Review of Undercover: The Expert Witness Rating *****
The movie Mercy starring Chris Pratt has a unique premise. In the film, Pratt’s character is secured to a chair — a chair designed to mete out a death sentence if he is unable to convince an AI judge of his innocence. Throughout the movie, the AI judge gives percentages of his guilt based on the evidence presented. This podcast reminded me a lot of that film.
I have no doubt that AI will be instrumental in the criminal justice system going forward. The Expert Witness tells the story of a software tool called Cybercheck, which claims to identify suspects in various crimes. Like the AI judge in Mercy, Cybercheck provides investigators with a percentage likelihood that the person it identifies is the correct suspect.
It all sounds very futuristic, and it was quickly adopted by numerous law‑enforcement agencies across the country. The podcast begins with several murder cases in Akron, Ohio. In a few of those cases, Cybercheck reports were offered as evidence. The creator of the software, Adam Mosher, even testified as an expert witness. But there are numerous problems with verifying the usefulness or accuracy of Cybercheck’s reports. The company behind the software, Global Intelligence, along with its creator Adam Mosher, is opaque about how the system identifies suspects. Mosher refuses to provide prosecutors or defense attorneys with source code or disclose the AI algorithms used. The claim is that the system uses open‑source data, including cell‑phone location information and wireless‑router data, to place a suspect at a specific location at the exact time a crime is committed.
That description alone seems far‑fetched. Phone companies don’t have location data floating unprotected on the internet, and the idea of scraping wireless‑router information is equally hard to fathom. Additionally, the software uses something referred to as “criminal indicators” to help identify suspects — things like if a suspect views porn, has a criminal record, or belongs to a gang. While there may be some limited validity to behavioral indicators, the podcast rightly points out that this approach introduces bias. The vast majority of suspects identified by Cybercheck are Black or minority.
As defense attorneys begin questioning this new technology, problems surface quickly. When any new scientific or technical method is used in criminal cases — such as when DNA evidence first emerged — the technology must undergo what is known as a Daubert hearing. Only then can it be admitted at trial. Cybercheck never went through a Daubert hearing.
Host Sam Mullins eventually interviews Mosher. Mosher tells him that early uses of the technology were “in error” and that Cybercheck should be considered intelligence, not evidence. But Mosher has other problems. Mosher claimed that the technology had been peer‑reviewed and even provided a report to support that claim. It turns out the “peer review” was actually conducted by a lone college student who had access to the software for only a brief period.
The real proof of Cybercheck’s validity would be clear examples where the tool helped solve a crime. The podcast team tries to find even one such case — and comes up empty‑handed.
This podcast grabbed me for several reasons: the criminal‑justice angle, the use of AI in crime solving, and the ongoing question of whether Cybercheck is a groundbreaking tool or total BS. By the end, I came away with the same conclusion as the podcast creators — Cybercheck is smoke and mirrors that has caused more harm than good.