Beyond Technology: The Human Responsibility in AI Powered Satellite Intelligence

Privacy

The capability to monitor locations in real time can raise profound concerns regarding privacy. Individuals and organisations may unknowingly become subjects of surveillance, leading to questions about consent and the right to privacy amidst growing technological advancements.

Regulation also plays a role: in the US, private satellites are currently allowed to create images of up to 25 centimetres of resolution, which is enough to discern something the size of a mailbox.

When satellite imagery facilitates intrusive monitoring of individuals' behaviours and locations, privacy concerns become unavoidable. A balance between security imperatives and personal freedoms needs to be found, where ethics plays a role together with comprehensive regulation.

Consent

In addition to privacy issues, questions of consent and data ownership further complicate the ethical landscape. Many individuals remain unaware that their movements and images are being monitored, raising concerns about the use of personal data for commercial purposes without adequate consent.

As technology evolves and the identification of individuals becomes possible, more debate is needed about what constitutes acceptable use and what crosses the line into intrusive monitoring from above.

The establishment of ethical guidelines and international standards is crucial to ensure that the deployment of satellite imagery and AI technologies aligns with societal values while adequately protecting individual rights and civil liberties.

Bias

AI systems used in conjunction with satellite imagery are not immune to bias. If algorithms are trained on skewed or limited datasets, they may perpetuate existing inequalities and result in discriminatory practices. This can lead to harmful outcomes, particularly for marginalised communities that may already face disproportionate levels of surveillance.

Two forms of bias appear frequently in geospatial AI:

• Geographic bias, linked to spatial autocorrelation: a model trained on imagery from a specific region may overfit to the characteristics of those countries and perform poorly in unseen regions.
• Seasonal bias: satellite imagery captures snapshots in time, and seasonal variations can significantly affect the appearance of fields and vegetation, causing models to develop a bias towards specific seasons or phenological stages.

Mitigation

Bias cannot be meaningfully addressed without first considering the intended outcome and potential use of the model. Ethical treatment of satellite imagery requires anticipating how the derived insights might be applied in real world contexts and by whom. Even technically sound models can perpetuate or amplify social inequities if deployed without careful consideration of their downstream effects and accessibility across different communities and power structures.

Beyond assembling more geographically diverse training datasets, two practical approaches can help:

• Augmenting existing data to increase variability
• Using GANs to generate synthetic data that better reflects diversity across regions, environments, and socioeconomic contexts

Ethical discussions highlight the need for transparency in AI algorithms and the importance of diverse training data to ensure equitable outcomes in spatial data treatment and surveillance practices.

Legal frameworks such as the General Data Protection Regulation (GDPR) and the EU AI Act aim to address some of these concerns by establishing guidelines for data usage, consent, and transparency. However, the rapid evolution of surveillance technology frequently outpaces existing laws, creating a protection gap for individuals in markets where AI integration remains largely unregulated.