Navigating the deep waters of AI
By Psysig Research
In May 2022, Lehua Kamalu took a satellite phone call from the deck of Hōkūleʻa, a double-hulled canoe a few hours out of Hawaii, bound for Tahiti. She had minutes to talk. Then she hung up, and there was no calling back. Ahead lay 3,000 miles and roughly twenty days of open ocean, navigated entirely by sun, stars, wind, and swell.
Kamalu is Hōkūleʻa’s first female captain and navigator. In 2018, she became the first known woman to captain and navigate a long-distance ocean voyage — 2,800 miles from Hawaii to California — without modern instruments. “Everything is done mentally,” she told National Geographic. “You are tracking the wind, you’re tracking your cruise speed, you’re adjusting the sails.”
The boat came first.
It won’t surprise most people that Polynesian voyagers did not wait for a perfected navigation system before building the boat. Ocean-going canoes — the hull, the lashings, the sail — came first.
The craft of crossing deep water on purpose took generations longer to develop. It required a star compass memorized by heart, the ability to read island positions in ocean swells, and routes through the sea’s invisible “on-ramps and off-ramps” — all of it carried in memory and passed from master to apprentice.
The technology to traverse the water existed for centuries. The knowledge to navigate it had to be built, tested, lost, rebuilt, and finally proven in public.
Capability first, navigation next
Figure 1 · Wayfinding
Capability first, navigation next
Voyaging and intelligent systems share an arc: the vehicle arrives before the craft of steering it. Today, we are mid-voyage on the second.
AT SEA
The hull exists
Ocean-going canoes capable of coastal hops, then deliberate settlement.
The craft assembles
The star compass, swell maps, and sea roads — built and tested over generations.
Deep water, on purpose
Hōkūleʻa sails Hawaii to Tahiti in 34 days, guided entirely by human craft.
IN THE SYSTEM
The model exists
Generative models can already cross an astonishing amount of data.
The craft assemblesWe are here
Assembling verification, boundaries, and routing logic.
Deep water, on purpose
AI execution that has actually earned our trust.
This is roughly where artificial intelligence stands today.
The boat exists, and we can set sail. A generative model can already traverse an astonishing amount of water: it can draft the contract, answer the ticket, and write the code. The boat floats, and it is fast.
What we are still assembling is the wayfinding system — not the capability to move across the water, but the systemic frameworks that tell you where you actually are, whether your heading holds under pressure, and what to do when the environment shifts.
Kamalu’s crew sails without mechanical instruments because they rebuilt the skill deliberately. The point of their voyage was never to reject navigation; it was to prove that navigation is a craft — learnable, teachable, and only trustworthy once it survives open water.
AI requires the exact same evolution. We don’t need bolt-on mechanical instruments; we need integrated frameworks of governance.
The hull was never the hard part
Figure 2 · Wayfinding
The hull was never the hard part
Deep-water craft is framework plus training. At sea, it took generations. In the system, it is the work in front of us right now.
01 · AT SEA
The compass
A ring of rising and setting stars, memorized to provide bearings without a physical instrument.
IN THE SYSTEM
Verification
Reasoning that proves exactly how an answer was reached, not just the answer itself.
02 · AT SEA
The swell
A trained hand feels how land bends the ocean, finding islands in wave patterns.
IN THE SYSTEM
Boundary checks
Signals that hold their integrity under pressure — including input designed to bend them.
03 · AT SEA
The sea road
Invisible highways: known on-ramps and off-ramps of wind and current, kept in memory.
IN THE SYSTEM
Consequence routing
The level of systemic care strictly matches what is at stake on this leg of the voyage.
The chart is still being drawn
When the Polynesian Voyaging Society sparked a renaissance in traditional wayfinding in the 1970s, master navigators didn’t just hand their students a map. They trained apprentices, who trained more navigators, across fifty years of voyages that tested the craft a little further each time. Capability and navigation grew together, with each successful voyage funding the confidence for the next.
We expect AI to work the same way. The models will keep getting bigger and faster — that part needs no encouragement.
The work that needs doing is quieter: charting the deep water, building the verification frameworks, and training the systems to route risk appropriately. That is the side of the craft we work on at Psysig.
Because the ocean does not care whether your boat is beautiful. It only cares whether you can find your way.