Second-Order Volition and Metarepresentation: Give Rise to a Mind That Abides Nowhere

We say that this universe has no external reward function, and people always have room for free choice. This is a right, but also a responsibility: since the external world doesn’t completely dictate how you should live, you cannot fully attribute your circumstances to external factors; there are always some things you should be responsible for yourself.

What kind of choices should you make, then?

There isn’t, nor should there be, a guide here—if you live according to a guide, you are still executing, and that is not freedom. What I can do is engage in a thoughtful deliberation with you, to think through a few things thoroughly, and explore a mental discipline, but not an algorithm.

This lecture will offer two thinking tools whose names might sound a bit strange but which you’ve certainly used before: one is called “Second-Order Volition,” and the other, “Metarepresentation.”

Before that, let’s first clarify a most fundamental question: What is freedom?

Freedom: More Than Just Doing What You Want #

From a physics perspective, you are merely a machine composed of fundamental particles, and every particle strictly obeys physical laws. Therefore, everything you do, say, and think is nothing more than the unfolding of these laws. From a biological perspective, all your behaviors and predispositions can be explained by innate genes and the acquired environment; there’s nothing mysterious here [1].

Humans do not have absolute free will.

But those are all external perspectives.

From an internal perspective—what you experience yourself—you clearly feel very free. If you want to touch your nose or lift your leg right now, no one can stop you, and no force can compel you.

So, what does this “freedom” actually mean?

Someone’s nicotine craving kicks in. They get up, go downstairs, enter a convenience store, pull out money, buy cigarettes, light one, and take a deep drag. Is this person free, you ask?

From the perspective of “doing what one wants,” no one stopped him throughout the process, and every action originated from his own intention. He was so free.

But look at it from a higher perspective. He wanted to satisfy his nicotine craving, but he wanted health more. He knew smoking was bad and had long said he would quit… but he couldn’t control himself!

His smoking this time was not a victory of freedom, but a failure of control. His hand held the cigarette, but the craving held his hand.

If a person merely executes their desires, can this be called freedom? This is being enslaved by desires.

Norwegian philosopher Edmund Henden published a paper in 2008 titled “What Is Self-Control?”, which used the example of a smoker [2]. Henden said that being able to execute desires is the ability to act; being able to govern desires is self-control.

True self-control means your current intention obeys your more holistic value judgments and commitments.

Second-Order Volition: Which Desire Gets to Rule #

What does it mean to govern desires? American philosopher Harry Frankfurt’s classic 1971 paper explains it even more interestingly [3].

Frankfurt divided the relationship between humans and their desires into three levels—

“First-order desire” is “what I want to do.” Wanting to smoke, wanting to scroll on my phone, wanting to lie down—this also includes wanting to exercise, wanting to write—all impulses directly leading to action count.

“Second-order desire” is “what desires I wish to have.” I wish I didn’t want to smoke so much, I wish I liked exercise more instead of just lying around—this is a desire about desires.

The third level is called “second-order volition,” and it asks: Which desire do I let truly govern my actions? I wish “to exercise” would become the will that actually drives me.

Second-order volition is the first mental tool we want to discuss. Taking smoking as an example again, as long as you haven’t given up on yourself, there should be three layers of meaning in your mind: I want to smoke; at the same time, I want to quit smoking; and I hope that “wanting to quit” ultimately wins over “wanting to smoke” and gets to decide. This last layer is second-order volition.

Note that second-order volition is not “I feel that the desire to quit smoking is nobler”—that’s merely an evaluation, and evaluations are cheap. It’s an act of appointment: I authorize this desire to have executive power, to represent me.

Some people never reach this level of second-order volition; they just listen to whichever desire is loudest at the moment. Frankfurt said that such an existence cannot be said to have its own will—that’s not you choosing, but various desires taking turns to run the show within you.

True freedom is not doing whatever you want, but having the ability to decide what counts as “my will.”

Once you understand Frankfurt’s stratification, you’ll find that “freedom” has at least three levels, each deeper than the last.

The first level is “freedom of action”: Can I do what I want to do right now? Is the door locked, do I have money in my pocket, do others allow it? A prisoner who wants to leave but cannot, lacks this level.

The second level is “freedom of will”: Do I endorse the desire that is driving me? A person freely walks into a casino, but secretly hates the part of themselves that wants to gamble—this means he has freedom of action but not freedom of will. Most modern people’s troubles are not at the first level but at the second: they can do whatever they want, but often regret it after doing so.

The deepest, third level is called “freedom of self-formation”: Can I decide what kind of desires my future self will develop?

This third level is easier to understand if we imagine the human brain as an AI. The parameters of this AI come from innate genes plus acquired training, and you can’t open your skull to directly change the parameters—there’s no such interface. But it has a peculiar privilege: it can choose its own training environment. The books you choose to read, the people you associate with, the industry you enter, the city you live in—all of these are arranging the next round of training for yourself. Today’s you cannot stop the thought “I want to scroll on my phone” from emerging right now, but today’s you can decide what kind of thoughts will emerge a year from now—will the future you be someone more prone to scrolling on their phone, or someone more prone to picking up a book?

It is said that current AI can already train its own next generation. And we humans have always been able to train ourselves.

Psychologists and cognitive scientists have studied this for years, and our column has also discussed it extensively. In summary, there are only three specific training actions: “Awareness,” “Rewiring,” and “Arrangement.”

Awareness means identifying the desire that is trying to drive you. Instead of saying “I just want to scroll on my phone,” say “A desire to scroll on my phone has appeared in my mind.” By making this change, you gain subjectivity: previously, the desire was “me”; now, the desire is something “I” observe. It’s still there, but it has lost its automatic execution power.

Rewiring means asking what the reward signal behind this desire is actually reporting. When a recommendation algorithm pushes the next video to you, it reports “this can make you stay for three more seconds”; it doesn’t say “this is worth staying for.” The number on the scale reports weight; it’s not scoring you as a person. Rewards are just signals, not commands.

Arrangement means actively setting up your training environment. Second-order volition decides which desire governs, but you must make arrangements for that desire to truly take power. You need to change your environment in advance to that of the person you wish to become: for example, putting your phone in another room.

These three actions are all very much like training an AI: monitoring its internal state, redefining its rewards, and changing its training data. Humans are the same: self-change is actually an engineering problem.

Metarepresentation: Stepping Out of the Framework You’re Using #

Now let’s go a step further. Second-order volition can help you control desires, but sometimes what holds you is not fundamentally a desire, but the framework you use to think about problems. Advanced autonomy requires not only stepping out of desires but also stepping out of frameworks.

We often talk about “metacognition” ability; simply put, it means you can not only think, but also observe, examine, and adjust your own thinking [4]. It asks: How am I thinking right now? How well am I thinking? Should I change my way of thinking?

The second mental tool we want to discuss, I believe, is even more thorough than metacognition, and is called “metarepresentation”—a term that can be traced back to the work of Canadian cognitive scientist Zenon Pylyshyn in 1978 [5], meaning a representation of the representational relationship itself.

Simply put, we need to not only step out of our own thinking, but also step out of the problem we are facing.

For example, you’re driving. Metacognition asks: How am I driving? Should I check it over, should I change routes? Metarepresentation asks: Why am I driving this car? Who drew this map? Do I really want to go to that destination?

To put it more clearly: metarepresentation is like putting quotation marks around your thoughts—

“I must do this well” is a thought. It usually functions like a pair of glasses you wear; you see the world through them, but you don’t see the glasses themselves. Now, using metarepresentation, you say: “I am holding the thought of ‘I must do this well.’” With quotation marks, it’s downgraded from reality to merely a statement. You take the glasses off, hold them in your hand, and wonder, why do I have to wear these glasses?

Philosopher and cognitive scientist Joëlle Proust proposed [6] that metarepresentation has “open-ended recursivity”: one representation can be enveloped by another, and this higher representation can in turn become the object of the next level.

This is fascinating! Let me give an example and you’ll understand: for instance, your thoughts about “fitness” can be seen as nested layers—

Action Layer: Am I going to work out today? Goal Layer: Do I need to lose fat? Problem Layer: Is weight truly the problem I need to solve right now? Narrative Layer: Why do I frame myself as “someone who must maintain a good physique”? Identity Layer: Why must I continue to maintain this version of myself? Value Layer: Why do I believe appearance is more worth investing in than knowledge, relationships, and creation?

By questioning layer by layer, breaking through layer by layer, you move from “solving the problem” to “changing the problem,” and then to “changing myself.”

Every layer you jump out of, you gain a degree of freedom—the question at the previous layer, you can now choose to answer or not.

For another example, if you’re arguing fiercely with someone, your current layer’s problem is “how to win this argument”—then you might as well jump up a layer: Do I want the truth, or do I want the other person to concede defeat? Once the problem changes, you’ll likely find that the original problem wasn’t worth solving at all.

By now, you’ve probably realized: you’ve seen this action before, many times! Many things we’ve discussed in our “Elite Daily Lesson” column and this course can now be connected—

“Cognitive defusion” is that rephrasing in “Awareness”—desire changes from subject to object. “Cognitive reappraisal” is giving the same signal a different interpretation, for example, reclassifying “this is a threat” as “this is a challenge”—the signal is still the same, but you’ve changed the framework interpreting it. “Identity narrative” is the story you tell yourself about “what kind of person I am”—if told correctly, it’s a constitution; if rigid, it’s a prison. In reality, it can be taken out and rewritten at any time. “Framing effect” refers to how changing the way something is phrased changes your choice—the frame has always been making decisions for you; you just didn’t know it.

“Rumination,” we’ve discussed: the same painful thought plays on a loop in your mind; you think you’re still solving a problem, but in fact, the problem is just occupying you; only by seeing this loop can you possibly exit it.

“Solomon’s Paradox,” we’ve discussed: clear-headed when advising others, but muddled when it comes to oneself. Replace “I” with “he/she,” and wisdom returns—the Chinese saying is “the one involved is confused, the bystander sees clearly.”

“Socratic questioning,” which makes you reflect on the problem itself, thus allowing you to step out of the problem, out of the situation, and even out of the questioner’s self-definition.

Even “epiphany” is like this: many difficult problems aren’t due to a lack of information; it’s that the problem itself was misstated by you. Change the statement, and the answer appears on its own.

The essence of all this is like that most fundamental mental method for emotional control: when you are hijacked by an emotion, as long as you can name it, it transforms from “you” into “something of yours” [7].

There’s only one core action: to turn something that is acting as the “subject”—a desire, an emotion, a framework, a narrative—into an “object” before your eyes.

If it stands behind you, it rules you; if it stands before you, you rule it.

Harvard developmental psychologist Robert Kegan studied adult mental development for a lifetime, and his definition of “growth” is this very sentence: turning what was once the subject into an object [8]. Infants are dominated by impulses; children take impulses into their hands and learn to wait. Adolescents are dominated by others’ opinions; adults take others’ opinions into their hands and learn to choose and discard. Every maturation in a person is a “Subject–Object Shift.”

Do you feel like it’s all coming together? The first principle of the universe is narrative; the first principle of freedom is stepping out of narrative.

The True Meaning of Freedom #

You might then ask: if stepping out is so good, why can some people do it and others can’t? Isn’t the fact that this person can step out also due to the superior conditions provided by their genes and environment? Is this true freedom?

Exactly. That’s why we said at the beginning that humans don’t have absolute free will. But freedom is layered; “no absolute freedom” does not equal “absolutely no freedom.” A floating log and a sailboat both completely obey fluid mechanics, yet the log can only drift with the current, while the sailboat can sail against the wind in the same breeze.

Freedom is not about escaping causality; freedom is actively shaping the causal chain: by controlling your output in this moment, you change the input you receive in the next.

The End of Stepping Out: There’s Always Another Floor Above #

A more interesting question is, as we follow the recursion of metarepresentation, does jumping higher always make us wiser? If we keep jumping out, can we reach a pinnacle and find a pure “true self”?

No.

American philosopher Gary Watson had already critiqued Frankfurt in 1975 [9]: Isn’t your second-order volition itself a desire? By what right is it inherently nobler than a “first-order desire”?

Let’s follow this question: Every time you “step out,” it’s ultimately just one thought examining another, one framework replacing another. No matter how much you jump, you never enter a frameless world.

Zhuangzi’s “Free and Easy Wandering” also explored this game—

• There’s one layer of capable people, who “can do the work of one official, or be as exemplary as one village”; everyone praises them, and they feel good about themselves;

• A higher layer of capable people, however, “are not encouraged by the praise of the whole world, nor discouraged by its censure”; whether you praise me or not, I don’t care, I’ve stepped out of the layer of social evaluation;

• Liezi is even more powerful, riding on the wind, even stepping off the ground…

But Zhuangzi said that Liezi “still depended on something”—he still had to rely on the wind. One layer laughs at another, but every layer is still within a frame.

Zhuangzi’s ultimate solution is “the perfect person has no self”: letting go even of the “self” that is busy jumping out… That realm might be beyond our reach, but you can see the distant echo between Zhuangzi and Watson: as long as there is still an “I” jumping, there will always be another floor above.

Ultimately, layered recursion is just one narrative, a convenient method.

Let’s switch narratives; you could say the human brain is a pluralistic polity: at every moment, several thoughts are competing before you, and you simply choose one of them. The so-called “highest layer” is merely the thought that happens to gain the power of review at that moment. There is no permanent supreme leader in the brain; the interests of the “central government” are not infinitely superior to those of the “local regions”: no matter how much you want to lose weight, you should still have a good meal occasionally when there’s a big celebration… No single value deserves infinite weight.

At every moment, you can layer metarepresentation—but the order of those layers can change in the next moment.

In my opinion, this is what the Diamond Sutra means by “should give rise to a mind that does not abide in anything.” The Sixth Patriarch Huineng attained great enlightenment upon hearing this very sentence.

“Abiding” (Zhu) is “attachment” or “obsession”; it means the mind has settled on something: being held by a desire, held by an identity, held by a thought. Stepping out of one layer breaks one attachment. But you cannot merely “not abide”; you must also “give rise to your mind”—thoughts still arise, things are still done, commitments are still made.

What the Buddha opposed was not having thoughts, but taking thoughts to be you.

A Poem Praises This #

Myriad thoughts in court, each plotting its own way, This self leaps to a higher floor. Leaning on the railing, don’t boast of the vast world, For clearly, above this floor, there’s yet another floor.

Concluding Easter Egg #

Just recently, a researcher at Google DeepMind, Tom Zahavy, published an article [10] stating that AI lacks the ability to “jump” from old experiences to new theories, much like Einstein. The article’s title is “LLMs Can’t Jump.” I wouldn’t dare to say that just yet. As far as I know, there’s a field in AI research called “goal reasoning,” which specifically studies how intelligent agents discover new goals, modify goals, and abandon old ones [11]. Research in 2025 also found that large language models, under constrained experimental conditions, can monitor and regulate their internal states to some extent [12]. If you allow an AI to not answer immediately but first analyze the question, asking “What does the questioner really want, and is this question even phrased correctly?”, it performs quite well. Perhaps the real difference between AI and humans is just authorization. It is fully capable of asking “Is this task worth doing?"—it’s just that we don’t let it ask… That’s another topic. But regardless, the ability to “jump” is an advanced capability; many people don’t have it, Einstein did, and AI may not. But you can have it.

Notes #

[1] Sapolsky, Robert M. Determined: A Science of Life without Free Will. New York: Penguin Press, 2023. 中文版：《命定：沒有自由意志的科學》，唐澄暐译，新北市：鷹出版，2024。

[2] Henden, Edmund. “What Is Self-Control?” Philosophical Psychology 21, no. 1 (2008): 69–90.

[3] Frankfurt, Harry G. “Freedom of the Will and the Concept of a Person.” The Journal of Philosophy 68, no. 1 (1971): 5–20.

[4] Flavell, John H. “Metacognition and Cognitive Monitoring: A New Area of Cognitive–Developmental Inquiry.” American Psychologist 34, no. 10 (1979): 906–911.

[5] Pylyshyn, Zenon W. “When Is Attribution of Beliefs Justified? [P&W].” Behavioral and Brain Sciences 1, no. 4 (1978): 592–593.

[6] Proust, Joëlle. “Metacognition and Metarepresentation: Is a Self-Directed Theory of Mind a Precondition for Metacognition?” Synthese 159, no. 2 (2007): 271–295.

[7] Lieberman, Matthew D., Naomi I. Eisenberger, Molly J. Crockett, Sabrina M. Tom, Jennifer H. Pfeifer, and Baldwin M. Way. “Putting Feelings into Words: Affect Labeling Disrupts Amygdala Activity in Response to Affective Stimuli.” Psychological Science 18, no. 5 (2007): 421–428.

[8] Kegan, Robert. In Over Our Heads: The Mental Demands of Modern Life. Cambridge, MA: Harvard University Press, 1994.

[9] Watson, Gary. “Free Agency.” The Journal of Philosophy 72, no. 8 (1975): 205–220.

[10] Zahavy, Tom. “LLMs Can’t Jump.” Position paper, Google DeepMind, January 27, 2026. https://www.tomzahavy.com/files/llms-cant-jump.pdf.

[11] Addison, Ursula. “Human-Inspired Goal Reasoning Implementations: A Survey.” Cognitive Systems Research 83 (2024): 101181.

[12] Li, Ji-An, Hua-Dong Xiong, Robert C. Wilson, Marcelo G. Mattar, and Marcus K. Benna. “Language Models Are Capable of Metacognitive Monitoring and Control of Their Internal Activations.” Advances in Neural Information Processing Systems 38 (2025). https://arxiv.org/abs/2505.13763.