More thoughts on trust - some rough scribblings

These thoughts are the expanded results of some scribblings I made when I couldn't get to sleep around Christmas 2002, and serve more as notes to me than anything more concrete.

1) In a p2p network, there's a likelihood of repeated interactions (Prisonner's dilemma) - but not always? How to achieve dignified closure of a set of interactions when, for instance, at least one of the participants knows that closure is coming.
2) There is likely to be some knowledge of other players' past behaviour (though it's likely to be incomplete, even with reputation servers). This means that modelling the systm as a pure set of Prisonner's Dilemma continuing interactions is not applicable.
3) Reputation can help (see 2) but there is also a strong requirement for trust of reputation providers themselves, which doesn't often seem to be covered.
4) Do wider contexts help or not? Does the human interaction and social grouping imply increased social capital growth?
5) Resource gambling. Changing stakes from different interactions - discrete or continuous (does it matter?).
6) Authority figures - do they make a difference? How are they created (by some Hobbesian magic sovereign creation process? Are they trusted more than other members - and why? How is punishment exacted on participants?
7) Identity issues are, of course, very important (note 4 - more social interaction may imply greater confidence in identification).
8) We are usually expecting explicit buy-in (which is rather different to Hobbes).
9) Self-defense measures can participants take? Can be in place for the system?
10) To what extent is Bruce Tuckman's 1965 team theory (forming, storming, etc.) relevant (see Tuckman model)?
11) What type types of resource will be brought to bear? This will often include personal engagement.
12) Recalculating trust matrices (graph theory?). If A:C=x, A:B=y, B:C=z... If A trusts B with confidence y, & B trusts C with confidenc C, and (importantly), A:C is a function f(y, z), should A's confidence in C (x) be dented if B defects? Note: A:C=x does not imply C:A=x, so trust is not commutative. If C defects on B, then surely x _should_ be affected. if C defects on A, both y and z should be affected.
13) How useful is reporting the transactions in 12 to a reputation server? How trustworthy are those reports?
14) How are aggregated/corporate entities different? Do they have a place in such a system?
15) How do different levels of resources brought to the community change matters? Note that the importance of resources to parties makes a difference as to how much they care about losing them. Cameron and relative values.
16) How do other power sources change matters?
17) Applicability of 12 to DNS and routing? Look for research.
18) Research into SIFT and other real-time failure systems may also be relevant.


Mike Bursell mike@p2ptrust.org

2003-01-14