One of my favorite novels is Scarlett Thomas’s The End of Mr. Y, in which a modern female graduate student in philosophy, while conducting research for her dissertation on “thought experiments,” accidentally discovers postmodern metaphysics is literally true and reality is a social construct—that can thus be changed by changing what people think. Which the novel carries out to expectedly bizarre results, from her doing a favor for a rat god some kids conjured in their backyard by changing reality so no animal experimentation ever happened in human history, to erasing Einsteinian Relativity and inventing the Garden of Eden and mothering the human race. It’s one of the most brilliant of thought experiments traversed into a novel. After all, if “reality is a social construct” were true… 

I’ve discussed before (in my Critique of Rea’s Argument against Naturalism) how this notion that thought controls reality is unfortunately (or fortunately) all too easily refuted in reality. Because things like that just don’t happen. Quite the contrary in fact.

Thought experiments have their place. They’ve produced all the world’s fantasy and science fiction. Almost all our decision-making and planning is actually based on them (whether we’re aware of this or not). They’ve even led to the formation of entire scientific subject fields (from Relativity Theory to Game Theory). And they can expose our faulty assumptions about a lot of things, by forcing us to think through scenarios we hadn’t considered in our logic before. They can expose overlooked possibilities and categories. They can hint at impossibilities and absurdities. They are useful analytical tools.

They can also be deployed so horribly badly as should embarass any solid thinker.

What is it that hoses a thought experiment? And what conditions do we have to meet to use them correctly?

The Trouble with Thought Experiments

Others have already explored the good, the bad, and the ugly of thought experiments. Wikipedia has a handy article, as does the Stanford Encyclopedia of Philosophy. And Roy Sorensen has written a whole book on Thought Experiments that may still be one of the best studies of the subject to date. For a great list of examples of use to the everyday philosopher is Julian Baggini’s The Pig That Wants to Be Eaten, who rightly doesn’t claim the results of any can be certain, but shows how they all force us to think about concepts in ways we hadn’t before, which can always produce insights even when we aren’t sure of any solutions to the imagined problems: one of the most valuable uses of thought experiments; a distinct use that differs from attempting to discern reality from them.

And for those not aware, this is a really important distinction in philosophy everyone should be aware of. Analytical statements and arguments and models are all about exploring possibilities and internal logic (“If we assume x, then what follows?”). They don’t teach us about reality, but they can help us frame questions to explore empirically that teach us about reality. Synthetic (or just “empirical”) statements and arguments and models are all about getting at not merely what’s possible, but what’s probable, and thus what’s true in reality, not just in our imagination (“If we observe x, then what follows?”).

Thought experiments can be used to both ends, but as such have very different reliabilities.

  • “Analytical” thought experiments are always useful if conducted correctly; we will understand a subject and concepts better, even though we won’t yet know anything more as to what’s true (that still requires going and looking at reality). Though we might thus know more about how to find out.
  • “Synthetic” (i.e. “empirical”) thought experiments, however, are not so trustworthy. They are in that role a dodgy way to bypass a paucity of good data to guess what’s true in the world; but those guesses still depend on implicit empirical data we might not realize we are introducing, and are highly prone to just being wrong. This is why science has always far surpassed philosophy in getting to the empirical truth of things. But science can only work on good data. And most things in life lack good data. Yet we still have to reach conclusions about them.

Keep these two different uses of thought experiments in mind.

Here I will just pick on a sample few experiments gone awry, and bring out a few general lessons. There are many famous or recent or commonplace thought experiments that have actually been incorrectly conducted or whose correctly-generated results are contrary to what their proponents assert.

Hilbert’s Hotel

Even now in the ongoing Carrier-Marshall Debate we see the folly of Hilbert’s Hotel deployed, where a hundred-year-old thought experiment about infinite hotels endlessly adding more guests mis-operates by erroneously assuming axioms only true of finite quantities apply to infinite ones, leading some to mistakenly conclude infinities should behave just like regular quantities do, when actually Hilbert’s very thought experiment is demonstrating they don’t. Rather than proving infinities impossible, this Hotel only proves them unusual. The wrong conclusion is thus reached even from a correctly described experiment. And that’s only one of several ways to hose a thought experiment.

The Chinese Room

Consider, for instance, Searle’s Chinese Room, which as I demonstrate in Sense and Goodness without God produced invalid results because Searle did not actually conduct his own experiment as described.

The basic idea is, you have a conversation in Chinese with “a room,” by inserting slips of paper into it with questions and statements in Chinese, and reading the slips that come out, and being convinced you are speaking to a real sentient person (like a Turing Test). But really, all that’s in the room is a guy who knows no Chinese, and a book full of instructions that tell him what characters to send out given which characters come in. So where is the consciousness in the room? Searle concludes from this that consciousness can’t just be a physical computer (“the book” in the room).

But Searle ignored what “the book” in the room had to actually be capable of doing to satisfy the stated conditions of the experiment; whereas doing the experiment correctly gets only a vacuous result: since the man in the room is analogous to the human circulatory system, the only demonstrable result, “that circulatory systems are not conscious,” is hardly a wowzer of a discovery. The book in that room, meanwhile, remains as evidently conscious as ever; so we’ve learned nothing from Searle’s experiment about the reducibility of consciousness to books.

Just-So Stories as Bad Thought Experiments

Consider likewise Evo-Psych fanatics who think thought experiments can replace science altogether, and thus conclude men will surely be promiscuous and women not “because sperm is unlimited and eggs are finite,” whereas all scientific evidence now shows such sexed distribution of behavior has not arisen in hardly any species on the planet, particularly among primates—including humans. Contrary to old beliefs and popular myths, all sexes of nearly all species are sexually promiscuous, refuting the thought experiment that ensures us this should not be.

Indeed most Evolutionary Psychology consists of inventing “just so” stories, and passing off what are essentially bad thought experiments as empirical and thus “proven” science (see my discussion of this and other goofs by EvoPsych fans in my extensive treatment in Is 90% of All Evo-Psych False?). Not checking the actual facts first and instead running a faulty thought experiment and calling it “science” is probably the biggest hosing of thought experiments possible.

The Confabulation of Disembodied Minds

Another example is something I pointed out before in The God Impossible: the readiness with which people imagine disembodied minds possible because they can imagine them. Surely if they were logically impossible, we couldn’t possibly perceive them in our imagination! But we’ve made another kind of mistake here. We’ve disregarded an important meta-issue with the experiment we are running: we are running the experiment in our mind.

We can indeed imagine a brainless mind…but only by cheating: by using a brain to do it. So really, we can only ever know what it is like for an embodied mind to imagine a disembodied mind; we can never imagine an actually disembodied mind—much less what it would be like to be a disembodied mind, because we will always be an embodied one, no matter what we are imagining. So disembodied minds may yet be logically impossible. It’s the same error as saying that because we can run a program on a computer, surely we can run the program with no computer, and so why do we even need computers? “Can’t we just have pure, disembodied programs?” Well, no. And for the same reason we probably can’t have disembodied minds.

Many errors in thought experiments derive from the experimenter not taking into account the baggage being brought in by the fact that they can only use their own mind for the experiment, and all the assumptions, requirements, biases, and lack of information that entails.

This is also why “I can’t imagine it” is often not by itself a good argument for something being false; your imagination is simply limited. And always will be. You can continually expand it, and thus imagine more than you ever had before. But there will always remain truths your brain is not yet equipped to imagine. And the more so, the less experience you have in any relevant capacity to what you are imagining; conversely, the less so, the more experience you do have with the subject being imagined. Hence experience can determine the strength of arguments from lack of imagination. And thus experience is important to successful thought experimenting (very much contrary to James Brown’s erroneous philosophy…wait, who? Oh right, we’ll get to that next).

In this respect, my discussion of “intuition” and its limitations in Sense and Goodness without God is required reading. There simply is no synthetic a priori knowledge. Every time you think you have some, it’s always just plain old synthetic knowledge being snuck in; or some analytical knowledge you are trying to pass of as empirical. Metaphysics is just physics with shitty data.

Galileo’s Goof

One philosopher, James Robert Brown, has even built a completely false worldview on top of getting thought experiments wrong, in his extensive study of thought experiments, The Laboratory of the Mind. He is thus fooled by screwing up thought experiments into thinking synthetic a priori knowledge exists, and we can figure out physics from the armchair, no science needed!

One example is given in a recent Aeon article on Brown. There he calls up the argument of Galileo that Aristotle’s theory of objects falling at different speeds owing to their different weights is impossible:

[S]uppose we connect the two objects with a short, stiff rod. One could argue that the lighter musket-ball acts as a brake on the heavier cannonball, slowing its fall. Then again, one could also argue that the composite body, whose weight is equal to the sum of the two original bodies, must fall faster than either body alone. This is obviously a contradiction. The only solution, Galileo says, is that all bodies fall at the same rate, independent of their weight.

In this case the conclusion is correct…but the experiment doesn’t produce it! Galileo hosed his experiment. And so thus does Brown, who gullibly thinks its amazing and proceeds to base an entire philosophy of synthetic a priori knowledge on it.

Of course, the time it takes for an object to fall does vary by its mass imperceptibly, as heavier objects pull the earth toward them (minutely) faster, but when you are dropping two objects next to each other, the lighter object benefits from the heavier one pulling earth faster to them both (their pull on the earth indeed is combined), so they still hit at the same time. If you dropped them from the same height at different times, the heavier object will fall “faster”—in the sense of hitting earth sooner (even if imperceptibly). Likewise objects dropped from different heights will fall at different rates (hence the noodle effect that Galileo’s thought experiment also failed to anticipate, despite being so close to). But these aren’t problems for his thought experiment as described, since it has already (conveniently) specified conditions that would avoid these effects.

Rather, what’s wrong with this thought experiment is that it presumes its conclusion (how gravity is supposed to work) before even conducting the experiment. Galileo actually made a lot of goofs like this in his writings. People just don’t talk about his mistakes—which all came to be empirically refuted. And here, the result wasn’t even refuted, because it was correct—so people praised the discovery, but overlook that this experiment doesn’t produce it.

One way to avoid this mistake is trying to prove yourself wrong with a contrary model. Imagine if Galileo were intent on proving Aristotle right, and thus wracked his brain for a way the world could be such that Aristotle’s law would hold. What could he have come up with? Indeed the value of the steel man argument in all forms of thought cannot be over-estimated. 

Here just think about it: what if indeed the gravitational force was projected only by objects toward the earth (driving them) and not by earth toward the objects? Then an object that weighed twice as much would generate twice the gravity. And indeed would accelerate toward the earth twice as fast. What would then happen if you linked them together by a bar? If they didn’t affect each other (if the force of attraction they projected was only toward the earth, and in Aristotle’s day, indeed even Galileo’s day, that’s all anyone was capable of observing), then this bar-bell would rotate, as the heavier object pulled more (like a rower in a boat working faster), and pull the smaller object down faster (like a faster rower pushing or pulling the slower), combining their forces, now transmitted through the bar.

It’s possible to illustrate this with the electromagnetic force: one highly charged ball will fly toward an oppositely charged but stationary plate faster than a lightly charged ball will; and if you anchor the charged balls together, the heavily charged ball will pull the lightly charged one faster than it would otherwise travel. Thus what Galileo imagined impossible is not only possible, it can actually occur in the real world.

Of course, this isn’t how it really is. Gravity is proportional to all masses, including earth’s; so the force is always in sum the same. Falling just isn’t that analogous to electrostatics. But this assumption is not stated in Galileo’s thought experiment; he therefore was assuming something not in the experiment, and thus he was not actually doing the experiment correctly. He was basically just assuming the conclusion as an unstated premise, thus arguing himself in a circle. Analytically valid, but useless for predicting how the world works. In this case he was simply sneaking in an observational conclusion he had already made; and thus wasn’t really reasonoing a priori as Brown mistakes him as doing.

And this was a mistake. Because when trying to explain the world, the function of thought experiments (even more than actual experiments, like dropping wonky barbells and watching what happens) should be to think through so as to compare different models, precisely so you can deduce some ways the world would be observably different in one model than another—which you can then go and test! Which is basically Step One of any sound Scientific Method. Step Two, incidentally, is doing your honest damnedest to prove your preferred theory false…and concluding your theory is likely to be true only when you fail. And in the end, Galileo just wasn’t as good at this as Newton was.

(For more on all this see Galileo’s Goofs.)

The Trouble with Mary

I’ve discussed the indomitable Mary the Scientist before. This imaginary heroine gets recruited in many thought experiments run badly. The original I think was Mary and the Colorless Room, a thought experiment supposed to prove conscious experience magical. The idea being that Mary lives all her life in a colorless room and knows all true propositions about the world yet still doesn’t know what “red” looks like!

A later iteration is Mary and the Mystery of Plus or Quuus, supposed to prove even propositional knowledge is magical. That one is more convoluted (follow the link for an explanation and analysis).

Both fail in the same way: everyone gets wrong what Mary knows in these thought experiments, by not actually thinking through what Mary would actually know—the thought-experiment equivalent of assuming what a woman thinks without doing the obvious and just asking her.

The first experiment only proves that some knowledge is experiential and not propositional, that the only way to “know” certain things is to experience them (I also discuss this in Sense and Goodness without God: II.2.1.1, pp. 30-31; III.6.4.4, pp. 146-48; III.6.5, 148-50). Just as all possible propositions about hearts will never pump blood—you still have to build a working heart. This does not mean hearts are magical, or that they are anything more than the parts they reduce to and their arrangement. It is simply teaching us something about propositional knowledge: it’s just codes for doing things; propositions about things, are not and will never be the things themselves.

And lo, all Mary has to do is use the propositional knowledge the experiment says she has to stimulate or reconfigure her brain or room to experience color sensations. And then she’ll know them. So it’s actually not true that she will only know colors when she leaves the colorless room. Just like she won’t know how to ride a bike by knowing all propositions about it, but all propositions about it are all she needs to build a bike and learn to ride it (assuming she isn’t disabled etc.); and she won’t be able to pump blood by knowing all propositions about hearts, but all propositions about hearts are all she needs to build one and pump blood (assuming she has access to the needed materials etc.).

Similarly, Mary can easily know the difference between Plus and Quus, buy using her propositional knowledge (as I explain in the previous link).

Thought experimenters using Mary in this way simply aren’t thinking through the difference between propositional (encoded) and actual (experiential) knowledge. Like the Chinese Room, these experiments actually don’t tell us anything about the irreducibility of experiential knowledge to purely physical systems. They simply teach us that experiential knowledge is different from propositional. Which we should already have known.

A lot of this has to do with not realizing that a mind (and thus “knowledge,” as distinct from just data) is a process and not an object. And that propositions about things are not the things they are about.

Sorry but…Zombies Are Impossible

A converse example are thought experiments that are on close analysis self-refuting. An example of this is the so-called “philosophical zombie” (see Wikipedia and the Stanford Encyclopedia of Philosophy for background).

But before I get to what a “philosophical” zombie is, let’s tour the taxonomy of zombies for a moment…

Among “movie zombies” there are two kinds: supernatural (it’s “magic” or “ghosts” or something) and science-fictional (it’s, uh, a fungus or something…in any event, supposedly physics can explain it, even if the writers handwave how). Regardless, all movie zombies are effectively impossible.

Supernaturally animated objects (much less corpses) don’t exist, of course; but they are probably impossible on present physics as well (insofar as physics has pretty well catalogued the forces and materials available for producing effects in the world, and what we have in result is naturalism). And they might even be logically impossible, insofar as the supernatural entails things that appear to be (even if we can’t yet prove them to be) logically impossible (see The Argument from Specified Complexity against Supernaturalism).

And “science fiction” zombies are physically impossible. Motion (much less perception and intellection) requires burning fuel (like, carbs) and removing the resulting exhaust and toxins that otherwise would chemically stifle combustion (like, CO2), and burning fuel requires oxygen to combust with, indeed in corresponding quantity to the energy-storing molecules that have to be burned to generate muscle contraction and neuron firing (for example). Which requires a functioning circulatory system. But you keep needing more oxygen to burn and to remove the exhaust. Which requires a functioning respiratory system. And you need to keep supplying fuel. Which requires a functioning digestive system. And if the blood system is under-pressured, it can’t get fuel and oxygen into cells; and if it’s over-pressured, it destroys the cells and they therefore can no longer perform any function (like move or think), so you need pressure regulation (and to clear from the blood any other toxins that will destroy the cells). Which requires a renal system. And so on.

Basically, trying to imagine how a zombie would work (at even just the most basic level of moving and seeing), you end up realizing you just built a human, and thus discover why we have all these organ systems—and why we stay dead when we don’t.

Well, okay, maybe you can have something like a nanorobotically animated corpse. That’s not physically impossible. It would just be weird. And it would still have metabolism and so on (and thus need oxygen or some kind of visible and thus destroyable or deprivable power source, power-train, brain, and so on). But other than that, even sci-fi zombies are impossible. Likewise, merely “crazy people” are possible (like that ever-imagined virus that drives people savage in some films), but they’re not actually zombies.

But then we get to the “philosophical zombie.” Which IMO is a full-on logical impossibility as defined, because it requires an impossible condition: a person who isn’t lying about their experiencing qualia yet isn’t experiencing qualia. Even if one tries to find a way to reconcile those two conditions, all they end up with is the realization that our mere belief we’re experiencing qualia is what qualia are. We don’t end up with magical qualia. Just a physical system that genuinely thinks it’s experiencing them. Which could simply be us. That’s an interesting discovery. But it isn’t what the inventors of this thought experiment were hoping for.

In any event, I’ve covered all this (and the Mary the Scientist cases as well) in The Mind Is a Process Not an Object.

Trolley Problems and Social Systems

One of the most famous and versatile thought experiments invented in the 20th century was the Trolley Problem, originated by one the greatest female philosophers of the 20th century, Philippa Foot (Wikipedia has a useful entry). It’s even led to real experiments. Which have importantly demonstrated, relevantly to today’s theme, that “people think differently in hypothetical dilemmas compared to real-life scenarios.” In other words, it’s so common for people to run thought experiments badly, that they often won’t reliably inform you of reality. An important lesson to remember when resorting to or analyzing thought experiments yourself. 

Plus, of course, what people do in experiments is also not thereby evidence it’s correct. The basic trolley problem only really tests how we “feel” about inaction that causes more harm than an action that causes harm but reduces it. “Would we kill one to save five? When and why?” But really, does it matter whether what does the killing is our action (flipping a switch) or inaction (doing nothing)? Objectively of course it shouldn’t matter; but in reality, humans are so terrified of causing harm by their actions, that they will often allow greater harms to prevail by their inaction. This describes nearly every election in United States history. And it’s arguably not at all rational.

But some versions of the experiment are conducted incorrectly, sometimes because they incorrectly limit the available options and thus deviate from what would happen in the real world. And sometimes because they overlook broader consequences that have to be factored into analysis, by artificially limiting the context of a decision. An example of the first error appears in the Hangman version: You stumble into a strange village; the villagers tell you they are going to hang five innocent people to ward off evil, but they will hang only one if you perform the hanging. What do you do? This experiment is badly conceived because it usually omits more obvious options: like, kill all the villagers. Or just about anything else that describes pretty much any nation on earth’s foreign policy.

But it’s the former kind of error that’s more common. This appears in the so-called Fat Man scenario—if you can save many lives by pushing a “fat man” off the train platform onto the tracks to stop a runaway trolley, do you? And the Mad Doctor scenario—if you are a doctor at a hospital and a patient comes in for a minor ailment but killing them to harvest organs would save several lives, do you? The same error occurs in both: they both seem to present only a straightforward question of utilitarianism: one or many? But this obscures the fact that both scenarios inherently take place in social systems, a context that radically alters the consequences to be considered, far beyond the narrow instance presented.

Quite simply, “train platforms” and “hospitals” are institutions that exist because of their social utility. If we accepted that people could be killed at them with impunity, people would stop using them. Which has the same effect as their no longer existing. The entire society loses the social utility of those institutions. Or else people who do use them, go armed. Creating an enormous risk of negative utility and a higher social cost for the positive utility. In no calculation is that a positive net gain.

In law this principle manifests as Duty of Care: in certain scenarios, we have a “duty of care” to others that we don’t normally. Such a duty exists for a utilitarian purpose: accepting (and thus enforcing) the duty creates social utilities that benefit everyone—and that would not exist but for the duty being accepted. If doctors exercised no duty of care, hospitals could not exist (or would cost a great deal more to use). If people using train platforms exercised no duty of care, train platforms could not exist (or would cost a great deal more to use). So these thought experiments by themselves risk not making clear that these are also consequences to be weighed. Merely adding up a single-instance body count does not match the real-world issues at stake.

I discuss this aspect of trolley problems more in my Open Letter to Academic Philosophy and Everything Is a Trolley Problem. But the overall point here is that thought experiments should not be treated in isolation: they often require placing the imagined scenarios in larger contexts that may well change how things look, and thus how the experiment is completed.

This realization, by the way, derives from Game Theory, one of the few thought experiments to lead to a rigorous mathematical science (see The Real Basis of a Moral World).

Conclusion

Thought experiments can be powerful tools. They allow us to explore artificially constructed models of reality, to explore “what if” questions, better understand concepts and assumptions, demarcate possibilities to explore or consider, write good fiction, and generate avenues for empirical and theoretical research—even to reach moderately reliable conclusions about life, ourselves, and the world, particularly when we need to base decisions on something but can’t do the full-on empirical scientific studies that would otherwise answer a question before us. Indeed most of human life consists of running thought experiments, just to decide what to do, what to believe, or what futures to pursue.

But thought experiments can also become a ready source of error, when they are not conducted correctly, or not treated as what they really are: intuitional guesswork, exploring artificial models we’ve invented, to test the boundaries of ideas or possible states of the world, ever limited by our empirically acquired background information.

And “correctly” here means at least four things in combination: 

  • (1) The experiment must be carried out as described (whereas in many cases, e.g. Searle’s Room and Mary’s Room, it is not).
  • (2) The experiment must not be carried out in isolation (for instance, if wider social or physical systems would be affected or affect the results in the experiment being proposed, that cannot be ignored; as with Trolley problems).
  • (3) We must admit the results tell us nothing about reality unless all the components of the experiment can be demonstrated to likely exist in reality, or to exhaust all possibilities in reality—including any implied limits on what can be done in them: because if reality lacks those limits, the results won’t apply to the real world; or will only in very limited cases, e.g. “Life Boat Scenarios” might tell us things about lifeboats, but most of our lives are not lived in lifeboats, so their conclusions do not generalize to anything broadly useful, outside rarefied analogs to “being in a life or death situation.” Yet even those situations may present more options than we think of in our experiments.
  • (4) We must admit the results are highly uncertain, i.e. we must not treat a thought experiment as a logical demonstration. It is not. It can often be wrong, or even refuted empirically.

This last is important. The results of a thought experiment can be overturned by actually taking the thought experiment into the realm of real, actual experiments. Or real, actual mathematical proofs. This latter is too often forgotten. We should always be asking: “How could we test this for real?” A conclusion reached correctly by a thought experiment that really does appear to track reality can be a useful “so so” philosophical result; but it is not a logical demonstration, nor a scientific one. It is sub-scientific. Its results will be less certain than even the results of a weak scientific experiment (much less a strong one; even less a whole series of strongly replicated scientific results). Because let’s face it, philosophy is always just science with shitty data (as I demonstrate in Is Philosophy Stupid?). Science is better. It’s just not very often what we have; good data is hard to get.

For other examples of botched thought experiments I’ve since covered:

Remember thought experiments cannot trump real experiments; whereas knowledge reached by the latter always trumps knowledge reached by the former (see my discussion of the ladder of knowledge in early sections of Sense and Goodness without God). So we really should do the actual experiments whenever we can. As long as, of course, we are doing them correctly, too; after all, a bad real experiment will be just as invalid as a bad thought experiment (for instance, again, some examples I explore in Is 90% of All Evo-Psych False?).

Failing to attend to this rule is why most Libertarian and Marxist economic theories fail to work in reality, despite their still-ardent advocates’ certainty they must be correct…owing to over-trusting or mis-conducting associated thought experiments. Thus we see seemingly irrefutable thought experiments giving up false results like “increasing the minimum wage will result in widespread unemployment” or “a centrally controlled economy is more efficient than a regulated free market” or “we can achieve world peace by abolishing centralized governments.” 

Arguments in defense of monogamy and gender hierarchy likewise rest on faulty thought experiments that fail to take into account real-world evidence (or even insert fake or false or cherry-picked evidence as premises). Or the random dufus who argues raping an unconscious person isn’t a big deal because he wouldn’t mind if we used his house and car without his knowing it—prompting us to want to take a shit on the floor of every room of his house to prove the error in his thinking. Thought experiments are tricky things. If you’re going to use them, don’t fuck them up.

Discover more from Richard Carrier Blogs

Subscribe now to keep reading and get access to the full archive.

Continue reading