerous thing (personal data) if it doesn't exist in the first place.
The second group is that of automated monitoring of large infrastructure sensor suites. It seems quite reasonable that using AI to monitor these vast data streams would be more effective than classical statistical methods or manual human work. Compared to these other two areas it would seem much safer to implement these assuming we have valid assurance performance is increasing.
The last are is that of using AI modelling capabilities to understand climate change and sea level rise. It seems that these systems are effective. However given the complexity of any climate modelling it seems to be of great risk these systems predict/suggest either too big of a response, or too small of an response. Both of these could have large real world costs.

Some of the question I would have are:
- What size model was used for the detection of roadside problems and were these models small enough that it would be plausible to use them on edge devices.
- What were the absolute response times for the cybersecurity paper. IS it on the scales of days, hours, minutes.