Making Dumb Phones Smart: Leveraging LLMs via SMS for Poverty-Affected Areas Idea

By Bernard van Tonder @ 2024-11-25T07:57 (+10)

Interactive Proposal: https://bernardvantonder.github.io/SMS-LLM/  (desktop view) 

Problem: 

• People in poverty-affected areas lack experts in useful domains e.g. medical care, education, entrepreneurship
• New LLM solutions target smartphone users as a usage medium
• 1.4 Billion people only have basic phones without smartphones

Candidate solution:

• Provide LLM over toll-free SMS (texting)

Issue:

• SMS cost is very high - $0.0035 per message in India https://www.textlocal.in/bulk-sms-pricing-india/

Scale: 1.4 Billion people

Impact: Depends on the intervention

Neglected: No one is doing this yet

Cost-effectiveness: Depends on intervention

Todo

• Complete implementation sections with experts.
• Make a much more detailed and complete theory of change, including adding requirements for each and justifying each causal connection along with a probability.
• Provide many more citations and justification for claims.
• Look at issues of similar interventions.
• Model micro-system dynamics, identify weak points, and visualize.

Issues with the proposal

• This is an early idea, not a detailed report.
• I lack experience and expertise in economics, intervention design, intervention implementation, behavioral science, operations, etc.
• I am biased towards tech-powered solutions.
• It lacks the rigor that interventions with months-long deep research have.
• Lack of specificity in intervention - consideration of a specific intervention requires deeper domain knowledge of systems and alternative interventions.
• It needs more evidence, research, deeper TOC with evidence and probabilities for outcome-transitions, planning, deeper cost-effectiveness analysis, benefit quantification per intervention, and alternate intervention comparison.
• I used o1 and Sonnet 3.5 to improve it.