Proof of Engagement: A People-centric Protocol for Human-AI Interaction
The PIN team
April 5, 2024
Abstract
This paper introduces Proof of Engagement (PoE), a blockchain pro- tocol designed to measure and reward user engagement with personal AI. Unlike traditional mechanisms that rely heavily on computational power, PoE evaluates the quality, duration, and depth of human-AI interactions. This protocol integrates human data and activity evaluations with a dy- namic scoring system, verified on-chain, which can be used as a sybil resis- tance and incentive distribution mechanism to bootstrap a fair economy around Human-AI interaction.
1 Introduction
Unlike traditional consensus mechanisms focused on transaction validation or computational power, PoE stands for a new protocol aimed at measuring the degree, length, and depth of human interaction with personal AI. Its purpose is to motivate individuals to more actively engage with personal AI. This en- gagement (1) transforms how individuals interact with the digital realm and (2) enhances the intelligence of personal AI, enabling it to understand users more effectively. People maintain full ownership of their data.
2 Personal Intelligence Network
Motivation on how personal intelligence needs a new model of data ownership and sustainable incentivization/engagement.
3 Proof of Engagement (PoE)
We start by describing the concept of PoE via its core components.
3.1 Engagement Metrics
A set of metrics to determine how users interact with the PIN AI platform in ways that can be quantified and rewarded. Currently including two principal factors: Human Bits and Humanhood Probability.
3.1.1 Human Bits
This metric evaluates human attention by counting the number of language tokens produced by an individual in their interaction with personal AI.
3.1.2 Humanhood Probability
This involves a series of guidelines and computational methods that utilize users’
data to assess the likelihood that the data owner is an actual human.
3.2 Data Evaluation
Process engagement data, verify its authenticity, and calculate data engagement scores.
3.3 Activity Evaluation
Evaluate user activity engagement with Personal AI.
3.4 PoE Score
Combines Data Score and Activity Score. PoE score is proportional to the amount of newly minted rewards allocated to the individual.
4 Engagement Metrics
4.1 Human Bits
Human Bits can be measured by the number of language tokens from human, it does not include the internal chain of thought language tokens for LLM rea- soning, but only the human input query language tokens length, and the final human facing output language tokens length (including text output and function call output to control other services).
4.2 Human Probability Evaluation
Prior to the project, for each account:
Genesis score: Considering the timestamp of the genesis block as a benchmark, we treat all prior user data as authentic and trustworthy. We establish an initial score for each user based on the data available up to that specific timestamp, which we refer to as the genesis score.
Genesis data analysis: We regard all data generated before the gen- esis block as genuine and reliable. This data should reflect typical user behavior, including aspects like frequency, volume, or recurring patterns. This analysis serves as a foundational guideline to help us determine whether newly generated user data is authentic or fraud- ulent.
Screening for Existing Bots on the Internet Prior and Post Genesis:
Temporal Patterns: Evaluate how user behavior changes over time, not just at the genesis block. This can help identify whether certain actions are consistent with the user’s historical patterns.
Anomaly Detection: Implement algorithms to detect anomalies in user behavior that deviate from established patterns, which could indicate fraudulent activity or data corruption.
Cross-validation across different data sources: Data from a single user should display commonalities or overlapping information across various sources. This consistency can be utilized to confirm and maintain a unified user identity.
Social Graph Utilization: We leverage the Social Graph to enhance the Humanhood Probability of all linked accounts, similar to so- cial networking algorithms like those used by Facebook, LinkedIn, or Google’s page rank, which evaluate user connections and follow- ers.
Outside Humanhood Authentication: If users provide their outside humanhood authentication, they can bump up the Humanhood Prob- ability of all the other accounts. For example, the user can link:
KYC-ed account: API key to an exchange account or other cen- tralized KYC account like a Coinbase account, or Cell-phone service provider credential;
Biometric verification: Cellphone FaceID or fingerprint or World- coin ID type.
5 Engagement Evaluation
Detailed explanation of how engagement data is processed, the authenticity verified, and engagement scores calculated.
5.1 Data Score
The data score is determined by the stake-weighted sum of humanhood prob- abilities for each data API and associated account. For example, if two email accounts each have a humanhood probability of 1, the resulting data score is 2.
The score for each API, such as email or Facebook, is computed by aggregating the humanhood probabilities across accounts:
This score is proportional to the PIN staked for acquiring more data for each API. Initially, an equal weight is given to a selected list of APIs from major Web2 legacy companies, with plans to adjust these weights in the future based on strategic requirements.
5.2 Activity Score
PoE-Scoreactivity = #Human Bits from user interacting with on-device personal AI (3)
The activity is measured by counting human bits either from a human’s query or instruction to personal AI, or from the personal AI making function call instruction to control the mobile APPs and AI native application/services.
6 Construct PoE Aggregate Score
The aggregate PoE score is constructed by aggregating the data score from each data source and combining it with the activity score from interaction with personal AI.
where the sum of the stake weights equals 1.
6.1 Dynamic Weight Allocation in the PoE Score
The aggregate PoE score strategically allocates weights between the data score and activity score, emphasizing stake-weighted contributions for each data con- nector API. The approach to stake weighting is designed to evolve through two distinct phases:
Bootstrap Phase: Initially employs fixed weights to establish the system and facilitate user onboarding.
Mature Phase: Implements variable stake weights that adjust based on user engagement and system growth, reflecting a more dynamic and re- sponsive scoring mechanism as the network matures.
7 Protocol Incentives
7.1 PIN Monetary Policy
At each epoch the protocol issues m PIN tokens. The amount can be fixed according to an issuance schedule or it can vary sublinearly with the aggregate data/activity scores.
7.2 PIN Rewards Distribution
The goal is to direct issuance rewards to the most valuable actions for the network. We want to reward volume of data but also diversity. The most straightforward mechanism is to distribute rewards proportional to the PoE score. Incentives can start even before the network token is launched in the form of points. These also offer more flexibility with data onboarding campaigns.
7.3 Protocol Economics
In the initial bootstrap phase, the focus is on increasing data volume and distribution. The protocol earns no revenue, users only pay for the minimal blockchain resources (Data Availability and Secure Compute) used to calculate and store their PoE score and to distribute associated incentives.
In the mature phase, the protocol can start earning revenue via direct fees or fee burn. Moreover, there is a wide array of AI services and data markets that can spin up once the PIN economy is at scale.
Last updated
