← Let's ReThink History
Salmon Counting: How AI, Drones & Sonar Are Learning to See the Salmon Run cover
LET'S RETHINK HISTORY

Salmon Counting: How AI, Drones & Sonar Are Learning to See the Salmon Run

Since 1955, one of the most consequential numbers in conservation — how many salmon 'escape' upstream to spawn — has been produced by a person on a wooden tower with a hand-clicker. This 8-part ReThink History series follows the technology now learning to watch the river for us: solar-powered 'drone-in-a-box' units over Bristol Bay, computer-vision models trained to pick a single sockeye out of glare and shadow, imaging sonar that sees fish in pitch-black glacial water, the 'echogram' that lets a cheap computer count fish where there's no internet, and Indigenous-led AI built into traditional First Nations weirs. Each new way of seeing does more than count faster — it reveals more of a life spent hidden underwater and thousands of miles out to sea, and it quietly changes who controls the count. The series runs from the counting towers and the salmon's own secret life cycle to the hard questions underneath — a warming climate, the Pebble Mine at the headwaters of Bristol Bay, and who owns the count — bookended by a summer on the cannery line in remote Naknek, where the run peaks on the Fourth of July and the bears watch from the fence. Narrated by AI voices from sourced, human-reviewed research. © 2026 Selway Solutions LLC · <a href="https://history.selwaysolutions.io/index.html#license">Educational & Institutional Use License</a> · <a href="mailto:nick@selwaysolutions.io">Contact</a>

Listen on Apple Podcasts, Spotify, YouTube, or paste the RSS link into any podcast app.

Episodes

  1. Climate, Pebble Mine & Who Owns the Salmon Count

    What does it all add up to? First the honest test now underway — run the AI counts against decades of human tower data and see if the machine can be trusted.

  2. SalmonVision: Indigenous-Led AI on Ancient Weirs

    The same AI revolution — but Indigenous-led, and pointed at sovereignty as much as science.

  3. The Echogram: Counting Salmon With a Single Image

    Here's the elegant trick that lets a cheap computer at a roadless field camp count fish: instead of grinding through hundreds of hours of sonar video, collapse hundreds of frames into one 2D picture — an echogram — and let a lightweight neural network read the count off the image in a single pass.

  4. Imaging Sonar: Counting Salmon in Pitch-Black Water

    When the water is too muddy, too deep, or too dark to see, sound takes over.

  5. How Humans Teach AI to Count Salmon on Video

    How does a machine learn to see a salmon? One click at a time.

  6. Drones and AI Are Now Counting Bristol Bay's Salmon

    On a Bristol Bay riverbank, a solar-powered box opens each morning, launches a drone on a pre-set path, and streams aerial video of red sockeye over sunlit gravel — a view managers never had.

  7. The Secret Life of a Sockeye Salmon

    What are we actually trying to count — and why is it so hard to see? A sockeye is born in freshwater, vanishes into lakes and the open Pacific for years, then finds its way home to spawn and die.

  8. Escapement: How Counting Salmon Runs Alaska's Fisheries

    A single number — escapement, the count of salmon that 'escape' the nets to spawn — governs a multi-million-dollar fishery and the families who live on it.