Oireachtas Joint and Select Committees
Tuesday, 13 February 2024
Joint Oireachtas Committee on Justice, Defence and Equality
General Scheme of the Garda Síochána (Recording Devices) (Amendment) Bill: Discussion
Ms Olga Cronin:
I will speak to the point about NIST and the most recent point. Mr. McGarr will talk to the law points.
It is important to note that the figures from NIST are the go-to when we talk about how accurate this technology is. NIST figures have been mentioned in the Dáil in that respect, but it is very important that when we are told there is an accuracy figure from NIST, and we should use it based on a figure of, let us say, 98%, we have to ask exactly how that testing was carried out. For example, we heard NIST figures being used in the Dáil to support bringing in facial recognition technology, but those figures related to a very clear mugshot compared with another very clear mugshot. The figures did not involve the use of images taken from CCTV, otherwise known as taken from the wild. Image quality obviously has a massive role to play in how accurate this will be. We have seen situations in America where some police forces have superimposed a cheek, chin, frown or open eyes just because that was missing from an image taken from CCTV. Those NIST figures should be interrogated accordingly. I urge the committee to do so. It is also important to note that NIST stated that the accuracy figure could worsen in excess of 20% when it relates to images taken from CCTV.
It is also important to note with regard to NIST testing that if at least one of the results returned from a facial recognition technology, FRT, search is a match for the probe image, the search is considered successful and counted as part of the true positive match. For everyone in the room, it is very important to know that when an FRT search is carried out in the manner the Bill seeks to provide for, there will be a list of candidates.
That is a list of people with a percentage score next to it, like a similarity score. Let us say there is someone who looks exactly like me or very similar to me and we are in the same database. Let us say she commits a crime and is caught on CCTV. A picture of her is taken and it is run through an FRT search. It is not guaranteed she will be first on the list and I will be No. 6 or 7, so when we talk about accuracy figures there is a lot to unpack there. The committee should take time to really get to grips with that.
The other point was on-----
No comments