Person of Interest - an action sci-fi show from 15 years ago - raises questions about AI and surveillance we still need to urgently grapple with

📺 Watched Person of Interest.

It took a while, but I finally got to the end of this 5 season sci-fi-ish thriller show that started in 2011.

Note the year. It was before the Snowden surveillance revelations of 2013. It was before the rise of LLM AIs and their contemporary use in suggesting which people should be arrested or killed.

Although actually the AI that dominates this show is a good AI, at least somewhat. The story goes that the government had a couple of computer geniuses invent a supercomputer that could accurately predict where and when terrorist incidents would take place and who the culprit would be.

The downside is that it requires constant mass surveillance of the entire population. The show started,of course, a few years after the post 9/11 “war on terror” began, so this didn’t seem like an unrealistic aspiration. One difference between that AI and current ones are that it basically doesn’t seem to make a mistake. But the authorities understand how the public might feel about being constantly spied on so it’s largely a secret.

One day, one of the inventors of the system realised the machine was outputting information not only about terrorist incidents, but violent deaths in general. The government, obsessed with terrorism, did not care to use that info so it was discarded as irrelevant.

Events, covered in later episodes of the show, transpired and the tldr is that one of the creators - perhaps not the one you might think - ended up putting in a backdoor to retrieve the identifiers of these “irrelevant” people and hiring an action-movie type guy with little to lose to go out and prevent them.

One idiosyncracy of the machine though is it can’t say whether the person it identifies is the one who is going to cause the violent death, or the one who will die, which introduces some interesting nuance. It isn’t brought up but it only seems to spit out about one person every few days, unless Finch is secretly ignoring the vast majority, so I guess it’s not all that comprehensive. But it’s basically always correct.

The earlier series of the show are mostly one-off episode stories where the machine identifies a number and the team tries to stop the forthcoming death. Later series introduce more of a longer-term storyline, with the main plot only somewhat featuring the one-off saves.

Whilst primarily a hacker/action series, some moral concerns we need to think about today are introduced. What is the trade-off between constant mass surveillance and “predictive policing” even if, unlike today’s version, it worked well? Do the public have the right to know? If they object, what then? Does the state need this power? Should corporations have it? But what if it is only a handful of private sector big tech companies that can provide it? Do the means justify the ends? How can we control the controllers of things we don’t understand? To what extent should the creators of such technology be allowed to enhance or restrict its usage? Are people capable of changing their predicted behaviour? The questions go on.

AI beyond human comprehension is already deciding, or at least influencing, decisions around who should live or die. The authorities try to algorithmically predict crimes. Private companies such as Flock constantly monitor people via camera, big tech companies silently surveil our behaviour, all feeds that the government has access to whether legally or not. Vast amounts of the public’s sensitive data are being given to extremely shady companies like Palantir who trick their way into inducing locked-in state dependence.

The show thus feels only a little further into the future than today, except perhaps the part where the AI is (almost?) never wrong. But even if in a world where that was true, all the above questions remain and need to be addressed urgently.