Reliable and not misleading

From Tournesol
Revision as of 18:44, 27 September 2021 by Aidjango (talk | contribs) (add logo)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

"Reliable and not misleading" is an optional quality criterion of Tournesol 🌻.

Reliability and not misleading logo

It aims to measure the trustworthiness of the information presented in a content, but also the extent to which the information may be misleading. In particular, a factual content that cherry-picks data to present to corroborate a world view and lacks completeness should considered partially misleading. Similarly, contents that lack epistemic prudence and suffer overconfidence may be considered less reliable than contents that adequately measure the extent of their ignorance.

Contents that score high on "reliable and not misleading" should make nearly all viewers improve their global world view, despite viewers' biases and motivated reasoning.


The multiplicity and diversity of sources and evidence plays an important role in the reliability of a content. It should ideally be also backed by peer-reviewed scientific publications AmericanChemicalSociety-19. However, as explained in this video by Veritasium, even published research usually has major shortcomings. More reliable information can be obtained by meta-analyses of multiple scientific publications, or by drawing on scientific consensus.

Even then, even meta-analysis may not be conclusive DeVrieze-18, especially when data are not compelling enough. A reliable analysis of the data should acknowledge uncertainty WassersteinSL-19 by showing epistemic prudence, thereby combatting overconfidence. Ideally, the level of certainty should be well calibrated.


Unfortunately, factual news can be deeply misleading. As an example, on October 21st, 2020, NBC News published a news entitled "Volunteer in AstraZeneca Covid-19 vaccine trial dies in Brazil". However, subsequent news reported that the volunteer did not receive the vaccine, as he was in the control group. Weirdly, this was reported by Healthcare Finance in a news entitled "28-year-old volunteer in AstraZeneca COVID-19 vaccine trial dies". Only the subtitle clarifies the story, as it asserts "The trial has not been paused as the volunteer did not receive the COVID-19 vaccine but was part of the control group.". Still, more generally, even a death of a single vaccinated individual should not radically affect our estimation of the dangerousness of the vaccine, as the cause of the death may be unrelated to the individual's vaccination. In fact, if tens of thousands of individuals get vaccinated, then a death of at least one individual should be expected in the months to come since; according at Statista, adults have a probability of dying within a year of at least 1 in 1,000.

ChevrePensante-21FR reports the way vaccine news have been reported in a disproportionally negative manner.

More generally, to avoid being misleading, information should arguably present on scientific background knowledge, set information within context, analyze the best available statistical data, discuss conflicting interpretations and use careful wordings.