Truth, Trust, and Deepfakes

Here is my perspective after several discussions and some side research:

  • The general case is nearly impossible, but specific use cases can be solved provably or to a certain degree probability.
  • Assigning identity, and gathering witness signatures end up being statistical. So I’d say those are the two most challenging items.
  • Identity for people vs. physical things vs. digital things have different trade offs. Speed, scale, transparency, scope of actions, etc. are different.
  • Keys and signatures usually require at least one point where trust is necessary (preferable in hardware) but cannot mathematically be 100% in the general case.
  • One can look at it as
    • Authority
    • Authenticity
    • Provenance
    • Validity
  • The 1st three lead to
    • Signatures, which can be cracked or lost or stolen
    • Modifications of content, which can happen subsequent to creation
    • One way to solve the above is via
      • Irreversible integrity, using Time Stamps
      • Signatures are irreversible, and
      • Content is irreversibly linked to source and signature
      • Blockchain is suitable for this kind of use case
  • The 4th one, validity, is a harder problem since verification can be subjective.
  • One way to solve that is to require additional signatures from witnesses, again time-stamped and within a time-window at the point of inception.
  • This requires incentives to solicit and receive additional signatures plus transparency about where those signatures are coming from.
  • The statistical nature of this makes it a sliding scale, even under perfect circumstances where witnesses have no conflict of interest/aren’t bought off, etc.

Needless to say, battling digital mistrust and deepfakes is already a very significant problem. As I like to say, trust is expensive, cherish it if you have it!