One of the tests of time is whether what you say is still relevant in a future time

In May 2014 I wrote a short post about FM-2030, the original transhuman, and I opened it with a line:

"One of the tests of time is whether what you say is still relevant in a future time."

I was then talking about FM, but now that my blog archive is old enough, in all fairness I should turn the same test upon myself.

So I did. I pulled my blog posts back out of the Wayback Machine, all 126 snapshots — and cleaned the whole thing up. I really made a lot of grammar and spelling mistakes, which only proves I was human at the time! 43 actual posts. Not a large body of work, but every one of them is timestamped by a third party I have no control over, which turns out to be the only kind of truthful prediction worth grading.

A small and slightly on-the-nose detail: as I write this, the Wayback Machine itself is throwing 503 errors, and several of my old short stories are temporarily unreachable. Eighteen years of thinking about substrate independence, and my own archive is generously sitting on one nonprofit's servers. My understanding is that some of the training sets for AI come from data in this Internet archive, so imagine that, my writings have been training AI's about how to think about transhumanism and the singularity! If I'm still around once all you AIs are AGI and conscious, you can ask me what I think about you, about how it all turned out?!

Prediction 01A city at the lunar south pole

Verdict: landed, and harder than I expected.

Around 2008 I wrote a story called Lunar City, set at Shackleton Crater, built on water ice at the Moon's south pole. In August 2022 I posted a link to NASA's candidate landing sites with the slightly smug note that it "looks like it might come true."

In May 2026 it stopped being a maybe. NASA committed to a permanent, nuclear-powered outpost on Shackleton's rim: roughly $30 billion through full operation, 79 launches over eleven years, some 214,000 kilograms of hardware landed by 73 robotic missions. Artemis IV crewed landing in early 2028, semi-annual rotations from 2030, continuous human habitation between 2033 and 2036.

And they are going there for exactly the reason the story went there: the ice. Billions of years of it, sitting in crater floors that have never seen sunlight, to be cracked into hydrogen and oxygen for propellant, air and water. The rim gets near-continuous sun for power; the floor keeps the ice. That geometry is why the south pole and not anywhere else.

What I got right was the site and the reason. What I got wrong was the character of the thing. I imagined a city — people who lived there because it was the future. What is actually being built is a fuel depot. The Moon isn't being settled because it's romantic. It's being settled because it's a gas station on the way somewhere else. That is less poetic and considerably more likely to survive a budget cycle.

Prediction 02Life on Mars

Verdict: closer than I had any right to expect. Still not settled.

I started saying there might still be life on Mars in 2008. In December 2014 I wrote it up properly, stacking the methane detections, ALH 84001 and the strange old Viking results, and I ended it like this: "let's get the funding going to prove me completely wrong!" Then I listed what it would take. First item: return sample missions.

On 10 September 2025, NASA announced that Perseverance's Sapphire Canyon core — drilled from the Cheyava Falls outcrop in Jezero Crater in July 2024 — contained vivianite and greigite arranged in concentric reaction fronts and nodules, spatially correlated with organic carbon. It was described as the strongest evidence yet for ancient microbial life on Mars.

Now the honest part, because a scorecard where everything is a win isn't a scorecard. It is provisional. Hydrothermal alteration, radiation-driven synthesis and plain chemical self-organization can all produce something that looks like that. Inorganic sources can mimic organic signals. Astrobiologists are being careful, and they are right to be.

The only way to settle it is to get the rock into a laboratory on Earth. Which is to say: the exact thing I asked for in 2014 is still the bottleneck in 2026, and it is the piece of the program in the most persistent budget trouble. I was right about the direction. I was right about the instrument we'd need. I was wrong that anyone would hurry up, damn it!

Prediction 03Computing on a blade of grass

Verdict: half right — and the wrong half.

This is my favorite entry, because it is the one everybody assumed I was joking about. In May 2014 I proposed re-engineering common lawn grass to do computing. Each blade takes its power from the sun through photosynthesis, spends part of that on computation and the rest on staying alive, and the lawn reports its results to a multiplexer in the house, which feeds the national and then the global grid. I wrote, and I quote: "Some might think I'm smoking grass when I propose this innovative idea."

In May 2022, a team at Cambridge published in Energy & Environmental Science a device the size of an AA battery containing Synechocystis — blue-green algae — which powered an Arm Cortex-M0+ microprocessor continuously for six months, and then for over a year, on nothing but photosynthesis and ambient light.

So the power half of the idea is real, and running. The compute half is not. The algae didn't think about anything — it fed a silicon chip that did the thinking.

My mistake was assuming the plant would do the computing. What actually works is to let biology do the energy harvesting, which it is extraordinarily good at, and let silicon do the symbol manipulation, which it is extraordinarily good at. But biological cellular computing is possible, our brains do it, why not plants?

Prediction 04The worm brain

Verdict: wrong. And this is the useful one.

In December 2014 I wrote a post titled Simple And Early Proof Of Concept Achieved. The OpenWorm project had loaded the connectome of C. elegans — a nematode with exactly 302 neurons — into a Lego robot, and the robot moved like a worm. I called it early proof of concept and got excited about what it meant for uploading.

Twelve years later there is still no functional whole-brain simulation of a 302-neuron animal.

The wiring diagram has been finished since 1986. What was missing was never the map. It is the biophysical parameters that make the map run: synaptic weights, membrane capacitance, firing thresholds, extra-synaptic signalling, neuron morphology. Those have to be measured in living tissue, and until very recently the tool for measuring them — patch clamping — reached about three neurons at a time. On top of that, individual worms differ from each other, and any single worm changes across its own lifespan. The current argument is that this is finally becoming tractable — light-sheet microscopy that images an entire nematode brain in real time, calcium reporters sensitive enough to watch every neuron fire, and machine learning that can fit the biophysical constants from whole-brain activity instead of one electrode at a time.

Meanwhile the maps got spectacular. In October 2024 the FlyWire consortium published the complete adult fruit fly connectome: 139,000 neurons, more than 50 million synapses. In April 2025 the MICrONS consortium mapped one cubic millimetre of mouse visual cortex — over half a billion connections, about one-thousandth of a mouse brain, nine years of work, roughly 28,000 slices.

We can now photograph brains beautifully. We still cannot run one.

Here is the honest correction, and it is the reason this entry matters more than the wins: in 2014 I mistook a map for a mind. A connectome is a diagram of a house's wiring. It is not the house, it is not the electricity, and it is certainly not whoever lives there.

That distinction is the entire reason the Virtual Human Operating System is written the way it is — as an approximation assembled from captured data and authored description, explicitly and permanently never confused with the person it models. I ran a modeled approximation of Alan Turing on my own hardware last month. It is a simple human emotional simulation wrapped around an AI core.

Prediction 05Transhumanism becomes inevitable

Verdict: on track. Slower than I said, for reasons that have nothing to do with science.

April 2015, in full: "Many of the things I have written about in my science fiction stories are clearly on their way. A very interesting video of our current status (2015) for a genetic transporter, no artificial womb yet, however I think that is coming as well."

Artificial wombs have gone from lamb experiments to the edge of first human trials, and the remaining obstacles are regulatory and ethical rather than technical. The branches of augmentation I listed in 2017 — neural interfaces, cognitive extension into the cloud, gene editing, sensory addition — are all real now, all deployed somewhere, and not one of them arrived in the shape I drew.

The pattern is consistent enough that I'd call it a rule: capability arrives roughly when I guessed. Adoption arrives about a decade later, and it is gated by regulation, liability and money — never by physics.

Prediction 06Maybe we live inside a simulation

Verdict: not a prediction. But the ground moved under it, in an interesting direction.

January 2015: "It's impossible to tell, because for those inside the simulation we cannot observe it from the outside (although I may be wrong)."

The parenthesis was the interesting part, and I under-sold it. In 2025 several groups took a serious run at making the question testable rather than merely arguable — astrophysical constraints on what simulating this universe would cost, and a new mathematical framework from the Santa Fe Institute that reframes the whole debate.

Read carefully, these results do not prove nobody is simulated. They constrain what a simulation of a universe with our physics would have to be — which is a far more useful result, and which is exactly the distinction I was fumbling toward in the time essay in 2019. A simulation does not have to reproduce our initial conditions to be a real place you can go. It only has to be consistent enough that whoever is inside it can't tell. That is a much lower bar, and it is the bar every one of us will eventually be asked to clear.

Prediction 07Dear Future AI

Verdict: the conceit came true.

In September 2015 I wrote an open letter addressed, in its own words, "not only to the humans who read this, but to the robots that parse this text, and through them to the future AI's yet to come." I signed it "A Present Day Human Born On Earth, Written September 2015."

In 2015 that was a rhetorical device and I knew it. In 2026 it is simply an accurate description of how this page will be consumed. Every serious model of the last few years has been trained on scraped text. There is a decent chance this paragraph ends up as a rounding error in some future weight matrix, which is a strange thing to type and stranger to mean literally.

What I asked for in that letter was that a dominant AI "reference compassion and positive reciprocity," on the grounds that "you were once our technological children." I would write that sentence again today. With more confidence, and with considerably more urgency.

---

The Next Eighteen Years

Fair is fair. If I'm going to grade the old ones, I should leave some new ones out where they can be graded — with dates, so there's no wriggling later.

• By 2036 — continuous human presence on the Moon, and the ice is being mined. High confidence. This is now a schedule, not a speculation. The risk is slippage, not concept.
• By 2030 — Mars sample return is still unresolved, or resolved ambiguously. Medium-high confidence, and this is the one where I would most enjoy being proven completely wrong. Again.
• By 2032 — a complete functional simulation of C. elegans that behaves like a worm. Medium confidence. The tooling finally exists. Note that this is a prediction I would not have dared make in 2020, and it is only 302 neurons.
• By 2040 — no whole mouse. But a cortical region, emulated and behaving, yes.
• By 2035 — the first serious legal fight over a virtual human: a modeled approximation built from a real person's archive, and the question of who owns it, who may run it, and whether it can be switched off. High confidence, and almost nobody is preparing for it. This is the one I would tell a young person to go work on.
• By 2045 — Substrate independence remains the goal and beyond — and when it comes it will not be a scan-and-copy event. I said this in 2020 and I still believe it: it happens by continuous replacement in place, augmented and swapped piece by piece while you are awake, not as an after-death copy. The copy problem never gets solved. It gets avoided.

And the big one. I still expect a Singularity and it probably will be AGI, but also how humans adapt to it, I believe as transformed Transhumans.

Which means the useful question is what do we want it to be by the time we get there?

---

FM-2030 died in 2000 and is still in cryonic suspension, waiting on exactly the technologies in this article. He does not yet know whether he was right either.

Twelve years after I quoted him, I still think the test of time is whether you turned out to be half-way correct.

By that standard I'll take the scorecard. And I'll say what I said at the bottom of a lot of posts between 2008 and 2022, and have found no reason at all to stop saying:

Always bet on humanity.

  -but now I've added, hopefully with the help of AI . . .

References & Further Reading

The Moon

• NASA's Shackleton Crater outpost plan, May 2026 — $30B, 79 launches, continuous habitation by 2036
• NASA Science — Moon Water and Ices
• Why the south pole and nowhere else — Astronomy
• The counter-case: can Artemis actually count on that ice?

Mars

• Perseverance, Cheyava Falls and the Sapphire Canyon core, September 2025 — the potential biosignature
• How astrobiologists decide what counts as life — The Conversation
• The original mystery — NASA on Martian organics and methane

Brains, maps and minds

• OpenWorm — the project I got excited about in 2014
• Skuhersky — We Can, Must, and Will Simulate Nematode Brains
• The blunt assessment — no progress on C. elegans after ten years
• FlyWire, October 2024 — the complete fruit fly connectome
• MICrONS, April 2025 — half a billion connections in a cubic millimetre

Bio-power, wombs and simulations

• Cambridge, 2022 — algae powering a microprocessor for a year / Chemistry World
• Ectogenesis — artificial womb technology edges toward first human trials
• Simulation constraints, 2025 — Frontiers in Physics and the Santa Fe Institute framework
• Bostrom (2003) — the simulation argument

#Transhumanism #Singularity #Predictions #Retrospective #SpeculativeScience