In The Machine Murders, Manos Manu visits the Greek island of Mykonos to attend the wedding of a friend. But half the action happens elsewhere: at his work office in Singapore, a team of talented programmers led by Mei Ni attempts to use the software that Manu has created to draft a shortlist of suspects and then track their whereabouts around the island. Fiction or reality? Well, although MANU is fiction, similar machine learning technologies are now used by law enforcement agencies around the world tracking calls, clicks, online pauses and suspect behaviors.

In July 2019, the second Global Meeting on AI for Law Enforcement took place in Singapore and published “Towards Responsible AI Innovation”, a 56-page report on the use of AI for crime and the development of tools addressing “cloud crime”. “Perhaps the biggest eye-opener, in terms of the malicious use of AI, concerned advances in programmatically generated fake videos and images, or so-called ‘deepfakes’,” say its authors. That’s exactly what Manos Manu and his team face when somebody (no spoilers, here) uses the social networks' social graphs to create fake profiles and plant stories that incite the killer to perform the “buoy murders”.

Despite the ML algorithms developed by the networks to prove if a profile is real or not, when we’re online we’re not sure whether or not we’re interacting with actual people. And we will never be 100% sure. In their effort to stop “fake news”, social media companies keep addressing this issue; in one quarter alone, Facebook shut down 2.2 billion (with a capital B) fake accounts from its service. This is a slightly lower number than its 2.38 billion monthly active users. The use of these accounts has been already detrimental: unconstrained lies (with or without a specific purpose like that of the perpetrators in The MM) can overwhelm truth online as the illusory truth effect tells us that the more something is said, the more it’s believed. The perpetrators don’t need to target anyone in particular (no spoilers, again). But in a pool of two million people who take their vacation on the island of Mykonos in a given summer, people who possess “dark traits” in their personality may account for 100 persons.

Manos Manu is a brilliant programmer who knows how to deal with overfitting and how to include other parameters which minimise bias and noise: his list goes down to sixteen. And then…

Interpol’s report cites the tools used by the Australian Federal Police (AFP) such as a search engine system, akin to Google, for investigative data that will support them in making both sense and use of the large amounts of structured and unstructured data in their databases. If you think that the social media channels are not onboard such law enforcement efforts, you are mistaken. Since facing cyber crime and fake accounts is in the interest both of the social networks and of society itself - this probably leads to authorised (and unauthorized for Mei) access to all important datasets about the behavior of a set amount of users.

Adversarial AI and police actions to deter cyber crime on a global scale have just started to provoke interest around the world. The rules of privacy need to be respected but also, the capacity of Police Authorities to navigate these huge datasets is our only assurance that we will not experience crimes such as the “buoy murders” described in The Machine Murders. But it will take time for the game to unravel on all sides. Police err to the side of caution, while criminals do not. “As far as we are concerned, these events are unrelated. There are no machine murders. Only machine solutions,” says fictional Interpol Unit Leader Dr. Daniel Novak to Manos, at the end of the novel. And then he goes on: