NOTE-1300: The Significance, Persistence, Contingency Framework

Contribution

A formal version of the heuristic that organises What We Owe the Future. To judge how much an act matters for the long run, ask how valuable the state of affairs it brings about is while it lasts, how long it lasts, and how far its presence depends on the act. The note shows that this three-way split is an exact factorisation of the change in value, gives marginal and expected-value versions, sketches an alternative, and places the result inside the effective-altruism importance–tractability–neglectedness framework.

Key insight

Contingency is what attributes value to a chooser. Significance and persistence describe the world; contingency asks “how much of the feature’s value can be attributed to a particular agent’s decision at a particular point in time” (p. 1). An act is worth what the world would lack without it. The framework’s “main goal is to help estimate the instrumental value (typically interpreted as the expected value) of an event, compared to the relevant counterfactual” (pp. 1–2).

Assumptions

• Expected-value maximisation is “the fundamental rule for choice under uncertainty” (§2.2, p. 5). It is assumed, not argued.
• Value comes in additive features. Overall evaluation sums features, Vtot = VF1 + VF2 + … (eq. 19, p. 9). The authors grant that such a list “may often be difficult to identify” (p. 9).
• Durations are finite. The book’s note (2.18) adds that an end date of the universe should be assumed, “otherwise we would have to deal with some states of affairs being infinitely persistent”.
• No theory of value. A feature is “any source of value (or disvalue)”: a headache, the future existence of civilisation, the prevalence of QWERTY keyboards (n. 1, p. 1). What is valuable, and for whom, is left open.

Key results

• The basic framework (§2, pp. 2–4). With ΔV and ΔT the changes E makes to F’s total value V and duration T, relative to the salient alternative or status quo:
  • Sig = ΔV/ΔT, Per = T, Con = ΔT/T (eq. 1), so ΔV = Sig × Per × Con (eq. 2).
  • It “make[s] most intuitive sense” when E’s main effect is to lengthen F’s presence without changing its value while it would have been present anyway (p. 3). Con is 1 when F would not exist but for E, 0 when E makes no difference to its duration.
  • Worked example: F contributes 1, 2, 4, 8 in four periods; it would be present in periods 2–3, and E adds periods 1 and 4. Then Per = 4, Con = 1/2, Sig = 9/2, ΔV = 9 (p. 3). Sig is neither the status-quo average (6/2) nor the average given E (15/4).
• Marginal form (§2.1, p. 4). For a marginal change in a parameter X, such as funding, Con = %dT/dX, the semi-elasticity of persistence.
• Uncertainty (§2.2, pp. 4–6). Either compute ΔV state by state and take its expectation, or use ex ante factors: Sig = ΔEV/ΔET, Per = ET, Con = ΔET/ET (eq. 6). These keep the product, but Sig is not expected significance and Con is not expected contingency, since “the expectation of a product is not usually equal to the product of expectations” (p. 5). The book’s note (2.30) says the same and adds that for its purposes this “will not be hugely important”.
• An alternative framework (§3, pp. 6–8).
  • Sig′ = V/T, independent of the counterfactual; Per′ = T; Con′ = ΔV/V, or %dV/dX at the margin (eqs. 10–12).
  • At the margin, Con′ = %dSig′/dX + %dPer′/dX (eq. 14): contingency is the sum of the semi-elasticities of significance and persistence.
  • The probability that F is present can be split out as Prob, so that EV = (EFV/EFT) × EFT × Prob and Con′ adds %dProb/dX (eqs. 15–16). The example given: “small changes in the probability that human civilisation continues beyond this millennium are often thought to be a significant source of value” (p. 7).
  • The two frameworks agree when F contributes a fixed value per unit time unaffected by E (p. 6).
• ITN (§4, pp. 8–9). Importance of marginal progress X on a problem is Imp = dEVtot/dX (eq. 18), counting all downstream effects. SPC estimates dEVF/dX feature by feature, and with additive features Imp is their sum (eq. 20). The book’s note (2.17) puts it as “a quantity proportional to the ‘importance’ dimension”.
• Tractability and leverage (§4.1, pp. 9–11).
  • Cotton-Barratt’s decomposition, dX/dW = (dX/%dW) × (1/W), takes neglectedness as 1/W. It is most intuitive with logarithmic returns; with linear returns his tractability grows with the work already done, which “could be somewhat counterintuitive” (p. 11).
  • The alternative: dX/dW = (X0/W0) × [(dX/dW)/(X0/W0)] (eq. 23). The first factor is the problem’s overall easiness, the second “leverage”, how much easier progress is at the current margin than on average. High leverage “does not imply that little work is being done” (p. 11).
• What the book’s notes add (second hand, by note number).
  • The framework “was created by Aron Vallinder and me and further developed by Teruji Thomas”, and appears in the book’s Appendix 3 (2.17). That appendix uses worlds X and O, a QWERTY–Dvorak example and the notation Vs(p), Ts(p) (A.4–A.6).
  • Note (A.6) concedes a gap: the requirement that a state’s value depend only on how long the world is in it “should perhaps be added to the definition”, or the definitions of significance and contingency make little sense.
  • The book’s ITN formalisation, “suggested to me by Teruji Thomas”, makes tractability depend only on the problem, not on the work already done (A.9). That matches the report’s easiness factor X0/W0.
  • “Trajectory change” is narrowed to lasting changes in civilisation’s average value, apart from changes to its duration (2.20), which the book calls “civilisational safeguarding” (2.22).

Claims

id	claim	strength	support
C1	The change in value an event makes through a feature equals significance × persistence × contingency	demonstrated: an identity from the definitions	§2, eqs. 1–2
C2	The decomposition is most intuitive when the event’s main effect is on the feature’s duration	interpretive; the book’s note (A.6) concedes a restriction may be needed	§2, p. 3
C3	Under uncertainty, ex ante factors keep the product, but are not the expectations of the factors	demonstrated	§2.2, eqs. 6–9
C4	Taking significance as V/T makes contingency the sum of semi-elasticities of significance, persistence and probability of presence	demonstrated: follows from log V = log Sig′ + log Per′	§3, eqs. 10–17
C5	SPC estimates, summed over features, give the importance factor of ITN	conditional on additive features, which the authors grant are hard to find	§4, eqs. 18–20
C6	Tractability is better read as overall easiness times leverage than as Cotton-Barratt’s dX/%dW in some cases	argued from two returns curves; offered as an option	§4.1

Concepts

• feature: any source of value or disvalue in the world, from a headache to the existence of civilisation.
• originating event: the choice or other event whose contribution is being valued.
• significance: average value per unit time, of the extra time due to E (§2) or of the feature given E (§3).
• persistence: how long the feature lasts, given E.
• contingency: how far the feature’s duration (§2) or value (§3) depends on E; at the margin, a semi-elasticity.
• semi-elasticity: the change in a quantity as a proportion of itself, per unit change in X; written %d/dX.
• leverage: marginal progress per unit of work relative to the average for the whole problem; a form of neglectedness.
• trajectory change / civilisational safeguarding (book’s notes): raising civilisation’s average value, versus extending its life span.

Connections

• What We Owe the Future (LIT-1747), unread. §2 explains “the framework as it appears in” the book (p. 2), so this is the book’s own accounting device. The notes show it organising chapter 2 and Appendix 3. It is not the book’s argument that future people count, nor its population ethics.
• Greaves and MacAskill (NOTE-1302). The same expected-value framing, with a benefit ratio in place of a factorisation. The book’s notes place the book short of strong longtermism: a critique by Mogensen “only applies to versions of longtermism on which our obligation to help future people trumps all other self- and other-regarding reasons for action, which is not the view I’m defending in this book” (1.50).
• Tarsney (NOTE-1247) and Thorstad (NOTE-1308). In SPC terms, and as an inference, their challenges fall on contingency and persistence: Tarsney’s decay of predictable influence lowers how much of a far-future feature depends on a present act, and Thorstad’s high background risk shortens expected persistence.
• Bostrom (NOTE-1314, NOTE-1321). The alternative framework’s Prob term is where reducing existential risk enters. MacAskill’s notes give his own credences: end of civilisation this century 0.1 to 1 percent, misaligned AI takeover about 3 percent, value lock-in at some point over 80 percent; and the expected value of civilisation’s continuation under 1 percent of the best feasible future, so that “trajectory changes have over one hundred times greater potential upside than civilisational safeguarding” (2.22).
• Syropoulos et al. 2023 (filed with this reading). Its Study 4 used a 385-word excerpt from the book’s introduction, the section “Future People Count”, as an intervention. It raised felt responsibility to future generations, by the smallest margin of five interventions.

Bearing on the record

• The owner’s thesis (CLAIM-491). These are inferences; the note does not discuss motive.
  • The framework values states of the world, not anyone’s good as theirs. Nothing in it is indexed to the agent except contingency, which attributes value to the agent’s choice. So the book’s ground, as formalised here, is the impersonal good. As QUESTION-299 already says, the thesis bears on such a ground only as a question of who is moved by it.
  • Contingency has the shape of a difference-making motive: what matters is the value that would be missing without me. If people are moved by their own contribution rather than by the value alone, a self-regarding element sits inside an impartial calculus. The report neither asserts nor tests this. GPI lists Greaves, Thomas, Mogensen and MacAskill, “On the desire to make a difference” (Philosophical Studies, 2024), not held, which is where the question is taken up.
  • The book’s notes name a ground the question does not list: reciprocity. “We benefit enormously from the actions of people in the past … Perhaps, then, this gives us reasons to ‘pay it forward’” (1.5). Syropoulos et al.’s sacrifice condition is a test of it.
• The triviality objection (CLAIM-492). Not addressed.
• Batson’s objection (CLAIM-494). Not addressed.
• A dated note is proposed for QUESTION-299, not for the claim.
• No THEORY filed.
• No ML instruction.

Limitations

• Not the book. Eleven pages of algebra. It does not argue that future people count, that there may be very many of them, or anything of population ethics, lock-in or extinction.
• The notes are not the chapters. They show sources, numbers and asides. What the chapters argue is inferred from them.
• One toy example. Applications to real events are left to the book (p. 2).
• Additivity of features is needed for the ITN step and conceded to be hard to meet (p. 9).
• Expected-value maximisation is assumed. The book’s notes grant a problem of “tiny probabilities of enormous amounts of value” (2.22).
• Version and date. Version 1.1 of August 2022, posted in June 2023. Version 1.0 was not found.
• Typos.
  • “occurence” (p. 2).
  • “tractabilility” (n. 4, p. 10).
  • In the book’s notes, (A.6) prints “Ts(p) − Ts(p)”, which must mean Ts(p) − Ts(q), q being the status quo (A.7).
  • Note (9.3) reads “Waiting for reference”.

Open questions

• Is a desire to make a difference, to maximise one’s own contingency, at odds with caring about the value itself? If so, which one moves people who say they act for the far future?
• Does the book’s Appendix 3 add the restriction note (A.6) proposes, or adopt §3’s alternative? Only the appendix’s notes were seen.