I vividly recall sitting glued to the screen during the Hollywood mystery thriller Knives Out, completely absorbed in solving the case as if it were my own. Just like Detective Blanc’s team questioning every individual at the Thrombey Mansion, I mentally eliminated suspects, only to bring them back into consideration after each unexpected plot twist. At the time, it never occurred to me that this classic whodunit was actually prompting me to do calculations in my head. While it might sound like a stretch, I firmly believe that Benoit Blanc’s method of investigation closely resembles Bayesian Inference. But anyone who recalls the interrogation scenes in the film will immediately notice that Benoit Blanc wasn’t even conducting the questioning himself. He sat quietly beside a piano, allowing his team (Lieutenant Elliot and Trooper Wagner) to handle the inquiries. So then, why do I claim that Blanc’s investigative approach had anything to do with Bayesian Inference? Blanc himself explained this during the movie, and I quote:
“I observe the facts without biases of the head or heart.” (Benoit Blanc, Knives Out [1])
This perfectly captures the core principle of Bayesian Inference, where your conclusions are guided by evidence rather than gut feeling. Let’s work through this murder mystery together using Bayesian Inference.
A brief note before we dive in. Throughout the film, contradictions appear in two different forms. There are contradictions presented through flashbacks, which are shown exclusively to the audience and remain mostly unknown to Blanc. Then, there are contradictions uncovered through verbal inconsistencies that Blanc directly observes during the investigation. Therefore, we’ll concentrate only on the verbal inconsistencies that Blanc personally notices.
Furthermore, a note on how probability weights are assigned and adjusted. These aren’t calculated using the formal Bayesian formula, since likelihood values are nearly impossible to assign to behavioral clues such as acting evasively or being dishonest. Instead, we’re using reasoned approximations as a learning tool rather than a mathematical demonstration. So, I hope you enjoy this exploration.
Setting the Stage — Establishing the Initial Beliefs
Detective Blanc was hired anonymously by someone in the family to look into the possibility that Harlan Thrombey was murdered. When his team begins the interrogations, Blanc quietly watches the potential suspects from the sidelines. When the questioning drifts in the wrong direction, he gently steers things back on track with a tap on a piano key.
He notices that each conversation is tangled with lies and contradictions. What he does right is refusing to dismiss a story as completely false while clinging to another based on instinct. He recognizes that even misleading statements might hold pieces of truth. He carefully evaluates each exchange, assigns significance to each observation, and then weaves them together to reach a conclusion. He starts with uncertainty but gradually builds toward the most likely truth, keeping his own biases out of the equation.
Blanc starts by outlining the likely causes of death. In Bayesian terminology, this is known as a Prior Model. A prior model is the collection of assumptions we hold before we have any evidence to go on. In this case, the prior model is the initial set of theories about Thrombey’s death before the investigation even begins.
Assessing the Completeness of Initial Beliefs
Let’s examine our initial beliefs to see if we’ve missed any other possibility. Have we overlooked the chance that this was an attempt to frame someone? If that’s the case, should we add that as a sixth hypothesis?
This is where the most critical rule (the MECE Principle) for constructing hypotheses in Bayesian Inference comes into play. Every hypothesis developed as part of Bayesian Inference should be Mutually Exclusive and Collectively Exhaustive (MECE).
Let’s reconsider the sixth potential hypothesis, ‘Attempting to Frame Someone.’ While the chosen hypothesis should explain what might have caused the death, this particular theory speaks more to the motive behind the death, assuming it’s proven to murder. So, it violates the mutual exclusivity requirement of the MECE principle and therefore can’t stand as a direct hypothesis.
Assigning Probabilities (Prior Probabilities)
Let’s continue with the hypotheses we established earlier, since they account for all possible causes of death (collectively exhaustive). The next logical step is to assign probabilities to our initial beliefs. This means we begin with an informed estimate about how likely each hypothesis is to have caused Harlan Thrombey’s death. Since we assign probabilities before we have any direct evidence or data, we refer to this as the prior probability. The visual below demonstrates us assigning equal weights to all hypotheses. Let’s take these as our prior probabilities for now.
A question that naturally arises is whether each hypothesis carries the same chance of occurring. No, not necessarily. It’s a common misunderstanding in Bayesian inference that we must assign equal probability to all hypotheses. In the absence of any prior evidence, we assume that Detective Blanc gives equal probability to each hypothesis. But that’s not always how it works.
We might also assume non-uniform (unequal) probabilities if we have prior knowledge indicating that a hypothesis is more likely than the others. General crime statistics can also be helpful for estimating prior probabilities. For example, according to FBI homicide data [2], it’s widely noted that in most homicides, victims know their killer. Homicides carried out by an outsider often require a motive involving burglary or some form of revenge. Therefore, H4 receives more weight, since family members had greater access to the victim. Moreover, in Harlan Thrombey’s case, the theory that a family member caused his death carries greater weight because his family members could be motivated by the prospect of inheriting his wealth and estate. The ideal prior probabilities in our scenario would be an unequal distribution.
Updating Probabilities based on Evidence
Let’s try to recall the scene where Marta is being interrogated. Marta has a peculiar condition that causes her to vomit whenever she lies. But since Marta initially believes that she caused Thrombey’s death by accidentally switching medications, she handles the situation by providing incomplete
The real twist? Detective Blanc already knows about her condition. Do Marta’s inconsistent answers raise red flags and shift the weight of suspicion? On one hand, Marta might have had a reason to want Harlan dead (supporting the outsider theory – H5). On the other hand, maybe Marta, as the nurse, accidentally made a fatal error that led to Mr. Thrombey’s death (H2). This is where the Bayesian Likelihood Function becomes incredibly useful. It evaluates how well each hypothesis accounts for the evidence we’ve gathered. Marta’s behavior alone isn’t strong enough to clearly favor H2 over H5, so the probability shifts only slightly. Both H2 and H5 see small increases, while H1 and H3 see slight decreases.
One key thing to remember about probability: once we receive any piece of evidence—big or small—and update our beliefs accordingly, we are now working with what’s called “posterior probability.” Based on the analysis above, we’ve adjusted the probabilities as shown below.
From the chart, you can see that the weight has shifted a bit toward H2, but nothing dramatic has happened yet.
Clear and Direct Contradictions — A Bayesian Goldmine
A major inconsistency emerges about who stood next to Harlan Thrombey during his birthday celebration. Linda, Harlan’s daughter, claims she was beside him along with her husband and son. Meanwhile, Walt insists that he and his family were the ones next to Harlan. While this doesn’t point the finger at any single person, it casts doubt on the entire family’s credibility, boosting the likelihood of H4.
Here are the updated probabilities:
Walt Points the Finger at Ransom
Lieutenant Elliot questions Walt about why Harlan pulled him aside during the party and why Walt seemed visibly upset afterward. Walt hesitates, then quickly redirects suspicion toward Ransom, claiming Harlan had argued with him. This deflection strongly suggests Walt is hiding something about his own conversation with Harlan. Let’s update the probabilities based on this new clue.
Mother-Daughter Story Doesn’t Add Up
When Blanc’s team asks why Joni arrived early, she says she needed to speak with Harlan about wiring her daughter’s school fees. But Meg, Joni’s daughter, contradicts this—saying her grandfather never once failed to send the money on time. This glaring inconsistency significantly strengthens the case for H4.
The Will Reading — Sharpening Our Theory
Up to this point, H4 (murder by a family member) has held the highest probability. But when the will reveals that every asset goes to Marta, the nurse and caretaker, all eyes suddenly turn to her. The likelihood of H5 nearly triples after this bombshell. The family now suspects she manipulated Harlan into changing his will. Updated probabilities are shown below.
This moment introduces a crucial idea: hypothesis refinement. Bayesian Inference doesn’t force you to stick with your original theories—it allows you to dig deeper and create more specific sub-hypotheses as new evidence appears. Originally, H5 was a broad category. Now, we can define a narrower version. Our updated list of hypotheses and their weights appear below.
Overnight, the family that once trusted Marta now sees her as the main suspect. But Detective Blanc isn’t convinced she had a motive—the toxicology report shows no morphine overdose. While the family reacts emotionally, Blanc follows the evidence, which leads him in a different direction: straight to Ransom.
The Climax — Evidence That Changes Everything
Throughout the investigation, nearly every family member—and even the household staff—mentions that Ransom had a bitter falling-out with Harlan and stormed out of the party early. His absence the day after Harlan’s death adds to the suspicion. Still, the motive isn’t clear until Ransom shows up at the will reading. Jacob, Harlan’s grandson, reveals he overheard Ransom say, “The will… I’m warning you,” before leaving in a rage. When confronted, Ransom admits he already knew he’d been disinherited. Watching this unfold, Detective Blanc realizes this could be Ransom’s motive. We now update the probabilities. Since H4 is broad, we refine it into a more specific sub-hypothesis. The updated hypothesis space and weights are shown below.
Notice how Marta’s likelihood of guilt plummets once the toxicology report rules out an overdose and Ransom’s anger over being cut from the will comes to light. The posterior probabilities shift whenever solid new evidence emerges—this is what makes Bayesian reasoning so intuitive. Built on conditional probability, it always asks: “Given everything I know right now, what’s the most likely explanation?”
In the diagram above, watch how Marta’s chances drop sharply at times, while Ransom’s surge toward the end as new facts come to light.
Conclusion — What About H3?
As we’ve seen, Knives Out is a perfect illustration of reasoning under uncertainty—the core idea behind Bayesian Inference. At first, the possibility of a family member committing murder rises as contradictions pile up. But as details about Marta emerge, suspicion shifts to her. Then, with Ransom’s arrival and revelations about his clash with Harlan, the odds converge on him. In reality, Harlan actually took his own life to protect Marta, under the mistaken belief she had accidentally given him a lethal morphine dose. So does Bayesian Inference fail because it didn’t settle on H3 (death by suicide)? Not necessarily. Sometimes the truth has hidden layers—here, Ransom deliberately swapped the medications and removed the antidote, intending for Harlan to die. While Ransom didn’t stab or poison him directly, he orchestrated the death. The Bayesian approach goes deeper than the surface cause of death—suicide. When applied with an open and objective mindset, Bayesian Inference can uncover truths buried beneath the obvious.
References
[1] Official Transcript of Knives Out by Director Rian Johnson
[2] FBI Homicide Data
