A bilingual deposit (English · 中文) of the SSRN working paper, with the full PDF of each language embedded, followed by the complete Markdown text of both.

English Version (PDF)

中文版（PDF）

The Non-Naturalizability of the Normative: A Single Line of Reasoning from the Definition of Wealth to Practical Legislation

Abstract

This paper follows a single line of reasoning that proceeds from the mercantilist definition of wealth to an unresolved core problem in practical philosophy. The argument establishes the following theses in sequence. First, wealth is not a self-subsistent entity but a framework-dependent institutional fact. Second, mercantilism, owing to its positivist and utilitarian undertones, inherits already-identified limitations including the problem of Hume, alienation, and performativity. Third, the utilitarian intertemporal aggregation of utility depends on a discount parameter that structurally resists naturalization, in the sense that it can neither be stipulated without arbitrariness, nor derived by computation, nor estimated unambiguously from experience. The paper further argues that this epistemological limitation should be reinterpreted at the ethical level as a condition for the possibility of practical freedom rather than as a negation of freedom. The argument finally arrives at a position of normative legislation that has no external foundation and yet admits of no arbitrariness, and it identifies the conflation of induction with legislation and the conflation of description with norm as two manifestations of one and the same Humean seam.

Keywords　political economy; utilitarianism; intertemporal discounting; the problem of Hume; the naturalistic fallacy; practical reason; normative legislation

1. Introduction: A Domain-Independent Meta-Framework

Any concept in the social sciences may in principle be examined within a domain-independent analytical framework comprising four levels. The ontological level asks what kind of being the concept is, namely whether it is an entity, a state, a property, a relation, a process, or an event. The epistemological level asks how it comes to be known and within which interpretive framework it becomes manifest. The level of practical theory asks how it is deployed in action and how it acts back upon action. The ethical level asks what normative judgment it carries.

The methodological claim of this paper is that a powerful analysis of an unfamiliar concept derives less from a retrieval of the established conclusions of the relevant field than from a structural reconstruction of the concept by means of the above meta-framework. The discussion below takes wealth as its point of entry and shows how such a reconstruction advances stepwise from the ontological level and finally closes its loop at the level of ethics and political philosophy.

2. The Ontological Status of Wealth

2.1 The Mercantilist Definition and Its Undertones

Mercantilism identifies wealth with money or with unminted precious metal, that is, with a stock of gold and silver. This definition has a mirror image at the level of folk custom. The arrangements of bride-price and the so-called five golds in traditional marriage customs are in substance a storage and transfer of value in the form possessing the highest liquidity and the lowest risk of depreciation. Such an arrangement is a concrete realization of the store-of-value function and the general-equivalent function of money in everyday life, rather than an anomaly.

The theoretical undertone of this definition can be rapidly identified. It bears a pronounced utilitarian and positivist character. Its advantage lies in being operationalizable and quantitatively assessable. Its cost lies in its inheriting, along with that advantage, the full set of already-identified limitations belonging respectively to positivism and to utilitarianism.

2.2 Wealth as an Institutional Fact

The ontological status of wealth should be characterized as a relational and dependent mode of being rather than a self-subsistent entity. Its status is analogous to that of debt, money, and sovereignty, all of which are institutional facts, that is, things that obtain only under the constitution by collective intentionality whereby something counts as a certain status within a given context [Searle 1995]. A banknote as a physical object is self-subsistent, whereas its qualification as wealth is wholly contingent upon an epistemic and institutional framework.

On this basis the usual formulation can be strengthened. Wealth depends on a framework not merely at the level of being known, which constitutes an epistemological dependence, but also at the level of its very being, which constitutes an ontological dependence. When war renders a currency into waste paper, the mechanism is not that cognition has changed but that the institutional fact carrying that currency has collapsed.

3. The Cost of the Positivist Undertone

In exchange for operationalizability, mercantilism must reify wealth into an observable and countable stock. The cost of such reification is at least threefold.

The first cost is the Humean problem of induction. Positivism cannot guarantee that the gold and silver of today will still constitute wealth tomorrow. The substance of this limitation is that the temporal stability of value lacks any a priori guarantee.

The second cost is alienation. Relational goods, such as love, trust, and recognition, are excluded from the category of wealth because they are not quantifiable. The reason the monetized negotiation of bride-price damages the affective bond is that it measures, by the logic of the general equivalent, that which does not belong to the domain of exchange. This point converges with Sandel’s argument concerning the moral corrosion wrought by markets [Sandel 2012], and likewise with Marx’s critique that money brings about an inversion and confusion of the human and natural properties [Marx 1844].

The third cost is performativity and reflexivity. Economic concepts differ from natural objects in that they change by virtue of being measured. Once wealth is measured by gross domestic product, agents turn to acting for the sake of that very indicator, which is what Goodhart’s law characterizes [Goodhart 1975]. Economic theory exercises a formative power upon the behavior it describes, which is the performativity thesis [MacKenzie 2006]. Positivism presupposes that observation does not perturb its object, whereas in the social sciences observation itself constitutes intervention.

4. The Intertemporal Problem of Utilitarianism

4.1 The Infinite Horizon and the Ill-Definedness of Value

If one computes, in utilitarian fashion, the present benefit of a given object, one must in principle bring into consideration both its present value and the value of the states reached at every future step. Let $m_t$ denote the state at time $t$, let $\text{value}(\cdot)$ denote the single-step value functional, let $w(t,t’)\ge 0$ denote a weighting kernel, and let $Z_t$ denote a normalization coefficient. The general form of the aggregate value may then be written as

$$
V(m_t) ;=; \frac{1}{Z_t}\sum_{t’=t}^{\infty} w(t,t’),\text{value}(m_{t’}),
\qquad
Z_t ;=; \sum_{t’=t}^{\infty} w(t,t’).
$$

The normalization coefficient should not be fixed in advance as $1/N$. Its choice depends on the semantics the aggregation is intended to express, and distinct choices correspond to distinct ontological commitments. If $w(t,t’)\equiv 1$ and the horizon is the finite $N$ steps, then $Z_t=N$ and the aggregation reduces to a mean, which is the case of a finite expectation. If the aggregate value is to be normalized to the unit interval, then $Z_t$ should be taken as the sum of the values or as their supremum, so that $V$ becomes a weighted proportion or weighted average. If a geometric decay kernel $w(t,t’)=\gamma^{,t’-t}$ with $0<\gamma<1$ is taken, then $Z_t=1/(1-\gamma)$ is a finite constant and the aggregation becomes the standard discounted sum. The several choices differ mathematically and differ normatively as well, because the choice of coefficient itself already embodies a judgment about the relative weight of the value of each period.

More generally, the corresponding form in continuous time is

$$
V(m_t) ;=; \frac{1}{Z_t}\int_{t}^{\infty} w(t,\tau),\text{value}(m_\tau),d\tau,
\qquad
Z_t ;=; \int_{t}^{\infty} w(t,\tau),d\tau,
$$

where the weighting kernel $w(t,\tau)$ reduced to a constant corresponds to the undiscounted integral, taken as $w(t,\tau)=e^{-s(\tau-t)}$ corresponds to exponential discounting, and taken in the hyperbolic form $w(t,\tau)=1/(1+k(\tau-t))$ corresponds to the hyperbolic discounting observed in behavioral economics.

This expression compresses at least three mutually independent difficulties.

The first is the infinite horizon and convergence. When the weighting kernel does not decay with $t’$, for instance $w\equiv 1$, the infinite sum and the normalization coefficient $Z_t$ diverge simultaneously, and the aggregate value takes the indeterminate form $\infty/\infty$, which constitutes no well-defined real number. Introducing a decay kernel is the necessary means of rendering $Z_t$ finite and thereby rendering the aggregation convergent, and the introduction of a decay kernel is equivalent to discounting, as detailed in Section 5.

The second is the finitude of computability. Even granting that the aggregate value is mathematically well-defined, an agent of bounded rationality cannot traverse all future branches so as to compute its value, which is the bounded rationality discussed by Simon [Simon 1955].

The third is the ill-definedness and mutability of value itself. Utilitarianism tacitly presupposes a stable, exogenous, and additive single-step value functional. Yet value itself is time-varying, is endogenous to the chosen path, and is shaped by prior choices, which is the situation characterized by adaptive preference [Elster 1983]. Under these conditions the above aggregation fails to hold mathematically, because the single-step value $\text{value}(m_{t’})$ depends on the path chosen from $t$ to $t’$, while that path in turn depends on the present choice that takes the aggregate value $V$ as its criterion. The definition of $V$ thereby contains a dependence upon $V$ itself, constituting an ill-founded circularity.

4.2 The Attempt at Integral Transformation and Its Diagnostic Significance

An intuition originating in engineering is to handle the infinite time domain by means of an integral transform. The following analysis may be offered.

The Fourier transform does not eliminate the infinite but merely relocates it. Applied in the strict sense to $\text{value}(\tau)$, the transform requires that it satisfy absolute integrability or square integrability, whereas in the undiscounted case $\text{value}(\tau)$ generally does not decay, so that it possesses a transform only in the sense of a tempered distribution [Lighthill 1958]. In this sense the total value over all time is the zero-frequency component in the frequency domain, $V(0)=\int_{-\infty}^{\infty}\text{value}(\tau),d\tau$. When this integral diverges, $V(\omega)$ exhibits a singularity at $\omega=0$, for example a term with a Dirac measure as coefficient. The scalar one seeks is precisely the value at that singularity, and therefore the transform has not eliminated the difficulty but has merely localized it at a single point of the frequency domain.

What truly renders the infinite sum convergent is the Laplace kernel. This kernel is the mathematical substance of exponential discounting. Its form is

$$
V(s) ;=; \int_{0}^{\infty}\text{value}(\tau),e^{-s\tau},d\tau,
\qquad s>0.
$$

For a $\text{value}(\tau)$ that grows at most exponentially at infinity, there exists an abscissa of convergence $\sigma_0$ such that the integral converges for $\operatorname{Re}(s)>\sigma_0$. Introducing the discount rate $s$ is equivalent to multiplying by the exponential decay window $e^{-s\tau}$, thereby forcing the divergent integral into convergence. From this an important judgment follows. The discount rate is not an empirical fact read off from preferences but a regularization parameter introduced in order to render the aggregate expression well-defined. The same structure appears in the form of a discount factor in the value functions of dynamic programming and reinforcement learning. The existence and uniqueness of the solution to the Bellman equation $V(m)=\text{value}(m)+\gamma\sum_{m’}P(m’\mid m),V(m’)$ depend on the contraction-mapping property secured by $\gamma<1$ [Bellman 1957; Puterman 1994], which again indicates that the role of the discount factor is in the first instance mathematical regularization rather than empirical description.

The significance of this attempt lies not in solving the problem but in precisely diagnosing the level at which the problem becomes insoluble.

5. The Non-Naturalizability of the Discount Parameter

Following the notation of the preceding section, $\gamma$ denotes the discount factor in the discrete case, with $0<\gamma<1$, and $s$ denotes the discount rate in the continuous case, with $s>0$, the two corresponding via $\gamma=e^{-s\Delta t}$.

5.1 The Decay Coefficient Treats Only the Symptom

Introducing a discount factor less than unity does render the weighted sum convergent, the scalar well-defined, and comparison possible. This treatment is complete at the technical level. Divergence, however, is not where the genuine difficulty lies. The genuine difficulty is that no objective fact can answer how great a weight the future ought to occupy. The discount factor is not something discovered but something stipulated. At the moment of stipulating it, the agent has already completed the very ethical judgment that utilitarianism sought to evade by computation, namely the adjudication of the rate at which future suffering is to be discounted against present suffering.

Substituting the same flow of utility into different discount factors reveals the divergence of consequences. When the discount factor approaches unity, the several generations are treated almost equally, which is the position taken by the Stern Review, whose policy implication is that climate action ought to receive heavy investment in the present [Stern 2007]. When the discount factor is markedly less than unity, the distant future is exponentially compressed, which is the position taken by Nordhaus, whose policy implication is that a gradual response suffices [Nordhaus 2007]. The two employ the same formula and the same data, yet reach opposite conclusions, their difference lying solely in the value taken by the discount factor. The substance of this divergence is an ethical judgment rather than an empirical measurement, and both Ramsey and Parfit hold that a positive rate of pure time preference is ethically difficult to defend [Ramsey 1928; Parfit 1984]. From this it is evident that the discount factor is not a parameter of the model but a conclusion of the model. Whatever conception of intergenerational justice the agent presupposes is the discount factor the agent reads back out. Utilitarianism promises an objective computation of the optimal scheme, yet the premise of that computation must be supplied precisely by the normative position it claims to be superseding. The ethical judgment, expelled at the front door, re-enters through the channel of the discount factor.

5.2 The Exponential Decay Form Is Itself an Undefended Ontological Commitment

The exponential decay form presupposes that time preference is geometric and time-invariant. Yet the actual intertemporal choices of agents exhibit the character of hyperbolic discounting, namely a very rapid decay in the near term and an almost negligible decay in the far term [Ainslie 1992; Frederick, Loewenstein & O’Donoghue 2002]. Hyperbolic discounting gives rise to preference reversal and time inconsistency, that is, the optimal path computed at the present is overturned when recomputed thereafter by the same criterion, a dynamic inconsistency first formally characterized by Strotz [Strotz 1955; Laibson 1997]. Exponential decay can maintain time consistency, but it is not an empirical truth, only the mathematical convenience selected in order to maintain consistency. Under these conditions even the utilitarian foundation that there exists a well-defined optimal solution available for computation is shaken.

5.3 The Humean Problem Encountered in Estimating the Discount Factor from Experience

In order that the discount factor not lapse into arbitrariness, one natural recourse is to estimate it from empirical data, that is, to observe historical intertemporal choices, including saving behavior, interest rates, and experimental discounting data, and on that basis to read back a behaviorally revealed discount factor, the methodological warrant for which is the theory of revealed preference established by Samuelson [Samuelson 1938]. This step touches simultaneously upon a double Humean problem.

The first is the problem of induction, the epistemological version of Hume [Hume 1748]. From the fact that discounting was thus in the past, one cannot infer that discounting will be thus in the future, nor that it ought to be thus in the future.

The second is the gap between the is and the ought, the ethical version of Hume [Hume 1739], which Moore terms the naturalistic fallacy [Moore 1903]. Even granting an accurate estimate of how people in fact discount, one cannot infer that people ought to discount thus. The myopia of people may belong precisely to what ought to be corrected rather than to what ought to be enshrined as a normative benchmark.

That the two belong alike to the Humean problem is no coincidence. Within the empirically given there is neither necessity nor oughtness. Induction intends to extract the former from experience and naturalism intends to extract the latter, and both fail at one and the same seam.

There is, furthermore, a still deeper difficulty, namely misspecification. The estimated discount factor is a point estimate obtained by forcibly projecting a flow of behavior that is time-varying, hyperbolic, and drifting with circumstance onto the model assumption of exponential decay and time invariance. Where the model specification is itself in error, parameter estimation converges not to the true parameter but to a pseudo-true value that minimizes, in a specified sense, the divergence between the specified model and the data [White 1982]. What it converges to is not the true discount factor but the constant that best fits the historical data under the assumption that it is a constant. The object of estimation is itself a fabrication of the estimation framework.

5.4 Summary

Taking the foregoing together, the discount factor has nowhere to flee. It cannot be stipulated, for to stipulate it is arbitrary. It cannot be computed, for without discounting it diverges. It cannot be estimated, for estimation encounters at once the double Humean problem and misspecification.

From this the core thesis of this paper follows. The core parameter of utilitarianism structurally resists naturalization. It is a normative quantity disguised as an empirical quantity. The genuine problem therefore shifts from how to compute the parameter correctly to a problem of an altogether different character, namely who has the standing, on what grounds, and answerable to whom, to make this decision of intertemporal trade-off. The latter problem lies beyond the reach of mathematics and belongs to the domain of political philosophy, bearing on intergenerational justice, the representation of the unborn, and the fiduciary responsibility owed by the present generation toward the future.

6. From Epistemological Limitation to Practical Freedom

6.1 The Unreachability of Real Norms and the Conditions for the Possibility of Action

A powerful philosophical tradition holds that real norms cannot be finally reached, that the agent can act only upon the beliefs it holds, and that otherwise practice cannot get underway. Kant’s primacy of practical reason [Kant 1788], Weber’s value rationality [Weber 1922], and, within the pragmatist tradition, Peirce’s account of belief and inquiry [Peirce 1877] together with James’s ethics of belief [James 1896], all point to one and the same judgment, namely that action cannot await the final guarantee of epistemology. Since the Humean seam ensures that this guarantee never arrives, acting upon belief constitutes not a cognitive concession but a condition for the possibility of practice itself.

This judgment dovetails precisely with the conclusion of Section 5. Since the discount factor resists naturalization, that which bears it can only be a believing and responsible commitment rather than a discovered fact.

6.2 An Inference Concerning Free Will and Its Disambiguation

One common inference runs as follows. Real norms are independent of the will of the agent; hence if one acts solely upon real norms the will plays no part; hence free will does not exist; hence ethics admits of no discussion. Within this chain of inference there lies a crucial equivocation, namely that the term real norm is assigned simultaneously to two distinct things.

The first is the descriptive real regularity, that is, natural law in the deterministic sense. The second is the normative moral reality, that is, the fact, independent of the agent, concerning what is right.

The relation of the two to free will is exactly opposite. Kant’s entire enterprise argues in the reverse direction [Kant 1785]. Its gist is that it is only because there exists a moral law not determined by natural causality, namely the fact of practical reason, that freedom becomes possible. Within the Kantian framework freedom is not equivalent to escaping norms but is equivalent to acting upon the law the agent legislates for itself, which is autonomy. That which is driven by the causal chain of natural desire is what Kant terms heteronomy, and heteronomy is what is unfree.

From this it is evident that what constrains freedom is not the existence of norms but the source of norms, namely whether a norm is imposed from without or legislated by reason itself. Acting upon moral norms therefore does not negate freedom but constitutes the form in which freedom is realized.

6.3 Epistemological Limitation as the Condition of Freedom

Bringing the foregoing disambiguation into the main line of this paper yields a position far more powerful than a resigned retreat. The fact that the discount factor and all norms cannot be naturalized and can obtain no real guarantee does not constitute a threat to freedom but constitutes the very space in which freedom can exist.

Were the discount factor capable of being fixed at one stroke by a real norm, the agent would possess no freedom whatever in intertemporal trade-off and could only transcribe that single correct answer. It is precisely because the real norm is unreachable that the position occupied by the discount factor is left vacant, to be borne by the will. The conclusion of Section 5 concerning the discount factor’s resistance to naturalization may be translated at the ethical level as follows: in the matter of intertemporal trade-off, reality does not make the decision on the agent’s behalf, and therefore the agent must, and can, make the decision itself.

Accordingly, the ending of the original chain of inference, namely that free will does not exist and therefore ethics is meaningless, should be replaced by a dual proposition. The unreachability of real norms preserves rather than abolishes the meaning of ethical discussion. Were everything fixed by real norms, then belief, responsibility, and ethical discussion would truly lose their meaning, for in that case nothing would depend on the agent.

7. Induction and Legislation, Description and Norm

The whole of the foregoing difficulties converges finally upon a single judgment, namely the conflation of induction with legislation and the conflation of description with norm. These two are not two parallel errors but two faces of one and the same error, the two sharing one and the same Humean seam.

The conflation of induction with legislation belongs to the epistemological side, namely the slide from things have repeatedly been thus to therefore they are necessarily thus or ought to be thus, which is what was encountered in Section 5.3 in estimating the discount factor. The conflation of description with norm belongs to the ethical side, namely the slide from things are in fact thus to therefore they ought to be thus, which is what was encountered in dismantling behaviorally revealed preference. The two are two cross-sections of the same Humean principle, their common ground being that within the empirically given there is neither necessity nor oughtness.

7.1 Guarding against Overcorrection: The Constraint of Description upon Norm

This insight harbors a dangerous slope. Once alert to the fact that description is not norm, one readily overcorrects into the view that description is wholly unrelated to norm, and proceeds to maintain that norms must be legislated out of pure reason from a void, thereby sliding into a free-floating normativism. The following qualification must be made. Although description does not entail norm, it does constrain norm. The principle that ought implies can shows that a norm one is in fact unable to discharge does not constitute a norm for the agent. Description is therefore not the premise of norm but the boundary condition of norm. Legislation cannot be read off from induction, yet it cannot disregard the space of possibility that induction delimits. To misread the non-derivability between the is and the ought as an irrelevance between them constitutes an error in the other direction. The two cannot be derived from each other, yet they must be calibrated against each other.

7.2 The Two Disguises of Norm

Vigilance against the conflation of induction with legislation severs one error, namely the passing off of an external regularity as a norm. Yet such vigilance cannot guard against the other error, namely the passing off of an internal and already-contaminated arbitrariness as autonomy. Self-deception too is an act of legislation. Whether the law legislated by a subject implanted by ideology, or shaped by adaptive preference, may still count as autonomy is itself in question.

The former error passes off the world is thus as a norm and is a norm disguised as objective. The latter error passes off I simply judge thus as a norm and is a norm disguised as subjective. Both are disguises of norm, their difference lying only in that one is disguised as objective and one as subjective.

7.3 The Position of Legislation That Has No Foundation and Yet Admits No Arbitrariness

Accordingly, the foregoing judgment should be tightened into a version at once more rigorous and more honest. A norm can neither be induced from the is of the world, on pain of falling into the naturalistic fallacy, nor be legislated arbitrarily from the agent’s subjective conviction, on pain of passing off willfulness as autonomy. A norm therefore occupies a peculiar position, namely that it has no foundation and yet admits of no arbitrariness.

This position is precisely the point at which the whole of this paper’s conclusions converges. The discount factor cannot be naturalized and therefore cannot be derived from description; the discount factor is likewise not freely fillable and therefore cannot rest purely upon the subjective. That position, which has no foundation and yet admits of no arbitrariness, is the core terrain over which the whole of practical philosophy contends. Kant attempts to fill that position with the principle of universalizability, so that arbitrariness is filtered by the form of universalization [Kant 1785]. Habermas attempts to fill that position with intersubjective consensus under the ideal speech situation, so that arbitrariness is filtered by communicative reason [Habermas 1981]. The success or failure of the two approaches may be contested further, but the problem has been driven to its very center.

8. Conclusion

This paper has unfolded its argument along a single thread, whose course is as follows. The argument set out from the mercantilist definition of wealth, first establishing the ontological status of wealth as a relational institutional fact, then identifying the threefold cost of the positivist undertone, then exposing the difficulties of infinitude and ill-definedness in the utilitarian intertemporal aggregation, then completing the non-naturalizability argument concerning the discount factor’s non-stipulability, non-computability, and non-estimability, then reinterpreting this epistemological limitation in reverse as the condition for the possibility of practical freedom, and finally arriving at the position of normative legislation, demarcated by the judgment against conflating induction with legislation and description with norm, that has no foundation and yet admits of no arbitrariness.

Its methodological import is that the four-level framework of ontology, epistemology, practical theory, and ethics set out at the opening closes its own loop at the terminus of the argument. The ontological inquiry into wealth is, by stepwise advance, compelled to be handed back to ethics and political philosophy. Its substantive conclusion may be condensed into a single judgment, namely that the genuine defect of utilitarianism lies not in its inability to complete the computation but in its having to pretend that the decision which only the agent can responsibly make can be outsourced to a formula. Any effort to outsource this burden, whether to mathematics, to data, or to how people in fact behave, will fail at some Humean seam.

9. Coda: An Exit Not Yet Resolved

If that which bears the discount factor is belief, then how is the harmful belief to be distinguished from the responsible belief. Harmful beliefs include self-deception, ideologically implanted preferences, and adaptive preferences. Kant draws the distinction by the principle of universalizability, yet that principle in turn presupposes a certain real standard the agent cannot reach. The problem therefore transforms into this: under the premise of no guarantee by a real norm, how is the free self-legislation to be distinguished from a heteronomy that has merely changed its outward appearance. This problem is precisely the return, at the ethical level, of the adaptive preference discussed in Section 4.1, and is the borderland between the theory of autonomy and the critique of ideology, where the tension is most pronounced among the self-critique of enlightenment [Horkheimer & Adorno 1947], the diagnosis of ideological distortion by communicative reason [Habermas 1981], and the psychoanalytic thesis that the subject is constituted by the symbolic order [Lacan 1966]. The paper halts here and leaves this problem as the entrance to subsequent reflection.

---

规范的不可自然化：从财富定义到实践立法的一条推理线

摘要

本文沿一条单一的推理线索，自重商主义对财富的定义出发，逐层抵达实践哲学中一项尚未获得解决的核心难题。论证依次确立如下命题。其一，财富并非自立实体，而是框架依赖的制度性事实。其二，重商主义因其实证主义与功利主义底色，承袭了休谟问题、异化与述行性等业已被辨识的限制。其三，功利主义的跨期效用加总依赖一个贴现参数，该参数在结构上拒绝被自然化，亦即它既不能被任意规定，亦不能被计算导出，更不能从经验中无歧义地估计。本文进而主张，此项认识论限制在伦理学层面应被重新诠释为实践自由的成立条件，而非对自由的否定。论证最终落于一处既无外在依凭、又不容任意的规范立法位置，并指出将归纳混同于立法、将描述混同于规范，乃同一道休谟接缝的两种表现。

关键词　政治经济学；功利主义；跨期贴现；休谟问题；自然主义谬误；实践理性；规范立法

1. 引论：一个领域无关的元框架

社会科学的任一概念，原则上均可置于一个领域无关的分析框架之下加以审视。该框架包含四个层次。本体论层次追问该概念属于何种存在，即它是实体、状态、属性、关系、过程抑或事件。认识论层次追问它如何被认识，以及它在何种解释框架之下方才显现。实践理论层次追问它如何在行动中被运用，以及它如何反作用于行动。伦理学层次追问它承载何种规范判断。

本文的方法论主张为，对一个陌生概念的有力分析，其来源往往不是对该领域既有论断的检索，而是以上述元框架对该概念施行的结构重建。下文以财富为入口，展示此种重建如何自本体论层次出发逐层递进，并最终在伦理学与政治哲学层次完成闭环。

2. 财富的本体论地位

2.1 重商主义的定义及其底色

重商主义将财富等同于货币或未铸成货币的贵金属，亦即金银存量。此定义在民俗层面存在一个镜像。传统婚俗中的彩礼与五金等安排，其实质是以流动性最高、贬值风险最低的形式储存并转移价值。此种安排是货币的价值贮藏职能与一般等价物职能在民间生活中的具体落实，而非一种反常现象。

该定义的理论底色可被迅速辨识。它带有鲜明的功利主义与实证主义性质。其优点在于可操作并可定量评估。其代价在于它一并承袭了实证主义与功利主义各自业已被辨识的全部限制。

2.2 财富作为制度性事实

财富的存在论地位应被刻画为关系性的、依赖性的存在，而非自立实体。其地位类同于债务、货币与主权，皆属制度性事实，即唯有在某物在特定语境中算作某种地位这一集体意向性建构之下方才成立 [Searle 1995]。一张钞票作为物理对象是自立的，其作为财富的资格则完全依附于一个认识论与制度的框架。

由此可对通常的表述加以强化。财富不仅在其被认识的层面依赖于框架，构成一种认识论依赖；它在其存在本身的层面亦依赖于框架，构成一种本体论依赖。战争使货币沦为废纸，其机制并非认识发生了改变，而是承载该货币的制度性事实发生了坍塌。

3. 实证主义底色的代价

重商主义为换取可操作性，须将财富物化为可观测且可计数的存量。此种物化的代价至少有三项。

第一项为休谟归纳问题。实证主义无法担保今日之金银于明日仍构成财富。此项限制的实质是价值的时间稳定性缺乏先验保证。

第二项为异化。关系性的善，例如爱、信任与承认，因不可量化而被排除于财富范畴之外。彩礼的货币化协商之所以损害情感，其原因在于它以一般等价物的逻辑度量本不属于交换领域之物。此点与桑德尔关于市场对道德的侵蚀的论证相通 [Sandel 2012]，亦与马克思关于货币使人的特性与自然的特性发生颠倒混淆的批判相通 [Marx 1844]。

第三项为述行性与反身性。经济概念区别于自然对象，会因被度量而改变。以国内生产总值度量财富，行动者随即转而为该指标本身而行动，此即古德哈特定律所刻画者 [Goodhart 1975]。经济学理论对其所描述的行为具有形塑作用，此即述行性命题 [MacKenzie 2006]。实证主义预设观测不扰动对象，而在社会科学中观测本身即构成干预。

4. 功利主义的跨期难题

4.1 无穷视界与价值的非良定义

若依功利主义计算某一对象在当下的效益，原则上须将其当下价值与未来各步所达状态的价值一并纳入考量。设 $m_t$ 为时刻 $t$ 之状态，$\text{value}(\cdot)$ 为单步价值泛函，$w(t,t’)\ge 0$ 为加权核，$Z_t$ 为规范化系数，则聚合价值的一般形式可写为

$$
V(m_t) ;=; \frac{1}{Z_t}\sum_{t’=t}^{\infty} w(t,t’),\text{value}(m_{t’}),
\qquad
Z_t ;=; \sum_{t’=t}^{\infty} w(t,t’).
$$

此处不应将规范化系数预先固定为 $1/N$。该系数的取法取决于聚合所欲表达的语义，且不同取法对应不同的本体论承诺。若 $w(t,t’)\equiv 1$ 而视界为有限的 $N$ 步，则 $Z_t=N$，聚合退化为均值，此为有限期望情形。若欲将聚合值规范至单位区间，则 $Z_t$ 应取价值之和或其上确界，使 $V$ 成为一种加权占比或加权平均。若取几何衰减核 $w(t,t’)=\gamma^{,t’-t}$ 且 $0<\gamma<1$，则 $Z_t=1/(1-\gamma)$ 为有限常数，聚合成为标准的贴现求和。各取法在数学上互异，在规范上亦互异，因为系数的选择本身即蕴含了对各期价值相对重量的判断。

更一般地，连续时间下相应的形式为

$$
V(m_t) ;=; \frac{1}{Z_t}\int_{t}^{\infty} w(t,\tau),\text{value}(m_\tau),d\tau,
\qquad
Z_t ;=; \int_{t}^{\infty} w(t,\tau),d\tau,
$$

其中加权核 $w(t,\tau)$ 退化为常数时对应无折扣积分，取 $w(t,\tau)=e^{-s(\tau-t)}$ 时对应指数贴现，取双曲形式 $w(t,\tau)=1/(1+k(\tau-t))$ 时对应行为经济学所观测的双曲贴现。

此表达式压缩了至少三重彼此独立的困难。

第一重为无穷视界与收敛。当加权核不随 $t’$ 衰减时，例如 $w\equiv 1$，无穷和与规范化系数 $Z_t$ 同时发散，聚合值取不定式 $\infty/\infty$，不构成良定义的实数。引入衰减核是使 $Z_t$ 有限从而使聚合收敛的必要手段，而衰减核的引入即等价于贴现，详见第五节。

第二重为可计算性的有限性。纵然聚合值在数学上良定义，具有限理性的行动者亦无法遍历所有未来分支以求其值，此即西蒙所论之有限理性 [Simon 1955]。

第三重为价值自身的非良定义与流变。功利主义暗中预设了一个稳定的、外生的且可加总的单步价值泛函。然而价值本身是时变的，是内生于选择路径的，并且会被先前的选择所塑造，此即适应性偏好所刻画的情形 [Elster 1983]。在此条件下，上述聚合在数学上不成立，因为单步价值 $\text{value}(m_{t’})$ 依赖于自 $t$ 到 $t’$ 所选择的路径，而该路径又取决于以聚合值 $V$ 为准则的当下选择，由此 $V$ 的定义中包含了对 $V$ 自身的依赖，构成一个缺乏良基的循环。

4.2 积分变换的尝试及其诊断意义

一个源于工程的直觉是，以积分变换处理无穷时域。对此可作如下分析。

傅里叶变换并不消除无穷，而仅迁移无穷。该变换以严格意义施于 $\text{value}(\tau)$ 要求其满足绝对可积或平方可积条件，而无折扣情形下 $\text{value}(\tau)$ 通常不衰减，故仅在缓增分布的意义上具有变换 [Lighthill 1958]。在此意义下，全时段总价值即频域中零频分量 $V(0)=\int_{-\infty}^{\infty}\text{value}(\tau),d\tau$。当该积分发散时，$V(\omega)$ 于 $\omega=0$ 处呈现奇点，例如以狄拉克测度为系数的项。所欲求取的标量恰为该奇点之值，因此变换并未消除困难，而仅将困难定位于频域的单一点上。

真正使无穷和收敛者为拉普拉斯核。该核即指数贴现的数学实质。其形式为

$$
V(s) ;=; \int_{0}^{\infty}\text{value}(\tau),e^{-s\tau},d\tau,
\qquad s>0.
$$

对于在无穷处至多指数增长的 $\text{value}(\tau)$，存在收敛横坐标 $\sigma_0$，使该积分于 $\operatorname{Re}(s)>\sigma_0$ 时收敛。引入贴现率 $s$ 等价于乘以指数衰减窗 $e^{-s\tau}$，从而将发散积分强制拉至收敛。由此可得一项重要判断。贴现率并非自偏好中读出的经验事实，而是为使聚合表达式良定义而被引入的正则化参数。同一结构在动态规划与强化学习的值函数中亦以贴现因子的形式出现，贝尔曼方程 $V(m)=\text{value}(m)+\gamma\sum_{m’}P(m’\mid m),V(m’)$ 的解之存在唯一性依赖于 $\gamma<1$ 所保证的压缩映射性质 [Bellman 1957; Puterman 1994]，此再次表明贴现因子的作用首先是数学正则化，而非经验描述。

此尝试的意义不在于解决问题，而在于精确诊断问题不可解所处的层次。

5. 贴现参数的不可自然化

沿用前节记号，下文以 $\gamma$ 指称离散情形下的贴现因子，满足 $0<\gamma<1$，以 $s$ 指称连续情形下的贴现率，满足 $s>0$，二者经 $\gamma=e^{-s\Delta t}$ 相互对应。

5.1 衰减系数所治者为表象

引入小于一的贴现因子确使加权和收敛，使标量良定义并可供比较。此项处理在技术层面是完备的。然而发散并非真正的困难所在。真正的困难在于，未来究竟应占据多大权重，没有任何客观事实能够回答。贴现因子并非被发现者，而是被规定者。在规定该因子之时，行动者已然完成了功利主义本欲以计算回避的那项伦理判断，即裁定未来之苦相对于当下之苦折算为何种比率。

将同一效用流代入不同的贴现因子，可见其后果之分歧。当贴现因子趋近于一时，各世代被近乎平等地对待，此为斯特恩报告所采立场，其政策含义为气候行动应于当下即投入重金 [Stern 2007]。当贴现因子明显小于一时，遥远的未来被指数级压缩，此为诺德豪斯所采立场，其政策含义为渐进应对即可 [Nordhaus 2007]。二者使用同一公式与同一数据，结论却截然相反，其差别仅在贴现因子之取值。此项分歧的实质为伦理判断而非经验测算，拉姆齐与帕菲特均主张纯时间偏好率的正值在伦理上难以辩护 [Ramsey 1928; Parfit 1984]。由此可知，贴现因子并非模型的参数，而是模型的结论。行动者预设何种代际正义，即反推出何种贴现因子。功利主义承诺对最优方案作客观计算，而该计算的前提恰须由它声称将予以取代的规范立场提供。伦理判断被自正门排除，复经贴现因子这一通道重新进入。

5.2 指数衰减形式本身为未经辩护的本体论承诺

指数衰减形式预设时间偏好为几何的且时不变的。然而行动者的实际跨期选择呈现双曲贴现的特征，即近期衰减极快而远期几近不衰减 [Ainslie 1992; Frederick, Loewenstein & O’Donoghue 2002]。双曲贴现导致偏好反转与时间不一致，亦即当下所计算的最优路径，于其后以同一准则重新计算时即遭推翻，此种动态不一致最早由斯特罗茨予以形式刻画 [Strotz 1955; Laibson 1997]。指数衰减能够维持时间一致性，但它并非经验真相，而仅为维持一致性所选取的数学便利。在此条件下，存在一个良定义的最优解以供计算这一功利主义根基亦随之动摇。

5.3 自经验估计贴现因子所遭遇的休谟问题

为使贴现因子不流于任意，一条自然的退路是自实证中估计该因子，即观察历史中的跨期选择，包括储蓄行为、利率与实验折现数据，并据以反推一个由行为揭示的贴现因子，其方法论根据为萨缪尔森所奠立的揭示偏好理论 [Samuelson 1938]。此步骤同时触及双重休谟问题。

其一为归纳问题，属休谟的认识论版本 [Hume 1748]。自过去如此贴现的事实，推不出未来将如此贴现，亦推不出未来应如此贴现。

其二为实然与应然之间的鸿沟，属休谟的伦理学版本 [Hume 1739]，摩尔称之为自然主义谬误 [Moore 1903]。纵然准确估得人们事实上的贴现行为，亦推不出人们应当如此贴现。人们的短视恰可能属于应被纠正之物，而非应被奉为规范基准之物。

二者同属休谟问题并非偶然。经验所予之中既不含必然性，亦不含应然性。归纳意图自经验榨取前者，自然主义意图自经验榨取后者，二者皆在同一接缝处失效。

此外尚有更深一层的困难，即设定误指。所估得的贴现因子，乃是将一段时变的、双曲的、随情境漂移的行为流，强行投影于指数衰减且时不变的模型假设之上所得的单点估计。在模型设定本身错误的前提下，参数估计所收敛者并非真实参数，而是使所设定模型与数据之间的偏离在特定意义下最小的伪真值 [White 1982]。它所收敛者并非真实的贴现因子，而是在假设其为常数的前提之下最能拟合历史数据的那个常数。被估计的对象本身系由估计框架所虚构。

5.4 小结

综合上述，贴现因子无处可逃。它不能被规定，因为规定即任意。它不能被计算，因为无折扣则发散。它不能被估计，因为估计同时遭遇双重休谟问题与设定误指。

由此得出本文的核心命题。功利主义的核心参数在结构上拒绝被自然化。它是一个伪装为经验量的规范量。真正的问题因此自如何将该参数算对，转化为一个性质全然不同的问题，即何者有资格、依据何种理由、并向何者负责地作出此项跨期权衡的决定。后一问题为数学所不能触及，它属于政治哲学的范畴，涉及代际正义、未出生者的代表权，以及当下世代对未来所负的受托责任。

6. 自认识论限制到实践自由

6.1 实在规范的不可抵达与行动的成立条件

一项有力的哲学传统主张，实在规范无法被最终抵达，行动者只能依据其所持的信念而行动，否则实践无从展开。康德的实践理性优先 [Kant 1788]、韦伯的价值理性 [Weber 1922]，以及实用主义传统中皮尔士关于信念与探究的论述 [Peirce 1877] 与詹姆斯的信念伦理 [James 1896]，皆指向同一判断，即行动不能等待认识论的最终担保。既然休谟接缝保证该担保永不到来，则依信念而行动便不构成认知上的退让，而构成实践本身的成立条件。

此判断恰好接续第五节的结论。既然贴现因子拒绝被自然化，则承担该因子者只能是一项信念性的、负责任的承诺，而非一项被发现的事实。

6.2 关于自由意志的一项推论及其辨析

一种常见的推论如下。实在规范与行动者的意志无关，故若仅依实在规范而行动则意志不起作用，故自由意志不存在，故伦理无可讨论。此推论链条之中存在一处关键的混淆，即实在规范一词被同时指派给两种不同之物。

第一种为描述性的实在规律，即决定论意义上的自然法则。第二种为规范性的道德实在，即独立于行动者的关于何者为正确的事实。

二者对自由意志的关系恰好相反。康德的整个工程正是反向论证的 [Kant 1785]。其要旨在于，唯因存在一个不被自然因果所决定的道德法则，亦即实践理性的事实，自由方才成为可能。在康德的框架中，自由不等于摆脱规范，而等于依据行动者为自身所立之法而行动，此即自律。被自然欲望的因果链所驱使者，康德称之为他律，他律方为不自由。

由此可知，限制自由者并非规范的存在，而是规范的来源，亦即规范究竟系自外部强加，抑或由理性自身立法。依据道德规范而行动因此不否定自由，反构成自由的实现形式。

6.3 认识论限制作为自由的条件

将上述辨析接入本文主线，可得一个较之无奈退守远为有力的立场。贴现因子以及一切规范无法被自然化、无法获得实在的担保，此项事实不构成对自由的威胁，反构成自由得以存在的空间。

倘若贴现因子能够被实在规范一举确定，则行动者在跨期权衡上将毫无自由，只能照录那唯一正确的答案。正因实在规范不可抵达，贴现因子所处的位置方才空出，以待意志承担。第五节关于贴现因子拒绝被自然化的结论，在伦理学层面可译述为，跨期权衡一事，实在不代行动者作出决定，故行动者必须并且能够自行作出决定。

据此，原推论链条的结尾，即自由意志不存在故伦理无意义，应被一个对偶的命题取代。实在规范的不可抵达恰好保全了伦理讨论的意义，而非取消该意义。倘若一切均被实在规范所确定，则信念、责任与伦理讨论方才真正丧失意义，因为彼时无一事取决于行动者。

7. 归纳与立法，描述与规范

上述全部困难最终汇聚于一项判断，即将归纳混同于立法，并将描述混同于规范。此二者并非两个并列的错误，而是同一错误的两副面孔，二者共享同一道休谟接缝。

将归纳混同于立法属于认识论一侧，即自过去反复如此滑向因而必然如此或应当如此，此即第 5.3 节估计贴现因子时所遭遇者。将描述混同于规范属于伦理学一侧，即自事实上如此滑向因而应当如此，此即拆解由行为揭示的偏好时所遭遇者。二者是休谟同一原理的两个切面，其共同根据为经验所予之中既无必然性亦无应然性。

7.1 对过度矫正的防范：描述对规范的约束

此项洞察存在一处危险的滑坡。一旦警觉于描述不等于规范，便易过度矫正为描述与规范全然无关，进而主张规范须凭空自纯粹理性立出，由此滑向一种悬空的规范主义。对此须作如下限定。描述虽不蕴含规范，却约束规范。应然蕴含能够这一原则表明，一个事实上无法履行的规范不构成对行动者的规范。因此描述并非规范的前提，而是规范的边界条件。立法不能自归纳读出，但不能无视归纳所划定的可能性空间。将实然与应然之间的不可推导关系误读为不相关关系，构成另一方向上的谬误。二者不可互相推导，却必须互相校准。

7.2 规范的两种伪装

警惕将归纳混同于立法，斩断了一种错误，即以外在的规律性冒充规范。然而此种警惕无法防范另一种错误，即以内在的、已被污染的任意性冒充自律。自欺亦是一种立法行为。被意识形态所植入的主体，或经适应性偏好所塑造的主体，其所立之法是否仍可计为自律，本身即成疑问。

前一种错误以世界本来如此冒充规范，是伪装为客观的规范。后一种错误以我即如此认为冒充规范，是伪装为主观的规范。二者皆为规范的伪装，其区别仅在于一者伪装为客观，一者伪装为主观。

7.3 既无依凭又不容任意的立法位置

据此，前述判断应被收紧为一个更为严格亦更为诚实的版本。规范既不能自世界之实然中归纳而出，否则即陷入自然主义谬误；亦不能自行动者之主观认定中任意立出，否则即以任性冒充自律。规范因此处于一个特殊的位置，即它无所依凭，却又不容任意。

此位置正是本文全部结论的汇聚点。贴现因子不能被自然化，故不能自描述导出；贴现因子亦非可任意填充，故不能纯然取决于主观。该既无依凭又不容任意的位置，即整个实践哲学所争夺的核心地带。康德以可普遍化原则试图填充该位置，使任意性为普遍化的形式所过滤 [Kant 1785]。哈贝马斯以理想言谈情境之下的主体间共识试图填充该位置，使任意性为交往理性所过滤 [Habermas 1981]。两条进路的成败均可进一步争论，但问题已被逼至其正中央。

8. 结论

本文沿一条单一线索展开论证，其行程如下。论证自重商主义的财富定义出发，先确立财富作为关系性制度事实的本体论地位，继而辨识实证主义底色的三重代价，再揭示功利主义跨期加总的无穷与非良定义难题，进而完成关于贴现因子不可规定、不可计算、不可估计的不可自然化论证，复将此项认识论限制反转诠释为实践自由的成立条件，最终落于将归纳混同于立法、将描述混同于规范这一判断所标定的、既无依凭又不容任意的规范立法位置。

其方法论寓意在于，开篇的本体论、认识论、实践理论与伦理学四层框架，在论证终点处自行闭环。对财富的本体论追问，经由逐层递进，被迫交还于伦理学与政治哲学。其实质结论可凝练为一句判断，即功利主义的真正缺陷不在于无法完成计算，而在于它须假装那个只能由行动者负责作出的决定可以被外包于公式。任何将此项承担外包出去的努力，无论外包于数学、外包于数据，抑或外包于人们事实上的行为，均将在某一道休谟式的接缝处失效。

9. 余论：一处尚未解决的出口

若承担贴现因子者为信念，则有害的信念与负责任的信念之间应如何区分。有害的信念包括自欺、被意识形态植入的偏好与适应性偏好。康德以可普遍化原则为界加以区分，然而该原则又预设了某种行动者无法抵达的实在标准。问题因此转化为，在缺乏实在规范担保的前提之下，如何区分自由的自我立法与一种更换了外观的他律。此问题恰是第 4.1 节所论适应性偏好在伦理学层面的回归，亦是自律理论与意识形态批判的交界地带，在该地带启蒙的自我批判 [Horkheimer & Adorno 1947]、交往理性对意识形态扭曲的诊断 [Habermas 1981]，以及主体由象征秩序所构成的精神分析论题 [Lacan 1966] 之间的张力最为显著。本文于此止步，将该问题留作后续思考的入口。

概念索引

主题	关键概念	相关思想者
财富本体论	制度性事实、本体论依赖、关系性存在	Searle；Marx
实证主义代价	物化、异化、述行性、古德哈特定律	Marx；Sandel
跨期难题	无穷视界、双曲贴现、时间不一致、适应性偏好	Simon
贴现的数学实质	拉普拉斯核、正则化、贴现率之争	Stern；Nordhaus；Ramsey；Parfit
双重休谟问题	归纳问题、实然与应然鸿沟、自然主义谬误、设定误指	Hume；Moore
实践自由	实践理性优先、自律与他律	Kant；Weber；Peirce；James
规范立法位置	可普遍化、交往理性、自律与意识形态	Kant；Habermas；Adorno；Lacan

本文系对一段自由思考的整理。其论证均由第一性原理重建。所标注之思想者旨在为论题定位文献坐标，而非主张本文文本以其为依据。

参考文献

Ainslie, G. (1992). Picoeconomics: The Strategic Interaction of Successive Motivational States within the Person. Cambridge: Cambridge University Press.

Bellman, R. (1957). Dynamic Programming. Princeton: Princeton University Press.

Elster, J. (1983). Sour Grapes: Studies in the Subversion of Rationality. Cambridge: Cambridge University Press.

Frederick, S., Loewenstein, G., & O’Donoghue, T. (2002). Time Discounting and Time Preference: A Critical Review. Journal of Economic Literature, 40(2), 351–401.

Goodhart, C. A. E. (1975). Problems of Monetary Management: The U.K. Experience. In Papers in Monetary Economics, Vol. 1. Sydney: Reserve Bank of Australia.

Habermas, J. (1981). Theorie des kommunikativen Handelns. Frankfurt am Main: Suhrkamp.

Horkheimer, M., & Adorno, T. W. (1947). Dialektik der Aufklärung. Amsterdam: Querido.

Hume, D. (1739). A Treatise of Human Nature. London: John Noon.

Hume, D. (1748). An Enquiry Concerning Human Understanding. London: A. Millar.

James, W. (1896). The Will to Believe. The New World, 5, 327–347.

Kant, I. (1785). Grundlegung zur Metaphysik der Sitten. Riga: Hartknoch.

Kant, I. (1788). Kritik der praktischen Vernunft. Riga: Hartknoch.

Lacan, J. (1966). Écrits. Paris: Éditions du Seuil.

Laibson, D. (1997). Golden Eggs and Hyperbolic Discounting. Quarterly Journal of Economics, 112(2), 443–477.

Lighthill, M. J. (1958). An Introduction to Fourier Analysis and Generalised Functions. Cambridge: Cambridge University Press.

MacKenzie, D. (2006). An Engine, Not a Camera: How Financial Models Shape Markets. Cambridge, MA: MIT Press.

Marx, K. (1844). Ökonomisch-philosophische Manuskripte aus dem Jahre 1844. （1932 年首次刊行。）

Moore, G. E. (1903). Principia Ethica. Cambridge: Cambridge University Press.

Nordhaus, W. D. (2007). A Review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature, 45(3), 686–702.

Parfit, D. (1984). Reasons and Persons. Oxford: Oxford University Press.

Peirce, C. S. (1877). The Fixation of Belief. Popular Science Monthly, 12, 1–15.

Puterman, M. L. (1994). Markov Decision Processes: Discrete Stochastic Dynamic Programming. New York: Wiley.

Ramsey, F. P. (1928). A Mathematical Theory of Saving. The Economic Journal, 38(152), 543–559.

Samuelson, P. A. (1938). A Note on the Pure Theory of Consumer’s Behaviour. Economica, 5(17), 61–71.

Sandel, M. J. (2012). What Money Can’t Buy: The Moral Limits of Markets. New York: Farrar, Straus and Giroux.

Searle, J. R. (1995). The Construction of Social Reality. New York: Free Press.

Simon, H. A. (1955). A Behavioral Model of Rational Choice. Quarterly Journal of Economics, 69(1), 99–118.

Stern, N. (2007). The Economics of Climate Change: The Stern Review. Cambridge: Cambridge University Press.

Strotz, R. H. (1955). Myopia and Inconsistency in Dynamic Utility Maximization. Review of Economic Studies, 23(3), 165–180.

Weber, M. (1922). Wirtschaft und Gesellschaft. Tübingen: Mohr.

White, H. (1982). Maximum Likelihood Estimation of Misspecified Models. Econometrica, 50(1), 1–25.

中文版

一

有时候，按计划应该完成的事情，会被一些计划之外的小事打断。比如走进图书馆之前，注意到门口的一朵花，蹲下来看了一会儿，半天就过去了。

这件小事或许值得展开。它牵涉到一个更大的问题：在一个描述工具空前丰富的时代，我们真正能听见的东西，是不是反而越来越少？

二、关于”目的过滤”

受过严格学术训练、又在AI时代成长起来的人，或许都会逐渐养成一种习惯，可以试着称它为目的过滤。

读书之前先问，这本书对我的研究有用吗。做事之前先问，这件事会带来什么产出。甚至与人相处之前都会有一点这样的反射，这次互动的意义在哪里。

这种过滤当然有它的功能，让人能在有限的时间里做更多的事。但它也可能有一个不太被注意到的代价，让人越来越不容易被任何事物真正触动。

因为真正能触动我们的东西，往往超出我们事先计划要去找的范围。它需要绕过我们的目的体系，才能到达我们。而一个把目的过滤训练得很熟练的人，这个过滤已经成为反射，比意识更快。一本书还没翻开，已经被判为”暂时不相关”。一朵花还没看清，已经被归为”路过的景观”。

我们以为我们在节省时间。
但同时，我们可能也在节省自己的存在。

Recently, while reflecting on the methodology of “evidence-based decision-making” (EBM), I became aware of an interesting philosophical structure.

English

EBM essentially belongs to positivist social science: it extracts regularities from patterns observed in the past and uses them as the basis for future judgment. But this framework has two important built-in limitations that improvements in method cannot fundamentally eliminate. These two limitations are not confined to EBM; they are something that all positivism-based epistemologies and methodologies of social science must think through and confront.

The first limitation is the Humean problem. In A Treatise of Human Nature (1739), Hume pointed out that there is neither logical nor empirical guarantee that causal relations observed in the past will continue to hold in the future. EBM cannot, in principle, escape this problem. It is especially severe for practices that aim at creating new domains or at disruptive innovation. Innovative action is, by definition, a deviation from past evidence.

Source document: Draft Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of Regulation (EU) 2024/1689, issued by the European Commission’s Directorate-General for Communications Networks, Content and Technology (DG CONNECT), through the Artificial Intelligence Office.

Nature of the document: The Guidelines are non-binding. Authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice of the European Union (CJEU). The present version remains in the stakeholder consultation stage, and the final text is yet to be adopted by the Commission.

Introduction (added by the present author)

Regulation (EU) 2024/1689 (hereinafter the “AI Act”) constitutes the first horizontally applicable comprehensive legislative framework for artificial intelligence at the supranational level. Within its risk-based regulatory architecture, Article 50 occupies a position intermediate between the “high-risk” tier and the “minimal-risk” tier. The provision does not prohibit any technical use, nor does it impose an ex ante conformity assessment procedure. Rather, it employs information disclosure as a comparatively restrained regulatory instrument designed to redress the informational asymmetry that obtains between AI systems and the natural persons exposed to them.

The present notes are structured as follows. Sections one to three constitute additions by the present author, addressing in turn the regulatory problem at which Article 50 is directed, the legislative motivations and normative aims of the provision, and the antecedent normative materials with which it interacts. Section four summarises the original Guidelines section by section, in the order of the source document. Section five provides practical implications and preliminary commentary by the present author on issues arising from the systematic interpretation of the provision.

Part One: The Regulatory Problem (added by the present author)

The regulatory problem to which Article 50 is addressed can be articulated at three distinct levels.

1. The asymmetry between generative capability and human discernment

As generative AI systems approach and in some cases exceed the threshold at which natural persons can rely on intuitive judgement to distinguish synthetic content from authentic content across the four modalities of text, image, audio, and video, the “native discernment capacity” enjoyed by individuals has been significantly devalued. This phenomenon implicates not only individual-level cognitive bias but also the credibility of the information ecosystem as a whole. Article 50 does not seek to impose a general prohibition on AI outputs that are indistinguishable from human-produced content. Rather, it requires that, through marking, disclosure, and detection mechanisms, natural persons be equipped with the metadata necessary to make informed choices regarding the trust they extend to such content.

2. The diffusion of responsibility along the AI value chain

The AI value chain involves multiple tiers, including model providers, system providers, deployers, distribution platforms, and end users. Specific harms such as impersonation, deception, and manipulation may arise at any point along this chain, yet no single actor is in a position to undertake compliance responsibility for the entire chain. The regulatory technique of Article 50 lies in allocating obligations to the actors most capable of implementing the relevant measures: the provider for technical measures (marking and detection), and the deployer for disclosure and labelling in the context of dissemination. Auxiliary actors along the chain (in particular online platforms) are addressed through connecting provisions that encourage coordinated conduct.

3. Cross-modal and cross-contextual regulatory granularity

The technical maturity of watermarking and detection techniques varies significantly across textual, visual, auditory, and video modalities. The risk profile of a given system also varies markedly across contexts, with the risks posed by a non-player character in a video game differing substantially from those posed by a politically charged deepfake on a social platform. Article 50 must therefore strike a balance between generality of application and avoidance of regulatory over-reach. It does so by introducing a substantial set of exceptions, including the “obviousness” standard, the exception for assistive functions of standard editing, the attenuated disclosure regime for artistic and creative works, the law enforcement exception, the exclusion for purely personal non-professional activities, the exclusion for scientific research and development, and the exception for human review under editorial responsibility.