Part 2 - What can AI see?
One of the major challenges facing the global community as it implements AI solutions to address critical challenges will be developing solutions that don’t create a digital divide between those who already have access to technology, and those who don’t. While AI technologies have immense potential to act as a social leveller by bringing down skill barriers, the technology could exacerbate existing inequalities and fail to improve the lives of the world’s poorest.
Electricity
Access to electricity is a significant challenge in the rural communities of many of the world’s poorest countries. In 2020, approximately 733 million people worldwide lacked access to electricity, with around 80% of them residing in rural areas (World Bank, 2023). For AI systems to work, you need a steady supply of electricity to power servers and data-gathering devices.
One of the challenges for LUMs was that whilsts the power grid could provide a consistent supply of electricity to their central servers, it couldn’t for the towers distributed throughout the forest. To avert the risk of the hardware on the towers losing power supply at crucial moments, LUMs combined, solar, sensors, and cloud-based solutions to more consistently maintain power throughout the system.
All the equipment on the tower is operated using solar-charged batteries. The tower has a sensor component, containing a camera and weather station, and it has an embedded computer that takes the data from the sensor and transmits it over the network to the cloud storage. Once the data is transmitted to the cloud it is instantaneously pulled by the Central Server where the AI module detects a smoke or fire event. The cloud, central server, their inter-communication, and the AI module are connected to grid-based power.
Cost
Cost is another significant barrier to ensuring AI technologies benefit everyone. Later in the report, we delve into the costs associated with developing computer vision solutions and what that might mean for international development programs.
For the LUMs team, the high cost of thermal cameras was a major challenge. Thermal cameras, which detect infrared energy instead of visible light, are particularly useful for identifying fires in environments where visual cues might resemble fires but lack the same heat signature. However, the expense of this hardware would have consumed a large portion of their budget, and as we’ll explore later, cameras are only one component of the overall cost of computer vision systems.
To address this, the team opted for more affordable PTZ (Pan-Tilt-Zoom) cameras, which capture images using visible light. While this approach had limitations, they mitigated these by integrating IoT sensors on the ground to collect additional data. This combination of lower-cost technologies enabled them to develop a functional and cost-effective system.
Further savings were achieved by introducing a 30-minute interval between image captures, reducing the burden of data transmission. This interval could be adjusted as needed, providing flexibility while keeping costs down. These measures collectively reduced the projected cost by tenfold.
This approach highlights a valuable lesson for managing costs in AI projects: state-of-the-art hardware is not always necessary to achieve excellent results. Conducting a detailed analysis of the technical requirements for specific use cases can help identify the most suitable and cost-effective hardware. Additionally, experimenting with combinations of lower-cost technologies can deliver effective results without exceeding budget constraints.
In the next section, you’ll learn more about some of the ethical challenges and solutions the team faced.