Short Message Service (SMS) text messages, which started as a way for Nokia engineers to test mobile phone network operations, has grown into a killer app – for everyone. At the SMS4Dev Technology Salon in San Francisco, we looked at three ways to apply SMS to pressing development projects.
SMS:Gov
First we discussed the issue of local government communications with their constituencies. The problem being that usually they don’t communicate with constituencies outside of infrequent in-person meetings.
By employing software like FrontlineSMS in a SMS:Gov usage model, local governments could offer two compelling services: 311 and MyObama.
- 311: By offering a single, simple text message menu tree using keywords, local governments can categorize constituent needs and wants into categories for prompt response, like the 311 systems in New York, Washington DC, and San Francisco.
- MyObama: Local politicians can use the same process to become knowledgeable on the electorate’s concerns, and individualize their message to respond to those concerns, like MyObama did to great success in 2008
Rob Munro discussing his SMS efforts
Categorizing SMS via Artificial Intelligence
But what if the categories for keywords are not known in advance, or a community doesn’t understand the concept of a keyword? Robert Munro faced this challenge in Malawi when implementing FrontlineSMS with rural Community Health Workers (CHWs) who mainly use the Chichewa language.
The doctors at a central clinic spend one hour each day managing incoming CHW text messages, but with a patient population of 250,000 this averages to just 5 seconds per patient per year, and so any automation for triage incoming text messages from CHWs can lead to huge productivity increases.
Robert developed self-learning artificial intelligence algorithms that parsed free form SMS text messages in three different ways:
- Normalizing spelling variants of keywords by learning linguistically
predictable alternations - Segmenting words into their component morphemes to identify key substructures (like “patient” as the key form of “patients”)
- Using the normalized/segmented data to classify each message to determine its urgency – patient-related vs. administrative texts
With the algorithm learning from just 600 text messages it was was able to achieve about 95% accuracy, which should hold across any language using an alphabetic writing system and improve as the volume of text messages increases.
Applying SMS to Private Industry
Stepping away from SMS itself, Zach Berke spoke about two ways in which his company, Exygy, is developing text messages to support private industry expansion into the developing world.
- Payment plans: Solar power can be expensive, but how do you have a payment plan for an installed system? Require owners to text in codes they buy from local retailers to unlock another set amount of usage.
- Pharmaceutical validity: Counterfeit pills are a huge issue for consumers, but a simple code printed on a package can be texted to a central verification system to confirm drug authenticy.
Now both of these systems have their challenges. For the solar system, how do you pre-set codes into the hardware, or keep someone from soldering around the payment device. For pharmaceuticals, its printing variable yet secret codes on a specific end-user level package, with each code unique yet short enough to text without error.
Yet it can be done. Unicef used RapidSMS to track the distribution of 63 million mosquito bed nets across Nigeria with test messages on ordinary mobile phones using no-charge SMS shortcodes.