130 likes | 235 Vues
www.altruists.org. Introduction to Imaginary Friends. Altruistic E conomics. A E 1 8 : The Imaginary Friends Test. v1.0. A presentation by Dr. Robin Upton (2006-01-28) Available for download at www.altruists.org/ae14
E N D
www.altruists.org Introduction to Imaginary Friends Altruistic Economics AE18: The Imaginary Friends Test v1.0 A presentation by Dr. Robin Upton (2006-01-28) Available for download atwww.altruists.org/ae14 Attribution – NonCommercial - ShareAlike Recommended Pre-requisite: AE10: Keeping Score
Real Individual User Artificial Construct . F2F Network Growth Introduction to Imaginary Friends • Altruistic Economicsis a decentralised system, so users have a lot of autonomy... . , such as the chance to create Imaginary Friends: • Rewarding such Imaginary Friendswould encourage deception.
Sybil Attacks F2F Network Growth TheFriend2Friend network grows organically, over real social relationships. Each new member creates an online server and is linked in by friends. The idea is, everyone has one online server. However,like WWW, there is no centre, noone in charge, no list of members.
The ‘Imaginary Friends’ Test Sybil Attacks ‘Sybil’ was the first person diagnosed with so-called ‘Multiple Personality Disorder’. A Sybil Attack is an attempt todefraud an online system by using fake identities.
FAIL: Incentive to Cheat 3. Imaginary Friends are an advantage: PASS: No Incentive to Cheat 1. Imaginary Friends make no difference: 2. Imaginary Friends are a disadvantage: . Media Sharing The ‘Imaginary Friends’ Test “Does it help to create ‘imaginary friends’?”
Aysha I value your opinion… Birbal I value your opinion… “Chinese Whispers” I know this film Chris Basic Framework Media Sharing MEDIA SHARING is a project to network people who review files and then benefit from one another’s opinions. Aysha wants an opinion about the film ‘Chinese Whispers’… How to quantify this process... ?
Aysha I value your opinion… 0.3 I value your opinion… 0.4 I rate this file… 0.8 Birbal “Chinese Whispers” Chris SoAysha has one recommendation: Value=0.12 Rating=0.8 Combining Recommendations Basic Framework Assume: • Individuals make 2 sorts of evaluations: 1) ‘I value your opinion’ is quantified as V: 0<=V<=1 2) ‘I rate this File’is quantified as R: 0<=R<=1 • Individuals who don’t rate a file defer to their friends • Friends of friends’ opinions are valued multiplicatively
So, if Aysha has 4 recommendations: Value=0.15 Rating=0.6 Value=0.8 Rating=0.5 Value=0.08 Rating=0.6 Value=0.12 Rating=0.8 0.15×0.6+ 0.8×0.5+ 0.08×0.6+ 0.12×0.8 0.15+0.8+0.08+0.12 Note the high value of the 2nd rating, which means it dominates the others. Applying The Test Combining Recommendations Ratings are averaged according to value, to give a single figure. Weighted average = ≈ 0.55
… Value=VB Rating = R Value=VB Rating=R Value=VB Rating=R I value your opinion… 1.0 I value your opinion… 1.0 I value your opinion… 1.0 Rating = R MBogo1 MBogo2 MBogoN … Rating = R “Chinese Whispers” { N separate ratings have Overall value = NVB Chris A Problem Applying The Test What if Chris made Imaginary Friends to rate the file for him?
Value=0.15 Rating=0.6 Value=NVB Rating=R Value=0.8 Rating=0.5 Value=0.08 Rating=0.6 If Nis big enough, no one else’s ratings count! Imaginary friends can increase an individuals’ impact, so the algorithm fails the imaginary friends test. A Modification A Problem So, if Aysha has 4 recommendations: Chris could use his Imaginary Friends to dominate the other ratings.
Value=VB V1 Rating = R v2 … vN Rating = R MBogoN MBogo2 MBogo1 … Nfriends give Chris N ratings, with a value of ∑Vf. Rating = R “Chinese Whispers” f Let’s require ∑Vf=1 Chris f Summary of The ‘Imaginary Friends’ Test A Modification To control the imaginary friends, we add a constraintto the model.
Chris …} N such friends Vf MBogof Rating = R Rating =Rf VB VB “Chinese Whispers” Overall rating: Value= VB∑VfRating=∑VfRf “Chinese Whispers” Since ∑Vf=1,1≥R,Rf≥0… f f f Chris So the modified algorithm passes the imaginary friends test. Modified Algorithm Examined Modified Algorithm Examined With N>0 imaginary friends: As an individual: Single rating: Value=VBRating=R imaginary friends confer no advantage.
Creating Imaginary Friends should not confer any advantage: Recommended Further Reading: AE19: Preventing Abuse http://www.altruists.org/ae19 Summary of The Imaginary Friends Test • Sybil Attacksare a feature of decentralised systems. • Applying the Imaginary Friends Test helps deduce algorithms suitable for decentralised systems.