Route 37
A Field Report from the Compressed Decade

Mary-Ann and I have been overlanding Southern and East Africa for close to three decades. We travel in a 2009 LC76 Station Wagon — mechanically simple, deeply repairable, and exactly as capable as we need it to be. No more.

At the moment we are camping on the banks of Lake Victoria in northern Tanzania. It’s late May and the rains have started to subside. The Land Cruiser LC76 — or Nzou Yedu (our elephant) — is facing the last of the afternoon. I’ve just started the fire, and the 1HZ engine has been silent for nearly three hours. The acoustic array clipped to the awning support has logged the third fish eagle of the day. From here we can see, on the dam below, one of the new floating photovoltaic rafts the Tanzanians installed in 2033 — Chinese panels, locally owned, generating power for three communities and a clinic that did not exist when we drove through here in 2019.

This is the world we move through now. The Compressed Decade — the name they ended up giving the years between 2025 and 2035 — has changed almost everything that can be changed by intelligence applied at speed. And almost nothing about why we are sitting here, in this exact spot, watching the sunset.

This is a field report from that paradox.

When we began overlanding Southern Africa thirty years ago, travelling Route 37 through Southern and East Africa, the constraints were geographic and informational. The roads were poor, fuel was scarce and reliable communications were non-existent. Paper maps were the only navigation tool. The decade just past has dissolved most of those constraints. What it has not dissolved, and what we did not expect, is the value of the practice itself. If anything, that has become clearer.

So this is a long account, written by a camp fire on Lake Victoria, about what changed, what didn’t, and what overlanding looks like to two people who chose to keep doing it through the strangest decade in living memory.

Dario Amodei, in Machines of Loving Grace, framed it like this: powerful AI would let humanity compress fifty to a hundred years of biomedical and scientific progress into five or ten. Most of us reading that in 2024 found it ambitious. By 2030 it was clearly a conservative prediction, at least for biology, materials, and a few other domains. By 2035, sitting here hearing the call of the fish eagle, it is the texture of ordinary life.

But not evenly. The compressed part of the decade has been ferociously uneven, and that is the second thing you have to say about it.

In Southern Africa, here is what AI actually delivered.

That is the Machines of Loving Grace side of the ledger, applied here.

The other side is also real.

The second Amodei essay — The Adolescence of Technology, written in early 2026 — was the one that named the part we didn’t see coming, or didn’t take seriously enough. In the chapter called the “Player Piano,” after the Vonnegut novel, Amodei’s warning was this: the same general cognitive capability that made the medical and scientific breakthroughs possible was, simultaneously, replacing human knowledge work across the economy in ways the labour market and the welfare state had no plan for.

This is what happened to most of our friends.

Between 2027 and 2032, the white-collar economy in Cape Town contracted in ways none of us had expected. Auditors went first — which is to say, my own former profession, the one I had retired from a few years earlier. Then junior analysts in finance, most of the legal associates, pharmacists, mid-tier journalists, translators, half the architects. Radiology, predictably. Software developers above and below the median. People who had thought of themselves, with good reason, as the strong middle class. The shock did not come from the technology being unreliable — it came from it being slightly more reliable than the people it replaced, and considerably cheaper.

The South African government, like most governments globally, lacked a coherent policy response. Scandinavian and Korean experiments with Universal Basic Income ran ahead of the SADC. Namibia and Botswana launched small pilots in 2031 and 2032; both remain modest and contested, and the rand-based equivalent in South Africa is a means-tested grant that arrived too late and was too small for too many. Many of our friends became what the economists now call the displaced middle. Some retrained. Some emigrated. Some did what people in earlier industrial revolutions did, which was to find work in the parts of the economy AI could not touch — sectors that are intrinsically embodied, relational, local, and slow. What the sociologist Hartmut Rosa calls the domains where resonance — the experience of being genuinely affected by the world, and affecting it in return — still depends on a human body being present.

The strangest result, and the one this letter keeps circling back to, is that some of them started overlanding.

Not many. But more than before. We have met them all over the subcontinent in the last three years — ex-analysts, ex-consultants, ex-academics, sometimes solo, more often in twos or small groups, working seasonally as field assistants for the conservancies, or doing soundscape annotation, or guiding for the small operators who specialise in what is now called attentional travel. Not all of them are happy. Most of them are not wealthy. But they are doing something the player-piano economy could not let them do, which is to put their attention on the world directly, and be paid, albeit modestly, for it. The Compressed Decade had taken something from them — the assumption that being competent meant being employed at a screen — and overlanding had restored a different competence: what Hannah Arendt called natality — the power every newcomer has to start something genuinely new. That capacity is older than any economy and survives the loss of any economy.

Amodei wrote that human purpose does not depend on being the best in the world at something — that purpose comes from stories and projects we love, and that we have to “break the link between the generation of economic value and self-worth and meaning.” He wrote this knowing the transition would be painful. And it has been. It is. But the people we meet on the road — the ones who found their way out of their old careers and into a quieter, more bodily relationship with the work of being alive — are some of the most steady, observant, and interesting humans we have travelled with in thirty years of doing this. They are not consumers of the wild. They are citizens of it, in the sense the ubuntu principle has always insisted on: umuntu ngumuntu ngabantu — a person is a person through other persons. The web that holds them is not virtual. It is the track, the camp, the fire, the people whose land they are passing through.

Overlanding did not save them. But it offered something the displaced middle desperately needed, which is a structure of meaning that did not depend on outcompeting an AI. And the particular technology that ended up making the practice possible, in its richest form, is the one I want to spend the rest of this letter describing.

Of all the technologies that emerged during the Compressed Decade, the one that has most quietly reshaped and deepened our travel is the one that began not with a map, but with an ear.

To explain its place on the LC76, I have to start with the practice it had to fit into rather than replace.

We are not campers. We camp so that we can keep travelling. The LC76 is our porch — not a tool but an organ of perception: open to weather, dust, and noise, built for participation rather than insulation. We travel solo — just the two of us — which means slow and deliberate. Two-night minimums at any camp worth staying at. Two hundred kilometres a day, not six hundred. We leave Cape Town with an empty fridge and buy from the people along the way. The itinerary is always a hypothesis. The road gets the final word.

We are drawn to the remote and the unhurried — not for the difficulty, though difficulty is part of it, but for what extended time in wild places does to the quality of attention. That is the whole point. We don’t travel to see different things. We travel to see things differently.

Attention is the practice. We have come to understand that paying close attention to a place — really listening to it, without rushing to name or interpret what we hear — is closer to meditation than to tourism. You sit still. You let the place come to you. You resist the impulse to classify and instead stay with the sound itself, the smell, the slow change in the light, before you reach for the name. Most of what is most worth knowing about a place is found in the slow time after the impulse to know it has been let go. What Bergson called durée — lived time, saturated time, the kind of temporal depth that speed structurally forecloses — is what the two-night minimum and the 200-kilometre day are designed to protect.

This commitment is older than the technology. The technology had to prove its worth against it.

Most of what we considered adding to the LC76 failed the test. Route optimisation, algorithmic camp recommendations, continuous content capture — each was a substitute for attention dressed up as an enhancement. The question we put to every new technology was simple: does this deepen what we already do, or replace it with something easier? Only one passed. The rest didn’t make it onto the vehicle.

The technology that succeeded did so because it addressed a problem that others could not. Every other AI technology we considered operated on the visual-navigational axis — maps, routes, terrain and logistics. Each extended the vehicle’s capabilities to what it could do in the world. The listening machine was the first technology to extend what we could receive from it. It is oriented inward rather than outward — towards permeability rather than capability.

Sound is the sense that carries the most time. A visual landscape is present-tense. A soundscape accumulates — the dawn chorus of today is partly a record of what happened last night, the water table last month and the fire last year. The biophony of a place is its biography made audible.

Our motto has always been about seeing — we travel to see things differently. The listening machine shifted the register, although not entirely. We had spent thirty years learning to see. Now we were learning to hear. That is not a refinement of the same skill. It is a different skill altogether, and it turned out to be the one the Compressed Decade most needed us to develop — because the losses that matter most in a landscape are not the ones you can see. They are the ones you can only hear, or fail to hear, in the silence where a dawn chorus used to be.

I call this technology the listening machine, though that is my name for it, not its actual one. The technical literature calls it an edge-based acoustic world model. The apparatus itself is a small array of directional microphones clipped to the awning support, feeding into the black box behind the dash. The chip doing the real work is the size of a fifty-cent coin and draws less power than a tail light. I have replaced more expensive components on the LC76 over the years for less impact.

The science underneath the AI is older than the AI by half a century. Soundscape ecology has been a serious discipline since the 1970s, built on a simple three-part distinction:

Every place on earth has its own signature combination of the three. Every signature changes with the hour, the season, the weather, and the presence or absence of every actor in it. There are people, a very small number of them, who have spent lifetimes learning to read these signatures unaided. Their ears can tell you, from a single minute of recorded forest, what time of year it is, what the weather has been doing, whether the dominant mammals are present or absent, whether the place is healthy or ailing. That capacity has always existed. It has been the work of a few patient lives.

What the AI added to this older discipline is not the capacity to listen but the power to interpret. The chip behind the dash can hear what we hear, plus a great deal more — sub-infrasonic ground vibration, ultrasonic insect frequencies, the directional shifts that pinpoint an alarm call before our own senses register it. A mastery that once required a lifetime of human devotion is now available to anyone with a Land Cruiser and four watts of power.

There is also a more urgent capacity: ecological diagnosis. Without a baseline, you cannot hear the loss. Driving through the Kaokoveld, I have no instrument for knowing whether the silence I am hearing is the profound quiet of an intact landscape or the eerie quiet of one that has already lost half its birds. The acoustic world model carries the baseline with it.

The question — the only question that matters — is what the machine is for.

Two versions dominate the market, and the difference between them is not technical. The chips are similar. The microphones are similar. The training manuals overlap. The difference is what Heidegger would have recognised instantly: the disposition the technology has been given toward the world it perceives. Heidegger spent a career distinguishing between technology that challenges the world into yielding its resources — what he called Gestell, the conversion of everything into standing-reserve — and technology that lets the world come forward on its own terms. The two listening machines are an almost textbook instance of the distinction.

The first version — the one most newcomers buy — is built to convert the soundscape into something legible at speed. Species checklists, alarm notifications, campsite acoustic-quality ratings, shareable data. The forest becomes a list of birds in it. The river shrinks to the seven species that depend on it. A silent night becomes a five-star rating. It offers what the anthropologist Clifford Geertz called a thin description — it catalogues what is present without illuminating what it means that these things are here, now, in this configuration, with you standing in the middle of it. Built for the content economy, it serves the audience back home rather than the listener in the camp. It treats the place as a standing supply of beautiful data, waiting to be extracted.

The second version — the one in the LC76 — is built for the opposite purpose. It is calibrated not to extract from the place but to deepen our presence inside it. It logs everything and surfaces almost nothing, speaking only when asked. When it does speak, it does not say “alarm” or “danger” or “species: lion.” It reads the ecosystem as a living text — what Geertz would have called a thick description, a reading that catches the webs of significance a place is suspended in. It says things like:

That is the entire output. No map. No alert. No recommended action. The machine has done what a good travelling companion does: it has sharpened our attention without replacing it. We are now more present than we were thirty seconds ago. The darkness has thickened. We are listening with our whole bodies.

These two versions are not merely different products. They embody two different relations to the place being listened to — one that reduces the world to a taxonomic inventory, and one that treats it as what Geertz called “an intricate web of significance” to be inhabited. One broadcasts its interpretation as a metric, which becomes a destination, which then becomes traffic. The other whispers them as invitation.

Here is the boundary no machine can cross. This is what Rosa means by resonance — the experience of being genuinely affected by the world around you, and affecting it in return. The listening machine can generate extraordinary description at extraordinary speed. But it cannot perform the act of staying in the encounter long enough for something to move between us and the world. That remains entirely ours. The technology doesn’t travel. We do.

We had to learn this. During the first six months of running the array, we made every mistake there was to make — checking the log every twenty minutes, treating notifications as commands, letting the system’s confidence become a substitute for our own attention. We turned off most of the features inside the first year. The configuration that survives is minimal and severe. The array listens. The chip thinks. The system speaks only when asked, or when something genuinely warrants the interruption, which is once or twice a week. Otherwise it is the quietest passenger we have ever travelled with.

To understand what the machine actually does on a real trip, the best thing is to describe a single place over three consecutive days.

Hlaro is a camp on the Runde River in Gonarezhou — Zimbabwe’s second-largest national park, still rough-accessed, still lightly visited. We were there six months ago on this same loop, before we worked north through Zambia and into the Tanzanian highlands and ended up here on Lake Victoria. What follows is what happened, with the small editorial liberty of compressing what the listening machine surfaced into prose.

The camp has one facility: a long-drop toilet behind a canvas screen. Everything else is the Chilojo Cliffs: four kilometres of red sandstone rising three hundred metres from the floodplain opposite, catching the last light each evening and holding it long after the rest of the landscape has gone dark. No fences. No other vehicles most nights. Elephants that know exactly what a Land Cruiser is and have decided opinions about it. We have been to Hlaro four times now. It is one of the places that changes us most.

We arrive, set up camp facing the cliffs, get the fire going. The acoustic array is unfolded and clipped to the awning support. The system begins listening immediately. We do not check it. We never check it on arrival. The first evening belongs to the unassisted ear.

We are each sitting with a glass of wine watching the Chilojo Cliffs go through their evening sequence — amber, then deep red, then the colour of dried blood — when the lions walk through. A pride of six: two adult females leading, then three sub-adults, then a large male bringing up the rear. They come from the cliffs on our right, cross within fifteen metres of the LC76, and continue unhurried along the river bed. The male pauses. He turns and looks at us with the complete, unhurried attention of something that has no predators. Then he follows the others…

Nobody speaks. The Chilojo Cliffs are still burning. The acoustic system is logging the entire event — the sub-infrasonic contact rumbles between the animals, the suppression of every other species across a 200-metre radius as the pride moves through, the gradual, tentative return of the biophony after they pass. We decline to read this data until the following morning. Tonight, the encounter remains entirely unmediated.

Over coffee, as the cliffs intercept the first level light of dawn, we open the acoustic log for the first time.

The model recorded the pride’s return upstream at 02:40. Between 03:15 and 04:00 it logged what it interprets as a large buffalo herd moving through the drainage to the south — the ground vibration, the low contact calls, the sharp alarm snorts of the oxpeckers riding the animals. A second biophonic suppression at 04:55, briefer, possibly the male lion returning alone. By the time we review these recordings, the espresso bubbling in the Bialetti, the dark hours have been reconstituted. Night is no longer an empty gap between days; it has been revealed as a populated, active space.

A herd of buffalo — we count close to fifty — materialises from the Mopane woodland and moves across the floodplain below the cliffs. The acoustic system has been tracking their approach for twenty-two minutes before they appear. I check the log afterwards: it detected the ground-contact signature of the herd at 800 metres, the dust-suppression effect on the high-frequency insect register, the directional shift in the oxpecker alarm traffic. Not a warning. A record. The herd already knew we were here. The system knew the herd was here. Only we did not. Now we do.

We take the LC76 out at first light and pick up the male lion’s tracks on the sand road south of camp. We follow them for nearly two hours, at walking pace, the 1HZ at idle. The tracks lead us into a thicket of Mopane, and there he is — sixty metres ahead, walking with the slow purpose of a predator on a known route. We follow him at a respectful distance for forty-five minutes as he moves through the bush. He stops twice. Looks back. Continues. The third time he stops he simply lies down in a patch of shade and watches us. We watch back. Nobody has anywhere to be.

When we get back to camp and pull the log, the system has caught the entire encounter. A sequence of contact calls between him and another lion (presumably one of the pride females, out of our visual range). A mild alarm exchange that the model interprets, with high confidence, as being about us. A reassurance sequence. Extended low-frequency vocalisations consistent with family cohesion calls. We were the subject of a conversation we could not hear. The model heard it.

Whether that knowledge enriches or diminishes the encounter is a real question. I think it enriches it. They were not simply an aesthetic event. They were a social world, deliberating about us, and arriving at a position. That is more interesting, not less, than the postcard version. And the machine did not see them for us. We saw them. The machine heard the conversation underneath. The hearing came after the seeing, which is the order it ought to come in.

Before we pack up, I spend an hour with the three-day acoustic log. This is the capacity no amount of attentive listening could replicate alone: the trend across time.

Dawn chorus, day one: 38 species active in the first hour of light. Day two: 37. Day three: 33, with a notable absence of the ground-level species that had been active on arrival. The model flags this against its baseline for Gonarezhou in the same week of the dry season: statistically anomalous suppression of mid-storey and ground-layer species, pattern consistent with sustained large-predator activity in the immediate area.

The lions have been here every night. The landscape has been registering them. We have been present to three days of that registration without knowing its shape. Now we know its shape. The three days feel different in retrospect — not reinterpreted, but thickened. We have not been sitting quietly in a beautiful place. We have been inside the nervous system of a place that has been continuously processing information about its own inhabitants, and we have been, occasionally and briefly, part of what it was processing.

On the second day, during the meditative lull before the buffalo arrived, I caught myself checking the system twice without any external prompt. Pure habit — the instinct to manage data rather than attend to the world. The technology did not cause this. I did. In 2035, as now, the discipline is the scarce resource — not the data. The listening machine is only ever as wise as the person who possesses the restraint to keep it quiet.

The listening machine works as well as it does — ethically as well as technically — because of three pieces of the Compressed Decade that locked into place around it. Each is, at root, an anti-Moloch structure — a deliberate institutional defence against the trap that no individual intended but that the collective logic of optimisation assembles: the conversion of every particular human or natural good into standing-reserve. As Liv Boeree has put it, wisdom outranks intelligence; knowing which game to play matters more than winning the wrong one. These institutions are attempts to choose the right game.

The Caprivi Linguistic Archive, mentioned earlier, runs on the same principle — a multilingual model built with elder oversight, owned locally, accountable to the communities whose languages it carries. It is situated knowledge given institutional form.

The Cross-Border Health Corridors (which is why the mRNA cassettes are in our fridge) and the Continental UBI Pilots in Namibia and Botswana sit alongside this as part of the larger architecture. None of it is finished. None of it is enough. All of it is the actual material from which an amazing future, in Existential Hope’s phrase, could be built.

Most of what we do has not changed. We still leave Cape Town with an empty fridge. We still drive two hundred kilometres a day, not six hundred. We still stay two nights minimum at any camp worth staying at. The LC’s 1HZ engine still tells us most of what we need to know.

But five practices have become deliberate, in ways they were not before — and four of the five exist because of the listening machine.

1. We declare our zonesBefore any trip, we map out the places we are going through that fall under Off-Spectrum jurisdiction, and the places we are choosing to treat as Off-Spectrum even where the law does not require it. The array gets folded down and stowed when we enter those zones — not just switched off, physically packed away — and it stays packed until we are out again. The box behind the dash respects the zone declaration. Discipline is easier when it is structural.
2. We pay the data taxWhen we are in a Conservation Acoustic Trust zone, we use the system and we pay for it. Where we are not, we generally don’t run the array at all. It is not that we are against the technology. It is that we have decided not to extract data from places that are not yet in a position to be paid for it.
3. We listen before we readWhen we arrive in a new place, the array is set to log silently, and we do not check its outputs until the morning of day two at the very earliest, and ideally not until day three. The first evening belongs entirely to the unassisted ear. We learned this slowly, and it has changed what we hear. The machine knows more than we do about most places. That is fine. But the order of perception matters: our ears first, the machine’s afterwards. The other way round and the machine’s interpretation overwrites our own before our own has had a chance to form.
4. We carry the cassettesEvery long trip carries at least one mRNA cold-chain cassette for a clinic somewhere on the route. We have done this for three years. It costs us a small fraction of fridge space and adds nothing to the driving. The clinics we have delivered to are some of the most interesting places we now visit. They are where the Compressed Decade looks most like what was promised.
5. We refuse the content economyWe do not post in real time. We do not share GPS coordinates of any campsite. We do not run the version of the listening machine that produces shareable acoustic-quality scores. We do not contribute to the algorithmic recommendation systems that are slowly turning the most beautiful places in Southern Africa into queues. This is the Moloch trap applied to landscape: optimise hard enough for a proxy of beauty and you eliminate the beauty itself. If we write about a place — like this — we write about it after we have left, and we leave out what would let an algorithm find it. This sounds like a small thing. It is not.

Amodei opens Adolescence of Technology with a scene from Contact, where the alien-contact protagonist is asked what single question she would put to the older civilisation. Her answer: how did you survive your technological adolescence without destroying yourself?

That is the right question.

We are not in a position, sitting here on Lake Victoria in 2035, to know whether humanity has come through its adolescence well. Most of the evidence will not be in for another generation. What we can say is that the Compressed Decade arrived more or less as Amodei predicted, that the upside has been almost exactly as radical as he hoped, and that the downside — the player piano, the displaced middle, the uneven distribution — has been almost exactly as painful as he feared. The technology was not the test. The test was, and is, the maturity of the systems and the humans wielding it.

Every technology that arrived this decade carried both its danger and its saving in the same circuit — the same chip, the same code, the same product. The gene therapy that cures sickle cell is the same molecular technique that could, in other hands, be used to do much worse. The grid that lights the clinic on the Runde is the same grid that, in its centralised form, would have entrenched the old gatekeepers indefinitely. The listening machine that deepens our presence in a place is the same listening machine that, in its other version, reduces that place to a stack of shareable data. The patient work, instance by instance, is telling them apart and insisting on the version that saves. We do not yet know if humanity is doing this work well enough. We know it is the work.

AI has made the logistics of remote travel far less frightening though. Predictive maintenance means fewer catastrophic failures. Translation tools mean border crossings that once cost half a day in sweating frustration now take forty minutes. Satellite connectivity has arrived even in the Khaudum and western Zambia. Routes that were genuinely unknown territory a decade ago now have crowd-sourced telemetry, seasonal flood models, and real-time fuel availability data. Merely reaching remote places has become less differentiating.

The consequence is that two sharply different types of overlander now populate the same tracks.

Optimised Mobility. Highly connected. Efficiency-driven. Route-maximising. Content-oriented. Travels with AI assistants that pre-interpret destinations. Covers more ground with less uncertainty. Arrives everywhere already knowing what to expect. The journey as execution.

Perceptual Travel. Slower. Relational. Selectively disconnected. Stays longer. Chooses depth over distance. Actively resists pre-interpretation. Uses AI only to deepen encounter, not to manage it. The journey as transformation.

Overlanding, at its best, is a small-scale practice of perceptual travel — moving through places without consuming them. Using capability to deepen attention, not to substitute for it. Refusing the optimisation logic when it does not serve the deeper purpose. Carrying what you need. Taking what you can use. Leaving what is not yours. Staying long enough to be changed.

For us, the listening machine has turned out to be the clearest instance of the larger principle. It is a move we didn’t expect, from an angle nobody had learned to look from. The listening machine’s best outputs have that quality. They are Route 37 moments — readings of a place that no human ear could have produced, but which, once heard, are obviously right.

It is a piece of the Compressed Decade that earned its place on the vehicle by becoming smaller, quieter, and more deferential to the human task of paying attention — not by becoming louder, more confident, or more recommendation-driven. It is the version of AI we hoped for and were not sure we would get. Most of the AI products that arrived this decade chose the other path. We do not use those products. We use this one. The distinction matters.

The motto we have travelled with for years — we don’t travel to see different things, we travel to see things differently — was an epistemological commitment before the technology arrived. It is now, increasingly, a political one too. In a world where intelligence has become abundant and attention has become scarce, the willingness to be changed by what you find is no longer a luxury or a sentimentality. It is the practice on which a liveable future depends. It is also, in its own quiet way, what Arendt called natality — the capacity to begin again, to arrive in a place and let the place re-make you, however slightly, every time. To keep becoming, together, through encounter rather than arriving, alone, through optimisation. That capacity is not optional. It is what makes us the kind of creature whose technological adolescence can have a survivable end.

We do not know if humanity is going to pass its test. We do not know if the gains of the Compressed Decade will reach the people who most need them, or whether the Off-Spectrum treaties will hold, or whether the UBI pilots will become real, or whether the next ten years will repeat the pattern of the last with even more force.

We know the motto. We know Nzou Yedu. We know the listening machine, and how to use it without letting it use us. We know the sunset on Lake Victoria at the end of a long day.

It is, when all is said and done, enough to keep driving.