Fénix-Starling AI Evidence Pillars

The AI Evidence Pillars were developed by Fénix Foundation in collaboration with Stanford Starling Lab

Corroborate evidence through diverse and varied sources, while remaining aware of ‘Impostor Bias’ and the ‘Liar’s Dividend’ and the potential for unduly high corroboration requirements.
Be aware of the ‘blackbox’ nature of AI, while refraining from disaraging AI-affected evidence solely on this account. It may be necessary to adopt an ‘explainable-enough’ standard.
As technical solutions continue to develop in the AI arm’s race between sophistication and detection, rely on corroboration to determine the probative value of digital evidence lacking in provenance information.
To avoid potential enhanced prejudicial effect of AI-affected visual and audio evidence, proceed with awareness — aim to interrogate technical and metadata markers before viewing or hearing the material itself.

Be aware of one’s ability to assess whether they are qualified to understand issues related to AI-affected evidence in a particular case.
Be aware of a threshold of inexplicability when scrutinising AI-affected evidence and explore ways to cure the ‘black box’ effect.
To ameliorate the resource strain of handling AI-affected evidence, endeavour to collaborate with industry expertise—with due consideration for collaborators’ privacy standards.
When adopting third-party tools and establishing partnerships, consider potential impacts on independence.