Cold Cases in the AI era

The vibe, dear reader, is not cozy

I like Agatha Christie and the Knives Out franchise as much as anyone, but the “cozy murder” or “cozy mystery” genre resurgence should have ended with the Obama era. As the manhunt for the recent Brown University shooting rampage and the murder of MIT professor a few days later showed, unexpected human witnesses crack the case. A Reddit post can matter more than drawing room staging. There is nothing cozy about murder.

More appropriate is the cold case genre. Two reasons: AI and politics.

I found myself watching two gritty UK cold case TV series this month, when I was under the weather: the Netflix Dept. Q (2025-) (an adaptation of the Danish films) and Unforgotten (2015-). I hadn’t ever watched the CBS Cold Case series (2003-2010). I went back and sampled a few episodes before writing this post. The series uses clever flashbacks (period music, period clothing, nostalgic lighting) that give the audience direct access to the past. I think that’s a mistake.

The cold case procedural begins where most detective stories end. Everything has been processed, from the crime scene to the witness recordings. All evidence has been catalogued, boxed, and shelved. Years and sometimes decades have gone by. The original investigators are retired or dead. Something sparks renewed interest. A box comes off the shelf. The files, clippings, photographs, tape recordings, ashtrays, a mitten, other material objects, are examined anew.

Dept. Q

You can see the connection to AI. The human detective who opens the box and pulls out the thick folders doesn’t need more information. The point is to see what it all means. What is the detail that was missed? The pattern that wasn’t applied?

This is an era of abundant information and scarce judgement. The need is for that gut feeling, that last human bit of interpretation. Cold case investigators are last mile detectives. They take contradictory, time-degraded evidence and come up with a new account that can stand up in court.

It’s not a great genre for prestige TV — cold case work requires static office and computer search labor. But it can be made interesting. All good art is a matter of working within limitations. The past is a foreign country. You need guidebooks and to speak that language. So, the detective reads, stares at photographs, visits locations and stands there. There are visits to old age homes and long scenes of rambling stories. A cross attendant stands nearby, worrying the resident is getting too agitated. A random detail emerges: a word or phrase that sheds light on the file. Perhaps it was misheard or mispronounced. The data is the same but the pattern suddenly looks different.

My favorite cold case drama is Tom Stoppard’s Arcadia (1993). It’s a country house intellectual comedy cold case. The play alternates between Sidley Park in 1809 and the present day, where two researchers arrive to do research in the family’s archives. One is investigating the history of the estate’s gardens and the identity of a hermit who lived on the grounds. Another believes he has discovered that Lord Byron killed a minor poet in a duel on the grounds and subsequently fled England to escape the consequences. This case rests on a letter about a duel, evidence of Byron’s visit, and the sudden disappearance of the poet from literary history.

The case pivots on the assumption that the minor poet is not the botanist with the same name who later died of a monkey bite in far-off lands. The first researcher, however … well, go see the play.

What might a pro-model AI system have done with the same evidence? Likely it too would have followed consensus logic from its training data that a poet could not have left the humanities for STEM. Such is the brilliance of the late Tom Stoppard, whose play toggles back and forth between 1809 and the present (not flashbacks), so the audience can weigh interpretation against fact. Some mysteries are solved. But others will never be known. Was there really a last waltz? Does it matter?

THOMASINA: ....the enemy who burned the great library of Alexandria without so much as a fine for all that is overdue. Oh, Septimus! —can you bear it? All the lost plays of the Athenians! Two hundred at least by Aeschylus, Sophocles, Euripides — thousands of poems — Aristotle’s own library!....How can we sleep for grief?

SEPTIMUS: By counting our stock. Seven plays from Aeschylus, seven from Sophocles, nineteen from Euripides, my lady! You should no more grieve for the rest than for a buckle lost from your first shoe, or for your lesson book which will be lost when you are old. We shed as we pick up, like travellers who must carry everything in their arms, and what we let fall will be picked up by those behind. The procession is very long and life is very short. We die on the march. But there is nothing outside the march so nothing can be lost to it. The missing plays of Sophocles will turn up piece by piece, or be written again in another language. Ancient cures for diseases will reveal themselves once more. Mathematical discoveries glimpsed and lost to view will have their time again…

Stoppard’s play is describing a genealogy of knowledge that doesn’t require any individual or institution to hold everything. The cold case detective picks up what was dropped.

An AI system could flag every inconsistency in the file, cross-reference every name against every database, and surface every connection. AI tools may get very good at generating reconstructions, producing multiple plausible narratives, complete with visualizations. The cold case genre is made for this moment. The detective’s job would be to evaluate competing narratives, to identify the reconstruction that is true, distinct from the one that is merely plausible. But ultimately it takes haphazard events to tie things together. A person who saw a random thing and decided it was important.

Recent events show the awfulness of watching the construction of a murder narrative in real time. Names emerge on social media. People pounce. Innocent people are caught up in churning events. Nothing about it is satisfying as real life or a genre. The cozy mystery genre tends to overlook all that. It doesn’t care about picking up the pieces of a broken life. The cold case genre is all about the pieces of a broken life. No one is paying attention.

The cold case genre delivers a kind of satisfaction more appropriate to the AI era than the cozy mystery genre, first because of the forensic software tech, the genealogy databases, document analysis systems, etc., and second, because audiences want stories about human judgment, about deciding what matters and what does not. The satisfaction is in closing an account that has been open for too long. Things will be re-torn and repaired.

Sci fi like Minority Report assumed that AI tools would prevent crime. Current AI tools are being used to track and identify every single person inside the U.S. borders. Whether you agree that this is a good thing or not, it is happening while hundreds of thousands are huddled under quilts with cocoa watching Only Murders in the Building.

How can I not like the cozy mystery genre, you ask? Like much of the detective genre it treats homicide as an aberration in an otherwise functional society. But in the cozy iteration, everyone is well dressed and the furniture is elegant. It is a subgenre of reassurance, fitting the optimism of the Obama years where order is inevitably restored and the disruption is merely a temporary glitch in a “feel good” world. That world is gone. The cold case genre acknowledges that the system failed and the truth was shelved. This is the hard looking back required in the present moment. There are cracks in the foundation that allowed the file to gather dust for decades. The detective does not restore a social equilibrium. There remain unopened files of shattered lives.