Making Dumb Phones Smart

Introduction

The proposed intervention leverages a text-based Large Language Model (LLM) accessible via SMS on basic mobile phones to deliver critical knowledge, expertise, and opportunities to people in poverty-affected areas. By transforming "dumb phones" into smart tools, this initiative aims to bridge information gaps and empower individuals across various domains such as education, agriculture, healthcare, and entrepreneurship.

Problem

In many poverty-affected regions, communities face a multitude of challenges:

Limited Access to Knowledge and Expertise

A lack of educational resources and expert advice hampers personal and professional development.

Exclusion from Smartphone-Based Interventions

Most modern interventions and communication platforms are designed for smartphones and require internet access. Solutions like WhatsApp and mobile apps are ineffective in these communities because they are inaccessible on basic phones.

Information and Opportunity Gap

The absence of accessible technology creates a significant information gap. Individuals cannot obtain essential knowledge on education, healthcare, agriculture, and economic opportunities, hindering personal and professional development.

Scarcity of Implementers and Data

There is a shortage of skilled individuals to implement interventions and a lack of data to inform decisions.

Objectives

Enhance Access to Information

Provide timely and relevant information across critical domains through SMS-based LLM.

Empower Decision-Making

Enable individuals to make informed decisions in areas like agriculture, health, and entrepreneurship.

Foster Personalized Support

Offer individualized advice and assistance tailored to the user's context and needs.

Collect Rich Data

Gather data on local issues and trends to inform future interventions and policies.

How It Works

Person

Basic Phone

SMS Gateway

Cloud API

LLM

Domains to Assist In

Teacher education (Banerjee et al., 2016)
Relationships
Sexual Health (L'Engle et al., 2016)
Per-Student Quizzes
Emergency Preparedness
Nutrition (Bassi et al., 2018)
Water Sanitation techniques, practices, risks
Female rights and awareness (Ippoliti & L'Engle, 2017)

Preparing for weather and pests (Cole & Fernando, 2021)
Solving issues (Aker & Mbiti, 2010)
Individualised Advice (Fafchamps & Minten, 2012)
Plant selection
Optimisation of agricultural variables
Context and budget considerations

Basic advice (Free et al., 2013)
Preventative healthcare (Aranda-Jan et al., 2014)
Child healthcare
Maternal care (Lee et al., 2018)

CBT techniques (Aguilera et al., 2017)
Counselling
Dispute Mediation
Mental wellness tips

Career Advice
Communication
Translation
Entrepreneurship (Donner & Escobari, 2010)
Government programs
MicroFinance Opportunities (Suri & Jack, 2016)
Job Opportunities
Training and Resources

Reach

Basic mobile phones remain the primary communication device for billions of people worldwide, presenting a massive opportunity for impact.
1.5 billion people are left behind by the LLM revolution.

Global Mobile Phone Usage (2022)

6.4B

Total
Mobile
Users

4.9B

Mobile Internet
Users

1.5B

Basic Phone
Users

Numbers in Billions. Source: GSMA Intelligence, 2022

Basic Phone Users vs Mobile Internet Users in a Region (2022)

Sub-Saharan Africa

49%

Asia Pacific*

24%

Latin America

21%

Greater China

19%

*Excludes Greater China. Source: GSMA Intelligence, 2022

Advantages

Access: Reaches 1.5 billion people currently excluded from smartphone-based LLM applications
No client upgrades needed: No client-side software updates are required for new interventions, changes, or new features.
Rich data collection: Every interaction provides valuable insights about community needs, challenges, and intervention effectiveness
Automatic trend tracking: System can identify emerging issues and patterns across communities through message analysis
Progress monitoring: Built-in feedback mechanisms enable continuous evaluation of intervention impact
Superior coverage: SMS networks typically have broader reach than internet infrastructure in rural areas
Viral adoption: Users naturally share the service with others, creating organic growth through word-of-mouth
Unified infrastructure: Single cloud platform can support multiple interventions and use cases efficiently
Efficient scaling: Serverless architecture automatically adjusts to user demand without manual intervention
Flexible trial support: Same system works for both small pilots and large-scale deployments
Personalization features: A profile system can enable contextually relevant responses based on user context, location, history and preferences
Platform expandability: System can be extended to support WhatsApp and other messaging platforms as needed
Low Technical Risk: Implementation is straightforward and can reliably be implemented

Issues & Challenges

SMS costs: SMS-based communication is significantly more expensive than internet data ($0.00355 vs $0.000001 per message). While client-side costs can be subsidized through toll-free numbers, the high server-side SMS costs remain a key challenge for scaling. This solution should not be used when smartphone and internet access is available.
Service awareness challenges: Difficulty in making target communities aware of the service and its benefits
Text literacy barriers: Limited ability of some users to read and write text messages effectively
Privacy concerns: User concerns about data security and personal information protection
Limited indigenous language support: Challenges in supporting multiple local languages and dialects
Capability limitations: Technical constraints of basic phones and SMS technology
Potential bias in responses: Risk of LLM generating culturally inappropriate or biased content
Trust building challenges: Difficulty in establishing credibility and trust with target communities
System accuracy concerns: The system could output factually incorrect information
LLM context limitations: Constraints in maintaining conversation context over multiple SMS messages
Global SMS routing complexity: Technical challenges in managing SMS delivery across different regions
Signal coverage limitations: Areas with poor or inconsistent mobile network coverage
UX constraints (160-char limit per text message): Restrictions on message length affects user experience. SMS limits to ~40 tokens per message for input, and separately for output. Need to support the user providing more information through multiple messages, ideally without the server responding by SMS before the user has finished submitting the full query.
System dependence risks: Communities becoming overly reliant on the service for critical decisions
Cultural adaptation needs: Requirements to tailor content for different cultural contexts
Language complexity handling: Managing nuanced communication across different languages
Bad actor prevention: The system needs to be protected from misuse and malicious activities
Non-poverty use management: Preventing system abuse by non-target users
Personal information security: Protecting sensitive user data and maintaining privacy
Community technology perception: Overcoming skepticism or resistance to technological solutions
Input medium limitations: Constraints of numeric keypad input on basic phones
Phone ownership dynamics: Issues related to shared phone usage and access
Local laws and customs compliance: Adhering to regional regulations and cultural norms
Harmful use prevention: Preventing system misuse that could harm users or communities
Academic integrity concerns: Preventing misuse for cheating or plagiarism
Spam management: Preventing and controlling unwanted or automated messages
Regulatory compliance: Meeting telecommunications and data protection requirements
Unforeseen impact management: Addressing unexpected consequences of system deployment

Theory of Change

Note: This theory of change is a general framework. Actual implementation will vary significantly based on local context, culture, stakeholders, resources, and specific needs of each community.

Inputs

Technological Infrastructure: Development of an SMS-based LLM platform accessible via basic mobile phones.
Content Development: Curated and localized content across various domains.
Partnerships: Collaboration with local governments, organizations, and frontline workers.
Funding and Resources: Financial investment for technology, marketing, and operations.

Activities

Development of the SMS LLM Platform:

Design, implement and deploy the software infrastructure.
Customize the LLM to function effectively over SMS with limited character support.
Ensure language support for local languages.

Content Localization and Specialization:

Adapt content to reflect local contexts, customs, and laws.
Engage with community leaders to ensure cultural relevance.

Awareness and Trust-Building Campaigns:

Implement SMS marketing campaigns showcasing examples of the service.
Conduct community meetings and workshops to build trust and demonstrate value.

Pilot Implementation:

Launch pilots in selected communities to test and refine the service.
Collect feedback and iterate on the platform and content.

Outcomes

Short-Term Outcomes:

Improved knowledge in critical areas like health, agriculture, and education
Enhanced problem-solving abilities at the individual level

Medium-Term Outcomes:

Behavioral changes leading to better health practices, agricultural productivity, and educational attainment
Increased utilization of government programs and microfinance opportunities

Long-Term Outcomes:

Reduction in poverty levels due to informed decision-making and opportunities seized
Empowered communities with improved socio-economic indicators

Key Performance Metrics

Behavioral Change Metrics:

Practice adoption rates in target areas
Sustained behavior changes over time
Community-wide impact measurements

Knowledge Assessment Metrics:

Pre and post intervention knowledge tests
Skill application success rates
Knowledge retention metrics over time

User Engagement Metrics:

Number of active users and growth rate
Message volume and interaction frequency
Response rates and average response times
User retention rates over time

Risks and Mitigation Strategies

Awareness and Adoption:

Risk: Low awareness may lead to limited adoption.
Mitigation: Implement extensive community engagement and education campaigns.

Language and Literacy Barriers:

Risk: Limited language support and literacy levels may hinder usage.
Mitigation: Incorporate local languages and simplify language complexity; use voice-based options where feasible.

Privacy and Security Concerns:

Risk: Users may be reluctant to share personal information.
Mitigation: Ensure data privacy through secure platforms and communicate these measures clearly to users.

Cost Constraints:

Risk: SMS costs may be prohibitive for users.
Mitigation: Subsidize SMS costs through partnerships with telecom providers or offer toll-free numbers.

Trust and Reliability:

Risk: Users may distrust the accuracy of the information provided.
Mitigation: Validate content through expert review and establish credibility through endorsements from trusted community figures.

Cultural Sensitivity:

Risk: Content may conflict with local customs or norms.
Mitigation: Engage local experts in content development and continuously adapt based on feedback.

Stakeholders

End Users (Community Members)

Individuals in poverty-affected areas using basic mobile phones to interact with the SMS-based LLM service

Local Community Leaders

Trusted figures within the community who aid in awareness, trust-building, and cultural adaptation of the service

Local Governments

Regional authorities providing support, approvals, and integration with local policies and programs

Telecommunications Providers

Companies offering SMS services and infrastructure essential for delivering the intervention

Content Developers

Experts responsible for creating and localizing content suitable for the target audience and context

Technical Team

Engineers and developers building and maintaining the SMS-based LLM platform

Funding Partners/Donors

Organizations or individuals providing financial resources to support development and scaling

Implementors

Field teams responsible for on-ground execution, training, and community engagement
Local coordinators managing day-to-day operations and stakeholder relationships
Monitoring and evaluation teams tracking impact and gathering feedback

Local Institutions

Schools and educational centers receiving support and resources
Local clinics and health centers accessing medical guidance
Farming cooperatives and agricultural groups getting expert advice

Research Partners

Academic institutions providing evidence-based insights and evaluation
Research organizations conducting impact studies and assessments

Local Businesses

Small enterprises that can benefit from and support the intervention
Local service providers integrating with the platform

Non-Governmental Organizations (NGOs)

Entities working in the area that can partner for outreach, implementation, and support

Cost Analysis

Basic Tech Monthly Cost Calculator (Rough Estimate in India)

Monthly Costs

Incoming SMS ($0.006 per SMS): $0.00

Outgoing SMS ($0.006 per SMS): $0.00

LLM Input ($1 per M tokens): $0.00

LLM Output ($5 per M tokens): $0.00

AWS Infrastructure (To be estimated, per million requests): $0.00

Total Monthly Cost: $0.00

Cost per Person: $0.00

Number of People: 1000

Incoming SMS per Person per Month: 10

Input Tokens per SMS (Prompt+User Query): 30

Output Tokens per Response: 40

Response SMS per Query: 3 SMS

LLM Model:

Cost per Incoming SMS: $0.006

Cost per Outgoing SMS: $0.006

SMS Provider: View ClickSend India Pricing

AWS Cost per Million Requests: $9.80

Additional Cost Considerations

Fixed Costs: Taxes, setup fees, toll-free services
Marketing: SMS marketing campaigns and community outreach
Development: Content creation and localization
Improved Model Capabilities:Retrieval Augmented Generation, Long Prompts, Database, etc
Personnel: Local coordinators and implementation partners
Quality Assurance: Monitoring and evaluation teams
Compliance: Local laws consulting and regulatory requirements

Implementation Strategy

Work in Progress

Deeply analyse and target deficiencies
Ideate solutions for context
Discuss with local community
Develop and test agents
Deploy and market solutions
Track metrics and iterate

Scaling Strategy

Work in Progress

Initial Pilot
Pilot RCT
Repeat solution process per local area

Key Questions

Intervention prioritization
Awareness strategies
Per-person-cost calculation per day
Possible partners - governments, organisations, frontline workers
Funding sources
Local laws compliance
LLM model selection and optimization