Bayesian thinking
Yes. The flaw is that the speaker is treating absence of personal observation as if it were evidence of nonexistence.
The argument is essentially:
I have never seen X.
Therefore, I do not believe X exists.
That works poorly because many things exist that any given person has never seen. Most people have never seen the South Pole, a blue whale in the wild, or the city of Youngstown, Ohio. Their existence is supported by abundant independent evidence.
The comparison between flying saucers and Youngstown is actually interesting because the two cases are not symmetrical.
For Youngstown, there is overwhelming converging evidence. One can look at maps, census records, satellite photographs, government documents, news reports, and testimony from thousands of residents. One need not personally visit it.
For extraterrestrial visitors in flying saucers, the situation is different. There are claims, anecdotes, photographs of varying quality, and disputed reports, but no evidence that has convinced the broader scientific community. A skeptic therefore has good reasons to doubt the claim—but not because they have never personally seen a saucer.
A more rational position would be:
I believe in Youngstown because there is strong public evidence for it, despite never having seen it.
I am skeptical of alien spacecraft because the evidence presented so far is weak or inconclusive.
The philosophical mistake is often called an argument from ignorance or, more specifically, confusing "I have no direct evidence" with "there is no evidence."
In daily life, almost everything we know comes not from direct observation but from trusting networks of evidence and testimony. Very few people have personally verified that Australia exists, yet they accept it because the alternative would require believing in a vast and implausible conspiracy.
The old joke works as humor because it pretends that Youngstown and flying saucers stand on the same evidentiary footing when they plainly do not.
You’re very close to a standard Bayesian move there, but there’s a subtle trap hiding in it.
Yes, in everyday reasoning we often use something like “base rate thinking”: Ohio cities are common, so prior probability that some Ohio city exists is high; flying saucers (as extraterrestrial craft visiting Earth) are rare in our background model, so the prior is low.
But the problem is that the comparison quietly assumes both propositions live in the same kind of probability space.
For “Youngstown exists,” we are talking about an entity in a well-populated reference class: cities in the United States. We already know roughly how many such cities there are, how they are documented, and what “existence” means in that system. So the prior is anchored in a dense, well-structured dataset.
For “flying saucers exist,” we are not dealing with “rare members of a known class of common things.” We are dealing with something more like: a claimed phenomenon whose underlying mechanism is unknown, undefined, and highly unconstrained. That matters a lot, because “rarity” is not well-defined unless you have a model of the space they are supposed to inhabit.
In other words, “Ohio cities are common” works because the hypothesis space is already well mapped. “Alien visitation craft are rare” is not something we can honestly quantify in the same way without already assuming a great deal about physics, extraterrestrial life, technological feasibility, interstellar travel rates, detection likelihood, and so on.
There’s also a deeper issue: you are implicitly treating “flying saucers” as a single well-defined category with a stable frequency. But in practice the category is fuzzy:
-
Are we talking about extraterrestrial spacecraft?
-
Misidentified aircraft?
-
Atmospheric phenomena?
-
Hoaxes?
-
Psychological reports?
Each of those has a very different base rate.
So the real Bayesian structure is more like:
P(alien spacecraft | reports of “flying saucers”)
not
P(flying saucers exist in general)
Once you condition properly, the “low prior” intuition either collapses into something much more nuanced, or becomes dependent on the quality of the evidence rather than the rarity of the label.
So your instinct—“use prior frequency”—is correct in spirit. The correction is that it only works when:
-
The reference class is well-defined
-
The hypothesis is precisely specified
-
You are comparing like-with-like events
Without that, “rare vs common” becomes a kind of rhetorical illusion: it feels statistical, but it’s not actually grounded in a coherent model.
And that’s why the Youngstown analogy is clever but misleading: it sneaks in a clean, well-structured probability space on one side, and a messy, ill-defined one on the other.
What you’re doing there is very natural reasoning—but it smuggles in a step that feels decisive and actually isn’t.
You’ve implicitly set up a ladder like this:
-
“Man on a bicycle” → ordinary, so accept
-
“Unicycle” → unusual but still within known physical behavior, so accept with mild doubt
-
“Floating in air” → violates strong background constraints, so reject
That feels like you’re tracking probability, but what you’re really doing is switching between two different kinds of reasoning without noticing.
The first two steps are fine because they stay inside a stable model of reality: human locomotion, balance, vehicles, etc. A unicycle is rare but not conceptually disruptive. It just has a lower prior but still high likelihood of being correctly reported or misperceived.
The jump to “floating” is where things change: you’re no longer just lowering prior probability—you’re implicitly saying the report would have to overcome a very strong likelihood penalty. And that’s the key distinction.
In Bayesian terms, the question is never “is this rare?” It is:
Is the evidence more likely under hypothesis A or hypothesis B?
So for your example:
-
H₁: “He saw a normal cyclist”
-
H₂: “He saw a unicycle rider”
-
H₃: “He saw a floating object”
You’re right that H₃ has an extremely low prior under ordinary physics. But the real question is: what is the probability that a person would report “floating” if the world is actually normal?
And that’s where things get messy. Because perception and reporting are noisy. Humans misidentify distance, motion, reflection, terrain, perspective, even sarcasm or exaggeration.
So “No” is only justified if:
(a) the prior is extremely low and
(b) the evidence is not strong enough to overcome the likelihood gap and
(c) alternative explanations (misperception, exaggeration, error) are more plausible
The trap in your reasoning is that you’re treating “violates intuition about physics” as if it directly translates to “can be rejected immediately.” But Bayesian updating doesn’t allow skipping the evidence step. Even absurd-seeming claims still get evaluated by asking whether the observation is more expected under “someone is mistaken” than under “physics has failed locally.”
Now, to connect this back to flying saucers:
The serious skeptical position is not “that can’t happen, therefore false.” It is:
Given what we know about physics, life distribution, detection systems, and human error rates, the probability that these reports are misinterpretations or artifacts is vastly higher than the probability that they are alien craft.
That’s a long, evidence-weighted argument—not a simple “rare things don’t happen.”
So your instinct about escalation (“bicycle → unicycle → floating → no”) is tracking something real: decreasing plausibility. The correction is that plausibility isn’t a hard cutoff—it’s a ratio between competing explanations, not a switch that flips to “impossible.”
If you want a compact way to say it:
Rarity lowers probability, but it never replaces evidence. Even very strange claims can be rationally entertained if the alternative explanations are even less likely.
No comments:
Post a Comment