Automated Reasoning and ACDCs Secure Decentralized Reputation Algorithms Leveraging Property Graphs in ACDC to reason with uncertainty ... securely Samuel M. Smith Ph.D. sam@prosapien.com @ProSop|en
REPUTATION Al Symmetric Multiple Elastic Constraint networks with Reinforcement-Learning Works with sparse data But can be tuned and optimized as more data is collected
Autonomous Underwater Vehicles ’ - & . | e l'"r.r 7/'/‘“‘”» e ..’,;’ 4
Uncertainty Imprecision (Possibility Measures and operators) Randomness (Probability Measures and operators) Ambiguity (Similarity Measures and operators) Hybrid (mixtures of imprecision, randomness, and ambiguity)
PROCESSING ISSUES Measurement Scoring, Rating and Ranking Static vs Dynamic Feedback effects Data Rate Data Provenance
MEASUREMENT THEORY Scales
Nominal (Labeled but not ordered)
aggregation operations: none Ordinal (Labeled and Ordered)
‘aggregation operations: median, mode Interval (Ordered with Distance)
aggregation operations: median, mode, mean and moments about mean, std) Ratio (Distance with True Zero)
‘aggregation operations: all
noun:
The estimation in which a person or thing is held, especially by the community or the public generally. root:
Latin word reputdre, which is equivalent to re + putdre, that s, to re-think or re-consider. usage:
A considered evaluation (measure) of past behavior used to predict future behavior. qualification:
Confidence improves with contextual similarity.
S RS ——— sganer
WHAT IS REPUTATION? WHAT IS REPUTATION AI? Contextual predictor of future behavior to enable a transaction Closed-loop automated reasoning, not just open-loop pattern recognition Means to filter and modulate transactions Curator, recommender, decision aid, 1A Contextual predictors are more powerful Behavior based predictors are more credible Transitive predictors are more portable
Computational generation of a reputation is to aggregate relevant instances of behavior. Instances of behavior = reputational events or reputes for short Reputation measures are inferred indirectly from reputes associated with an entity Contrasting Example: * Entities provide direct ratings of promptness of another entity * Collect instances of behavior of entity in context from which promptness can be inferred. Reputation from reputed behavior allows for re-scoping, re-weighting, re-combination, and re-evaluation of collected reputes (not so for ratings) Enables arbitrary levels of nesting, precision, and granularity in the data aggregation process (not so for ratings)
S RS ——— -
Symmetric Multiple Elastic Constraints Type of MCDM (Multiple-Criteria Decision Making) Conventions/ Symmotic Grisp consraits (nenval) puscbectie uctons) Bothgoats and constraints o xpressed i it constas Maximize objective function(s) in area allowed by crisp constraints Dacision-making consists of finding the confluence of goals and constans by aggrsgating the respectve membershipunctos, l OfF)=F(xn,) )= S5 Sxn, 53 S0} Degre Fast max] ., O(%) Suisfaction o oo Fast — i e
State Space State of Nature . uE€U state of nature U state space - decision alternative @ €K decison spuce Criteria Spaces e 1X0 X X ) i set of criteria C defined on X = {X,...., X} criteria spaces 1Ny n Decision Space aintC, = constra criteria spa Criteria elastic constraintC; = j“constraint on i“criteria space Alternatives = Relations P=re) membership function 4, (3): X, = [0,1] 1' M Q=X M,:Q = X, relation between decision space and i* criteria space .III €, induces a constraint D; on € through M o - - .'!347 If M, =m@) st x=m(®) Then .1' o (0) = e (1) = fie (m (@) YoEQ .III Hy (@): Q= [0.0] # (4):X, = [0,1] ml e Final deisin by aggregating consrains Elastic Constraint on Decision Space D, Ho(@) = Agg(Mp, (@), 4p (@)..... Up, (D),....Hp (@)) Criteria Sp:u:ccj o (@:0 ()‘lj i i, (0): 2 [0, e (1) X, = 10.1] e
Elastic Constraint Network Building Blocks Layered pren — ction Importance - I I - Revter Inputs R — s B o e .“H“‘"‘“ Vae orcu] o - . e N A Ko L ] ] Contgence ovor oo ovr g | s - v - m occr e el ey et ==
Aggregation Operators . e A . smomms 7 deezee averaging ~N/ o \ averagin menbertip operators opersiore tnorms e o - e t-norms and s-norms, weighted t-norms and s-norms averaging operators weighted aver- aging operators ordered weighted averaging operators (OWA) hybrid operators.
Aggregation Operators . .Y Mo N\ 4o s HEmy e P R SR e ";” == SR\ el R ] Bt S e . S [— . N T R e T o N " _— s, e e N S | W e W= %&r J Rl
Symmetry Enables Networks of Elastic Constraints
- ]
s o} ot} -
] o)
o =T «.JEW”‘ ‘ (maT ) | omin
e | ~— {owa )
) e
o o -
g
SMEC Features Works well with sparse data. Bootstrap using default logic. Tune-up with more data. Refine with reinforcement learning Flat and wide constraint space most robust. Modular, composeable, contextualizable without re-training. Elastic constraint network is explicable. GDPR compliant. Can be modeled for prediction, recommendation, optimization, anti-gamification.
Decision Trajectory Based Optimization Pattern Centric Reasoning Predictive discrete event simulation with reinforcement leaming to encode high level reasoning, strategy and tactics. D Tt . [ —— k > a Active " B = g GB T p— el e z i \ © - - g ¥ Actve Trectry priend ‘ Souton S
PORTABLE IDENTITY Security, Privacy, Agency (Trustworthy, Private Preserving, Self-Sovereign) Portable Identifiers & Attributes Decentralized (not in a silo)
PORTABLE REPUTATION Security, Privacy, Agency Portable Data & Algorithms Reputation system that spans verticals and applications Benefits from data network effects
OPEN & PORTABLE & PROPRIETARY Open Data Formats Open Frameworks Open Algorithms Open Governance Proprietary Contextualization, Parametrization, Tuning & Data
META-PLATFORM Two Sided Network Effects Curation, Filtering, Modulation Optimal Control
SINGLE-SIDED NETWORK EFFECT + Network + More consumers increases value which attracts more consumers Demand side driven
R ‘/ \ / \ T \ / more consumers drive demand which attracts more producers more producers drive supply which attracts more consumers
EXAMPLE / Va ™ L\ f | | N
NEGATIVE CROSS-SIDE NETWORK EFFECTS + s e [ \ / O\ i U N . man NN J e = N, T . s More supply choice increases friction e.g. customer confusion in producer selection thereby decreasing demand More demand choice increases friction e.g. producer failure in customer satisfaction thereby decreasing supply