Towards useful AI in remote sensing

Remote sensing has never lacked algorithms. What it still lacks is work that travels. Artificial intelligence and machine learning now appear in roughly half of recent titles or abstracts in International Journal of Remote Sensing (IJRS). That shift is real, and it is welcome. However, frequency is a poor proxy for useful AI. Useful, here, means a result another laboratory can reproduce and that an operational user could begin to apply, not a new network scored on a private split and closed with an aspirational application sentence. A reading of IJRS titles and abstracts from 2020 to 2025, across more than a thousand papers that use learning methods, finds reuse language scarce and accuracy claims dominant. Keep building new methods. However, stacking methods on private datasets should not be treated as progress towards useful remote sensing AI.