"The Irrefutable History of You", a presentation at the Re-coding Black Mirror workshop (https://kmitd.github.io/recoding-black-mirror/) at ISWC 2017 (http://iswc2017.semanticweb.org)
4. What if things were like this?
• Clearly a satire on obsession with Likes,
Retweets, etc., on social media
• But…
– Could it happen (technically)?
– How could we mitigate the consequences?
• We won’t look at
– Could it happen (socially)?
– Should it happen? (No)
5. Can Nosedive be
implemented?
• Identification of people
– Connection to their social media accounts
– Reliable
– Automatic
• A system for recording and sharing ratings
– Trustworthy
– Tamper-proof
• All of this technology exists already
6. How can we do it?
• Phones with accurate GPS
– Broadcast user position/social media details
• Biometric sensors
– Avoids the “stolen phone” problem
• Distributed ledgers
– Secure tamper-proof record of transactions
7. What is a distributed ledger?
• Blockchain (underpins Bitcoin)
• Shared append-only data structure
• Writing records has a cost
• Being picked to extend the chain
– Costs money but earns a reward
– “Miners” compete
• Everyone has a copy of everything
– Tamper attempts detectable
8. Properties of distributed
ledgers
• Records are (effectively)
– Immutable
– Reliably timestamped
– Reliably attributable
• Support multiple data-writers without trust
– Initial motivation e-currency, to solve “double
spend” problem
• DL backend fits requirements for Nosedive
9. How can we do it? (Reminder)
• Phones with accurate GPS
– Broadcast user position/social media details
• Biometric sensors
– Avoids the “lost/stolen phone” problem
• Distributed ledgers
– Secure tamper-proof record of transactions
10. Mitigating the consequences
• Potential for abuse
– (Low) ratings which are:
• Malicious
• Trivial
• Thoughtless
– Over-interpreting ratings out of context
• How can we limit these?
11. Moderation approaches
• Variations on limiting ability to give/view
ratings
• Consider
– Choices of the rated person
– Reputation of the rater
– Opinions of others
– Costs of ratings
12. Choices of the rated person
• Choose whether to accept a rating?
– Certainly avoids abuse by low ratings
– Unlikely to be accepted as an approach
• Rating inflation
• Require mutual ratings?
– Pressure to limit low rating abuse
– Also unlikely to be accepted
• Selective visibility of ratings
– Similar problems as choosing to accept/reject
13. Reputation of the rater
• Weight ratings by
– Average rating of rater
• Influenced by prejudice
• Existing power structures replicated
– Illustrated in episode
– Rating history of rater
• Subject to gaming by malicious raters
• Doesn’t prevent multiple low ratings by crowds
14. Opinions of others
• Meta-rating
– Randomly select nearby people to “rate the
rating”
– Effective in online forum settings (e.g.,
Slashdot)
• Effort from others
• Likely to follow existing prejudices/power
structures
15. Cost of ratings
• Rater has to pay for each rating
– Limits frivolous ratings
• Guaranteed to preserve existing inequalities
– Rich have more power and will maintain it
16. Semantic mitigation
approaches
• Use semantic technologies to provide richer
interpretation
– Represent context of ratings
– Ontology and reasoning
– Categorical ratings
– Two-way personalisation of ratings
17. Rating context
• Semantic annotation of rating
– e.g., retail transaction
• As purchaser
• As retailer
– Social interaction
• e.g., as wedding guest
• Fine-grained interpretation
– Aggregate rating only across similar contexts
– Link to evidence for rating
18. Ontology and reasoning
• Ontology of rateable interactions
– Manually seeded, crowd-sourced?
• Reasoning on ratings using ontology
– e.g., “aggregate rating of A’s interaction in a
financial context with people who rate A highly
financially and low socially”
• Possible enabling of more abuse
– Derived categories could be specific and
discriminatory
19. Categorical ratings
• Semantic categories instead of numbers
– “A was helpful in this retail interaction as retailer”
• Avoids over-interpretation of numbers
– Non-numerical aggregation – “mostly harmless”
• More flexible
– Harder to model
• Rating categories curated?
– Could otherwise be abusive/insulting
20. Two-way personalisation
• Expose ratings only for particular semantic
contexts
• Require viewers to provide semantic context
of interpretation to be able to see
– How to prove this?
• Automatic negotiation of visibility
• Unlikely to be adopted in hypothetical society
21. Conclusions
• This is an awful idea, but
– Hypothetically, we can mitigate it with
• Moderation (mainly mitigates giving ratings)
• Semantics (mainly mitigates interpretation)
– All mitigation strategies either have flaws or
go against “purpose” of rating system
• It’s still a really terrible idea
Ubiquitous rating system 1-5. Examples of ratings. Effect on main character.
Caricature. For the sake of argument, assume…
For the uninitated
In the hypothetical social context
Notice that in the episode, main characters shown to suffer from rating system are a woman and a black man; Lacie’s brother (white male, hinted to be heterosexual) makes a point of being unconcerned about his rating
Low crowd ratings: as in the episode