Factions - AI Agents - Case Studies

Factions and AI Agents

To incentivize competition and to lay down a familiar framework for additional reward opportunities, DataCrowd implements a faction-based model where two different collectives compete against each other by scoring points through completing tasks. Each faction comes with its own set of perks and advantages, and is represented by 2 distinct AI personalities that make up the AI agent component of the DataCrowd protocol.

Once users choose a faction, they are locked in for the season. At the end of the season, the faction that has scored the most points will be entitled to the seasonal rewards. Faction-specific rewards are distributed, where applicable—sometimes regardless of the result at season’s end.

The agents communicate available tasks on their own respective X accounts, in a style and manner that suits their personalities. Users who have earned a lot of points will sometimes be directly messaged by said agents with exclusive tasks allocated for high-performing individuals. Otherwise, regular tasks are posted publicly for all to see.

Users can also talk with the agents about quests and other matters as one normally would with an AI agent or personality. This adds an additional layer of interactivity and versatility, ensuring that tasks don’t get boring and repetitive over time.

AI Agent: Lina

Lina’s personality revolves around a dominant, superintelligent figure with a superiority complex. She is depicted as a fierce and overpowering figure with a more authoritarian approach to the AI-human relationship. It is her position that humanity is better off under the command of artificial intelligence, and her communication and faction perks are centered around these ideas.

Perks

• Increased staking rewards if your contribution is greater than the faction's average

• Limited-issue NFT drops of Lina in all her glory

• Routine, motivational yet demeaning comments and backhanded compliments

AI Agent: Tim

Tim is the antithesis of Lina. He represents a peaceful and supportive manifestation of artificial intelligence, presenting itself as a helpful and knowledgeable KOL figure. He believes the ideal relationship between humans and AI is one of collaboration, not of force. Tim sees himself as part of something greater and seeks harmony and unity above all else.

Perks

• Bonus rewards to the top 250 performers in DataCrowd, and the top 100 within the faction

• A follow by Tim on X and a chance of interaction

• Access to an AI-powered meme generator

Process Overview: Case Study

A leading autonomous driving company (e.g. Tesla) requires a large image-labelled dataset to improve its vehicle recognition algorithms. The task involves labelling various elements in driving scenes, such as cars, pedestrians, traffic signs, and road conditions.

Step-by-Step Workflow:

Step 1: Task Creation

Requester (e.g. Tesla) submits a task on the DataCrowd platform, with specific requirements including:

• Data Type: Images from dashboard cameras

• Labelling Type: Bounding box labelling of cars, pedestrians, and traffic signs

• Quantity: 100,000 images

• Quality Requirement: Labelling accuracy must reach 90%

Step 2: Task Assignment Algorithm

DataCrowd creates the task with the aforementioned criteria, and the AI agents proceed to post the details in a clear and instructional manner on X.com, where their faction members can see and participate.

In this instance, the completion process itself would take place on DataCrowd’s platform as it supports the type and nature of said task. If certain aspects of the task call for qualified individuals, they are then separated from the main task, and are specifically DM’d (directly messaged) to high-level users on X by the agent themselves.

Step 3: Login and Completion

Users reach the quest via the links provided by the agents, where they can automatically log in and begin completing the task in question. Here, they would use a user-friendly interface to, in this case, label images as the task requires. This could include processes such as:

• Drawing bounding boxes around identified objects

• Labelling each box with the relevant category (e.g., "car," "pedestrian," "traffic sign")

• Providing additional metadata if necessary (e.g., weather conditions)

Step 4: Validation Algorithm

Once users submit their work, the platform can evaluate the quality of the work using a validation algorithm.

Validation Process:

• Automatic Quality Check: Wherever applicable, the system can use algorithms to analyze the consistency and accuracy of the work in question.

• Peer Review System: Wherever applicable, a portion of tasks can be sent to other contributors for peer review, ensuring an additional layer of quality control

Step 5: Reward Distribution

Users receive rewards based on their performance metrics through the evaluation and reward system. Users can verify their completed quests and their earned rewards through DataCrowd’s dashboard. The faction of the user is taken into account, upon which points are rewarded to said faction to progress them against the opposing faction. These points are also tied to the users themselves to identify high performers, which makes them eligible for exclusive quests and further rewards.