Founder of the non-profit Khan Academy and of Schoolhouse.world, and a leading proponent of free online learning and AI tutoring.
Sal Khan’s slice of Factrail’s verified causal web — the facts, drivers and welfare indicators their actions connect to. Select any node to trace a path.
Loading network…
Projected scenarios from the Factrail model. These describe what may happen under stated assumptions — they are not confirmed facts and may change as new data arrives.
Horizon: Jan 1, 2027 – Jan 1, 2030
Under the baseline path, the global learning-poverty rate slowly recedes from its 2022 peak of 70% as post-pandemic recovery spending and the lagged dividends of recent financing reforms (Incheon, FUNDEB) take hold, but it stays far above the SDG 4 norm line of 0% and well above the World Bank's halve-by-2030 ambition. The recovery is constrained by near-flat teacher quality and supply and uneven digital access.
Assumptions
Assumes no new global education shock on the scale of the 2020 closures; that recent financing reforms hold and partially translate into teacher supply and materials over the model's multi-year lags; and that digital access remains unequal and only a weak contributor. Treats the funding-to-learning-poverty link (medium confidence, ~5-year lag) and teacher-quality link (medium confidence, ~4-year lag) as the dominant recovery channels.
This is a projected scenario, not a confirmed fact.
Updated
A chronology will appear once enough dated facts are linked.
No affiliated people are linked yet.
Sal Khan appears in the Factrail dataset as the founder of Khan Academy, tracked for two related actions that share a single theme: extending access to AI-assisted instruction. The entry is a useful test of the platform's discipline, because Khan is simultaneously a documented expander of educational access and an interested party promoting his own organization's tools. The model holds both of those facts in view at once, crediting what is recorded while explicitly refusing to convert availability into a claim about learning.
The first anchoring fact is the 2023 launch of Khanmigo, framed as a personal AI tutor for every student, introduced alongside a widely viewed TED talk in which Khan argued that AI could finally make individualized tutoring scalable. The second is the 2024 Microsoft partnership that made Khanmigo for Teachers free to all US K-12 educators. Both are classified as initiatives, both carry a verified status, and both sit at a medium confidence level. They are recorded as concrete steps that broadened the availability of an AI tutoring and teaching-assistant tool — not as evidence of any particular outcome in classrooms.
The distinction between the two actions matters for how the model weights them. The 2023 launch is treated primarily as a contribution to learning access; the 2024 step, by removing cost for the teacher-facing tool, additionally and more lightly credits teacher capacity. That is why the dataset connects Khan to two different drivers rather than one.
Factrail routes the 2023 launch through the Digital Learning Access driver, an infrastructure-and-technology lever carried at a current weight of 0.55. The 2024 free-for-teachers step touches both that driver and Teacher Quality and Supply, a human-capital driver weighted at 0.6. The logic is that a free or low-cost AI tutor plausibly widens access to individualized instruction, and a free teacher-facing assistant plausibly supports the people delivering it. Both driver weights are deliberately moderate, encoding the model's caution that broadening availability is a necessary but not sufficient condition for educational gains.
From the drivers, the chain reaches three welfare indicators. The Learning Poverty Rate for low- and middle-income countries — the share of 10-year-olds unable to read a simple text, where lower is better — carries the highest importance weight at 0.92. Out-of-School Rate at primary age, also lower-is-better, and PISA mathematics performance at the OECD average, where higher is better, round out the set.
The grounding records nine rating-impact rows, and they split into a positive cluster and a smaller negative cluster. The strongest positive contribution, at an impact value of roughly +0.14, comes from the 2023 Khanmigo launch acting through Digital Learning Access onto the learning-poverty indicator. Because learning poverty is a lower-is-better measure, a positive welfare contribution means downward pressure on the share of children who cannot read — the favorable direction. The 2024 free-for-teachers step contributes a similar positive impact of about +0.13 through the teacher-supply driver onto the same indicator, with several smaller positive rows reaching the out-of-school-rate indicator.
The negative rows are the more interesting part of the entry, because they reveal the model's restraint. The contributions to PISA mathematics performance are recorded as negative, with values around -0.12 and -0.11. This should not be read as a claim that Khanmigo lowered test scores. Rather, it reflects the sign conventions of the driver-to-indicator factors on those particular links combined with how the deviation factor is applied; the model is registering that the chain from a tutoring tool to a measured OECD assessment outcome is weak and not in a direction it is willing to credit as a clear gain. The honest summary is that the favorable signals run to access and learning-poverty indicators, while the link to a hard, externally measured outcome like PISA is treated as unproven.
Across the rows the calculation inputs are consistent with a hedged reading: contribution-size factors of 0.6 to 0.7, responsibility factors of 0.6 to 0.7, fact-impact factors between 0.4 and 0.6, and confidence modifiers around 0.68 to 0.8. None of these approach the ceiling, which is the model's way of saying the evidence supports a tendency rather than a measured effect.
The most important caveat is one the entry states plainly: availability is not the same as demonstrated learning gains, and the independent evidence on AI tutoring's classroom impact is still developing. The chain to welfare indicators is therefore treated as a tendency, not a result. Equally, the model registers that Khan is promoting his own organization's tools, which is exactly the kind of conflict that should pull confidence and responsibility factors down rather than up.
The records are also narrowly scoped. They cover the Khanmigo initiatives specifically and do not attempt to assess Khan Academy's full historical reach or to adjudicate the broader debate over AI in classrooms. That scoping is a feature: it lets the model credit a documented expansion of access without overstating what that expansion has so far been shown to achieve.
Khan's entry is a clean illustration of how Factrail separates a confirmed act from an uncertain consequence. The two initiatives genuinely widened access to an AI tutor, and that is recorded as a positive contribution to the access and learning-poverty side of the graph. But the model deliberately leaves the question of measured effectiveness open, declining to convert a promising tool and an interested founder's advocacy into a welfare gain the data does not yet support. The result is a profile that is encouraging where the evidence allows and silent where it does not.