The presumption of innocence, formalized in Article 11 of the 1948 Universal Declaration of Human Rights, serves two functions. Functionally, it allocates the burden of proof to the accuser. Symbolically, it expresses the moral priority of avoiding false convictions over punishing the guilty (Blackstone’s ratio). As legal scholar William Blackstone wrote, "It is better that ten guilty persons escape than that one innocent suffer."
This is a shift from adjudication to pre-crime analytics . As Crawford and Schultz (2019) argue, algorithmic systems "produce suspicion rather than respond to it." The user has no right to confront the algorithm, no discovery of the training data, and often no meaningful appeal. In Jasper v. Meta (N.D. Cal. 2024), the court held that Section 230 shielded Meta from liability, but noted that "the plaintiff was effectively tried and convicted by a statistical model." presumed innocent en ligne
Private online platforms (X, Meta, TikTok) moderate billions of content items daily. Their terms of service often include clauses allowing suspension or removal "at our sole discretion." In practice, automated systems flag content based on statistical risk scores. A user is not presumed innocent; rather, a post is presumed violative if it matches a pattern (e.g., certain keywords, account age, report frequency). The presumption of innocence, formalized in Article 11
A coherent response requires three levels of intervention. As legal scholar William Blackstone wrote, "It is
Moreover, forensic tools (e.g., cell-site simulators, hacking warrants) operate opaquely. The presumption of innocence requires that the accused can challenge the integrity of evidence. But when the evidence is an algorithm’s output or a proprietary tool’s analysis, meaningful challenge is often impossible. This creates a de facto reversal: the accused must prove the technology erred, rather than the state proving its reliability.