In court or in a police investigation, a digital photo is considered as solid evidence of an event having occurred, but a drawing by a person who has witnessed the event wouldn't be taken as seriously.

Yet from an information point of view, both processes are similar:

  • An image of the event was captured by the system (the camera lens, the witness's eyes).
  • This image was converted (camera: into 0 & 1s, witness: into neural signals) and than stored in the system (the camera's digital memory, witness's brain).
  • The image was later retrieved from storage and reproduced (camera: printed or displayed on a monitor, witness: doing the drawing themselves or using a forensic artist).

The processes are similar in principle, the only difference is the degree of accuracy, i.e. it is a quantitative difference. Yet they are considered to be entirely different qualities of evidence: The witness's reliability (even assuming good intentions) will be questioned and disputed by any prosecutor or lawyer. The digital photos will not be disputed in most cases, and would require an image processing specialist if they wanted to make the claim that it was digitally tampered with or photoshopped or something.

  1. Am I correct that this an epistemic question? Why is a digital photo considered more reliable than a human drawing?
  2. Does this have any implications for AI? We hear strong AI is impossible, computers can never do what a human does, blah, blah, but here we are trusting a very primitive computer (that's what a digital camera is) to be a much more reliable witness than a human.
  • “…it is a quantitative difference. Yet they are considered different qualities of evidence…” In most non-idealised settings, a large quantitative difference becomes effectively a qualitative difference. The mathematician Littlewood argues (flippantly, but I think sincerely) that the difference between qualitative and quantitative differences is itself a quantitative difference, not a qualitative one. I don’t know anywhere that such an argument is considered in more depth, but I’d be surprised if it hasn’t been. Nov 26, 2015 at 0:24
  • Reliable or valid? A reliable instrument is robust to random disturbances. A valid instrument has little systematic disturbance (bias).
    – ttnphns
    Mar 11, 2018 at 12:09

2 Answers 2


I suppose this could be filed as an epistemic question, but has other implications as well.

We certainly know that digital photos can be readily manipulated and that photos, such as Mathew Brady's "evidential" images, can be manipulatively staged. And we also use charts, graphs, police drawings, crime scene drawings, statistics, handwriting samples, recordings, models (as in Wittgenstein's example), and, above all, eyewitness testimony or "memory," which is generally considered the most reliable evidence.

The reason we may accept a digital photo as evidence is rather straightforward, but instead of confirming your second point, actually tends to refute it. The machine process is assumed to be a fundamentally physical-mechanical process with a linear causality and without the intervention of "subjectivity." Its product is objective or intersubjectively stable. That is, without the intermediation of a subject with freedom, values, choices, and all sort of possibilities for lying, deception, misrepresentation, moods, fears, greed, forgetfulness, or other "inaccurate" data transfers, including typically poor drawing skills.

The machine is simply "dumb" matter carrying out a predictable, "lawful" input-output transaction without the "ghostly," black-box operations of consciousness and memory. Machines don't lie. At least not without a human accomplice who "programs" or otherwise distorts their causal sequences. It is not the superior cognitive abilities, but precisely the inabilities of the machine that make it a "reliable" witness.

  • 2
    I know that digital photo evidence can be tampered with, but in that case the burden of proof typically lies with whoever made that claim, they will have to make a solid case (usually involving expert testimony and complicated proofs) that the photo was manipulated. Nobody requires such rigorous proof when a human witness is challenged, instead it is up to the witness to prove their own veracity. Nov 25, 2015 at 21:38
  • I get your point about intermediation of subjectivity, but that's really the gist of my question. Strong AI opponents are claiming that this is human's forte, but the photography example I gave hints at us subconsciously seeing our subjectivity as a weakness, not a strength. Nov 25, 2015 at 21:42
  • 1
    Whether subjectivity is a "strength" or "weakness" is entirely contextual. I am simply pointing out that it not commensurable with your machine model. And there are far too many dubious assumptions in your case. I do not think it is obvious at all that a photograph is always considered "more reliable" than a human witness. In any case, the photograph requires human attribution of "meaning" and is entirely ambiguous without it. The veracity of the evidence is ultimately referred to human judgment, and I don't think anyone would place confidence in a jury of 12 computers with video inputs. Nov 25, 2015 at 22:28

The question you raise has indeed an epistemic facet, and it could be break down to sub-questions: (1) What is the epistemic status of human-produced evidence (the drawing)? (2) What is the epistemic status of machine-produced evidence (the photo)? And then: (3) Of the two alternatives - why the first is taken to be in some circumstances less reliable than the second?

The premise accompany this question, if I understood you right, is that the two products are generated via similar processes yielding merely differences that amount to accuracy.

My conjecture is that the reason for taking the second as more reliable arouses from a desire to avoid error as much as possible - lack of accuracy could make us for example send some innocent person to jail. This desire can better be fulfilled when acknowledging that unlike humans, machines are not inherently biased, nor experience optical illusions, nor being distracted and so on. A machine-produced evidence thereby is preferable to a human-produced evidence when matters of trial and making-justice are at stake.

Contextual Note: from philosophical-historical point of view, during the 19th century intellectual world experienced epistemic-crisis in virtue of the discovery of non-Euclidean geometries which brought about the possibility that human's intuition cannot serve for matters scientific and cannot be trusted. It was thought before, that Euclidean geometry describes the space, and once having other consistent alternative geometries the question as to which geometry describes the space brought confusion. Furthermore, even before the discovery of non-Euclidean geometries, the gradual abandonment of religion and the secularization of societies, left humans with no epistemic-authority (God) and brought necessity to find trustworthy methods - mechanical methods or procedures evolved, and have been since preferable in light of the search for objectivity.

  • 2
    To conclude the line presented by Jordan S above, I would say that the epistemic principle is objectivity and the methodological principle is mechanism.
    – user18079
    Nov 25, 2015 at 19:51
  • I wouldn't say accuracy is the problem, because a memory could be very accurate on certain things a camera can't, or equally "bad" pictures (blurred, bad quality) can be not so accurate.
    – seldon
    Nov 25, 2015 at 23:10

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .