To see or not to see

I recently was introduced to a Canadian non-profit organization that mobilizes veterans to aid communities in disaster preparation, response, and recovery. In return, the organization provides a sense of purpose and community to the veterans who are transitioning back to civilian life. It's a fantastic win-win.

Naturally, I had to bring up the topic of AI. I wanted to educate myself on how the world of chatbots and generative videos can influence one of the least digitally reliant industries. Incredibly, AI automation dominated the majority of the discussion. I quickly found out that many humanitarian organizations have already identified opportunities to adopt AI and are actively seeking funding to explore these initiatives.

One strong practical use case of AI we identified is in the prevention of wildfires. Every year, the Canadian local government provides grants and financial assistance to property owners who wish to implement fire prevention practices - clearing overgrowth of vegetation, moving dry wood piles, etc. However, according to a study done by British Columbia FireSmart in 2023, over 50% of homeowners are unaware of the program and its benefits.

We considered leveraging computer vision and time-series satellite images in historically high-risk neighborhoods to assess government funding eligibility and notify the homeowners if they were eligible. In more traditional artificial intelligence practices, the Convolutional Neural Network (CNN) excels at identifying and learning spatial hierarchies in images. Each satellite image is passed through a CNN to extract spatial features related to vegetation density, distance to structures, and other relevant factors.

To make the model more sophisticated, we also thought of using Recurrent Neural Networks (RNN) that can easily handle sequential data, making them ideal for tasks involving time series or sequences. The idea is that by continuously analyzing satellite images, the CNN-RNN model can provide early warnings for regions at risk of wildfires. This approach allows for proactive measures such as controlled burns, evacuation planning, and resource allocation.

The technologist in me was extremely excited to apply AI to such a humanitarian cause. As I paced around the room, musing on a potential technical roadmap to a simple proof of concept, another thought occurred to me - what could be some other ramifications of this solution?

Back in 2015, Radiolab produced an episode of the podcast "Eye in the Sky". The gist of the story is about a powerful surveillance technology that allows its users to zoom in on daily life and rewind or fast-forward in time to solve crimes. This technology was originally developed to counter rising casualties from roadside bombs in Iraq. By deploying a small plane and a 44-megapixel camera to monitor an entire city continuously, this "eye in the sky" can trace back and identify bomb planters.

At the time, Radiolab raised ethical questions about the implications of such persistent surveillance. Should people know that they are being monitored? To what degree do we reasonably trade our privacy for public good? As I considered these questions, I couldn't help to recognize how eerily relevant they also are to analyzing temporal satellite images over vast geographical areas for wildfire prevention.

On May 20th of this year, Microsoft officially announced the Recall feature for Windows 11 during their special event introducing Copilot+ PCs. This feature takes periodic screenshots of user activity to help them "recall" past interactions with their computer through natural language queries.

While Recall has been praised for its potential to enhance productivity by allowing users to retrieve past activities quickly, it has also faced significant backlash due to privacy and security concerns. Critics worry that the constant recording of user activity could make sensitive information vulnerable to unauthorized access if a device is compromised.

Returning to the wildfire discussion, I suddenly wasn't too sure anymore. On one hand, using CNN and RNN to analyze sequential satellite images offers a cutting-edge approach to wildfire risk assessment. It also represents a significant step forward in leveraging AI for environmental protection and disaster management. On the other hand, technology solutions cannot be based only on the best-case scenarios. It is crucial to address the privacy and ethical implications associated with the surveillance capabilities.

Over the last 10 years, the general established trend for solution providers tends to lean towards action over inaction. It is easy to forget that we also have the responsibility to anticipate potential misuse and explore mitigating strategies long before the solutions are in flight. As AI rapidly evolves, creating a space for this discourse will be increasingly important to navigate the complex ethical landscape in modern innovations.