Karamad: A Voice-based Crowd-sourcing Platform for Under-served Populations

Document Title

Karamad: A Voice-based Crowdsourcing Platform for Underserved Populations

Document Information

• Subject Area: Human-Computer Interaction, Voice Technology, Technology Development and Social Impact
• Keywords: Human-Computer Interaction, Voice Crowdsourcing, Crowdsourcing, Low-resource Populations, Low Literacy, Mobile Phones, Supplemental Income

Research Background and Problem

• Identified Issues or Challenges: Traditional crowdsourcing platforms (e.g., Amazon Mechanical Turk) require the use of computers or smartphones and rely on internet access, yet approximately 46.4% of the global population lacks access to these devices. Additionally, around 750 million people are illiterate, and 220 million have visual impairments, making it difficult for these groups to participate in existing crowdsourcing work.

• Significance: These low-resource, low-literacy populations hold significant potential and can greatly benefit from the income generated through crowdsourcing work. However, existing large-scale crowdsourcing platforms have yet to effectively reach these groups.

• Research Motivation and Related Work: Voice platforms are considered a potential solution to address inclusivity and usability barriers. For instance, the BSpeak platform focuses on assessing task feasibility for low-income and visually impaired populations in India, while the ReCall platform evaluates the accuracy of voice transcription tasks. Against this backdrop, the Karamad platform aims to explore connecting low-resource populations using voice technology and interactive voice response (IVR) systems.

Solution

• Proposed Method or Solution: The authors designed and implemented a voice-based crowdsourcing platform named Karamad, enabling low-resource populations to complete tasks using basic mobile phones without requiring internet access. Upon task completion, participants are compensated through mobile airtime credits.

• Innovations:

  • No need for smartphones or internet access; participation is possible using basic phones.
  • An IVR system provides an easy-to-use work interface.
  • Users are attracted through word-of-mouth without additional advertising or training.

• Implementation Steps and Key Technologies:

  1. Platform Architecture: The platform consists of a web portal, an IVR platform, and a payment processing module.
  2. Task Types: Supports various tasks, including surveys, voice data collection, and voice translation.
  3. Payment Mechanism: Reduces mobile airtime recharge costs and transfers earnings through reliable payment processors (e.g., Meezan Bank).
  4. Dynamic Task Pathways: Offers dynamically adjusted question pathways based on users’ previous responses.
  5. User Interaction: Provides a user-friendly IVR interface while optimizing the "browse-select-complete" task flow.

Research Outcomes

• Specific Results:

  • Within six months of deployment, the platform attracted 725 users, completed 3,939 tasks, and collected over 43,006 data entries.
  • Users included women, unemployed individuals, low-literacy populations, and visually impaired individuals, with approximately 85% being visually impaired.
  • The average income per user was approximately PKR 637. Although relatively low, this income had significant economic value for the users.

• Advantages:

  • The platform achieved self-sustaining user engagement over time without advertising, training, or additional airtime subsidies.
  • Users demonstrated high participation rates in sensitive topics, such as HIV/AIDS surveys.

• Experimental or Evaluation Results:

  • Behavioral measurements showed that 35% of users engaged with the platform over multiple days, with retention rates comparable to social platforms.
  • Validation surveys revealed a high consistency of 74.4% in binary question responses, while multiple-choice questions (MCQs) showed lower consistency (below 39.2%) due to user input errors and cognitive load.

• Limitations and Future Directions:

  • Limitations:
    • The current payment method relies on prepaid airtime credits, with some users suggesting a shift to mobile wallets for greater flexibility.
    • The lack of advertising and notification mechanisms led some users to report difficulty in learning about new tasks.
    • User identity verification is lacking, making it easier for multiple registration numbers to be exploited.
  • Future Directions:
    • Increase task diversity: For example, adding general knowledge or religious-related questions.
    • Introduce notification mechanisms: Notify users of task updates via SMS or phone calls after registration.
    • Improve identity verification mechanisms: For instance, by linking user accounts to national ID numbers for unique identification.

Key Contributions and Conclusion

This study demonstrates the practicality and sustainability of voice-based crowdsourcing for low-resource populations, offering a crowdsourcing solution model that does not require internet or smartphone support. It is particularly suitable for regions with limited hardware resources but large populations. Karamad bridges human-computer interaction and social justice, providing profound insights and inspiration for the development of more equitable and diverse technological interventions in the future.