Timeline of the near future


2024: SpaceX Starship test flights. This may include SpaceX payloads, but not necessarily. Likewise, it may involve a launch to Mars (like the Falcon Heavy test flight), but not necessarily (the launch window for that is October-November), and I don't expect that launch to involve a successful landing on Mars. First Lunar fly-by more likely but still not certain.

2025: Useful Starship launches. First Starship attempt at the Moon is more likely here, I'd guess 80%. No Mars launch window.

2026: Businesses are hard to forecast even a few years ahead. Assuming SpaceX hasn't imploded by this point, I expect them to have landed on the Moon by the end of 2026. A Mars attempt may (67%) have been launched, but the 2026 launch window is November-December so a launch then won't arrive in 2026.

2026-2027: Someone will attempt to land a regolith processor on the Moon, experiment with in-situ aluminium production. The first attempt is unlikely to succeed, because we've not done it before. First demo will probably be a proof of concept that does 1 gram and then stops forever; subsequent units may be 1 kg/Lunar day (14 Earth days); needs to be able to output something like the mass of the processing factory itself before failing in order to be worth bothering with, and I can't guess how long that will take.

2028-2029: Next Mars launch window (December-January). If SpaceX doesn't have a demonstration Sabatier process plant that fits into a Starship by the beginning of 2028, they've stopped caring about going to Mars. If SpaceX didn't launch in 2026 (and are still around), they will attempt at least one Starship flight there during this launch window. (If SpaceX no longer exists, nobody else really cares that much about Mars). As 2028 will be a US presidential election, assume that there will be strong pressure to get human feet on the Moon before the January 2029 inauguration of the next one.

2029 onwards: Everything beyond this point is heavily dependent on unknowns, including the political — government leaders may decide they don't like Musk any more and start looking for ways to specifically undermine him personally. No Musk ⇒ No Mars.

If the economic environment supports it, there may be some early attempts at private space stations/space hotels around 2028-2030. We've already seen some work on this, but they can't succeed without a launch system at least as cheap as Starship.

The SpinLaunch system would be an excellent fit for the Lunar environment, as one of SpinLaunch's biggest limitations is the atmosphere both during spin-up (needing a huge vacuum chamber with a fast-acting door) and ascent (drag), while the Moon is famously airless. Also, the main payload arm can reasonably fit (or be designed to fit) into the payload bay of Starship, so it could be made on Earth prior to lunar ISRU being any good.

Because SpinLaunch can fit into a Starship, there's probably no chance of anyone making a prototype orbital ring — although this would be just about possible with Starship and made easier by having a SpinLaunch system on the Moon, it currently looks like SpinLaunch would be cheap enough that people will ask "why bother?" for an orbital ring. That said, an orbital ring can double as a power line, and any Moon base needs a way to get power during the 14-day-long Lunar night, so it's not impossible that someone will have at least seriously proposed it and be doing pathfinder missions for it.

By the end of the decade, we will either have a tiny permanent base (like McMurdo Station) on the Moon, or we will have learned that we really don't want to go to the Moon.

One thing to watch out for is Tesla's robotics ending up in SpaceX's missions. Partly because Musk is a salesman and he has previous form, but also because a Cybertruck is the kind of vehicle you want on the Moon or on Mars, while an android (even if used as a telepresence device, it doesn't need to have a full on-board AI) is a much better choice than a human for EVAs, as the entire process of an EVA is both clumsy and dangerous and space suits themselves are complex to the point of being almost single-person spacecraft.


2024 onwards: wildly unpredictable at every stage and in every regard.

Narrow AI: Lots of breakthroughs each year, but no obvious reason to expect them in any particular order. Which is going to happen first: decoding the language of the North Sentinel Islanders or automatically determining synaptic connection strengths in-vivo and without the current requirement of genetically modifying the brain you're scanning? And does it really matter?

GenAI — Strike action and economics: For much of the industrial era, the only leverage workers have had is to threaten to withdraw their labour. This works for issues of pay and conditions, but it cannot work where the issue is the owners wanting to have less workers for whatever reason, as you're accidentally threatening the employers with a good time. This also applies to GenAI: for now, LLMs are good enough to be interesting, even good enough to replace interns, but not really good enough to replace a skilled professional with years of experience. The image and video generators are a bit closer to the mark, in that a close eye can still spot the mistakes but a steadily increasing fraction of the population don't look for the things the GenAI do wrong.

Also note that the quality of AI varies by task, and this argument holds across all tasks: any given AI may be "good enough" to make the members of the Writers Guild of America unemployable before or after it makes me personally as a software developer unemployable. It's impossible to say at this point, as the kinds of mistakes that current (as of March 2024) LLMs make have different kinds of impact: a script saying something a little off may be corrected by an actor using "common sense" (whatever that means to you), or it may be a plot hole that unravels the whole thing; some C using a while loop instead of a for loop probably won't matter, but subtle code mistakes frequently cause disasters even at the best of times.

Cars/Self-driving systems: Again, no idea. Back in 2009, I was absolutely sure that by 2018 it would be possible for a normal person to go to a car dealership and buy a new model whose self-driving AI was so good that the car didn't have, or need, a steering wheel. This clearly didn't happen, even though Waymo has demonstrated limited robo-taxis that meet this requirement in some limited regions.

Androids: all the problems with self-driving cars, but with a more complex body with more degrees of freedom, and operating in even more diverse environments. Even ignoring the need for both governments and public liability insurers to be satisfied, I expect there to be a 5-10 year gap between the necessary compute requirements fitting into a self-driving car and fitting into a human-sized android.

AI safety/alignment: No two people can agree on what "good" is, what counts as "harm" such that any given output can even be considered a risk in need of being avoided in the first place, or if it is possible/good/bad to focus on short-term risks such as personalised propaganda or deepfakes vs. longer-term issues such as getting paperclipped (which some people insist can't be possible for an AGI, asserting that it "must" share human morals and values, but I've never seen this convince anyone who didn't already believe it). We still have people suggesting we can just turn the AI off at the same time as other people are criticising OpenAI for not releasing model weights even though releasing model weights basically guarantees someone somewhere will be running it at any given moment (even well past the point better models become available).

Therefore, I hope — hope — that we make absolutely no significant progress on state-of-the-art AI capabilities for a few years, even if we spent that time continuing to call each other by the adult equivalent of playground insults rather than double-cruxing to figure out what we even think we're talking about.


Hardware: No interesting developments. We've already got super-human speed and strength and have had it around the invention of the steam engine, the limit is purely in the intelligence to keep them safe around humans.

AI/Software: it will improve over the next decade, but I can't predict how far or how fast. Despite the current interest in humanoid robots, I think we're likely to get more quadrupeds (like Spot), wheeled robots, and flying drones, and that for any given role, one of these other form factors will be widely used for several years before androids get good enough at that specific thing to replace them (if indeed they do); and that the AI necessary to generalise over robot bodies will, once developed, take 5-10 years to shrink down to the scale of fitting in the power envelope of a robot.

Further note: while one can easily imagine separating the brains of a robot from its body, WiFi amongst many other possible solutions, this is not something that scales up to the level of being disruptive to a workforce — if a robot brain takes 2 kilowatts for safe operation at human speed, the current global electricity supply can support 1 billion such robots if that was all it did. If we take the European experience of "about half the population works about 40 hours per week" (i.e. just under 25% of the time) then any given robot could do the work of 8 people, but that's still using all of our electrical power just for the robots' brains. I do expect the growth of renewables to boost the available power supply, but on a similar timeframe that I'd expect the compute/model efficiency to improve to the point we don't need it — at earliest, late 2020s/early 2030s, even if the AI in cars (my starting point for the 5-10 year countdown to androids) was ready tomorrow.


Too far outside my field to guess — While there have been several interesting developments in the last decade enabled by AI (e.g. cheaper DNA testing and protein folding solutions), and of course mRNA getting famous (even though it wasn't new) for use in Covid vaccinations, I can't guess how this research will turn into things people encounter in their normal lives.

Long term, the worldwide life expectancy at birth has gone up 6.3 years in the 20 years between 1999 and 2019 (and the European life expectancy at birth by 5.7 years in the same time period), only to drop again 1.8 years worldwide (2.1 years in Europe) over the course of the Covid pandemic. Assuming no more pandemics (a big assumption given the public is more interested in picking between team "wet market" and team "lab leak" rather than team "why not fix both"), and that the recent dip is just a one-off with no lingering issues from the pandemic becoming endemic and/or long-COVID (again, these are big assumptions) that suggests a European life expectancy of 82 in 2029, of 84.8 in 2039, and of nearly 100 in 2093 (and worldwide, this would be 76.1 in 2029, of 79.5 in 2039, and of nearly 100 in 2100).

That said, I'd be surprised if other trends didn't get in the way of these first: tech enabling synthetic pandemics or suddenly making novel cures much cheaper; global climate changes causing famines, droughts, or lethal heatwaves; rapid introduction of clean energy reducing direct pollution from burning particulate matter and not just long-term climate-impacts; robotics ending dangerous labour, or AI being deployed in malicious or incompetent ways leading to anything from simple bad policy to outright genocide (and where on that spectrum do you put each of Pol Pot, the Irish Potato Famine or the Holodomor, and why?); or just straight up World War Three, a topic about which I hope to never learn via the medium of personal experience regardless of which weapons are used to fight it.

3D printing

Industrial: At this point, it looks like a material science question — which crystal structures do you get when adding these atoms in this way, and why. For example, 3D printing diamond: we can already grow synthetic diamonds quite easily, and even buy pre-made industrial equipment to do this, so I think it's a question of when we can be bothered to try 3D printing diamond rather than needing any actual scientific or technological breakthroughs.

Biological: The dream is bioprinting to replace lost limbs, substitute transplants, improve gender reassignment surgery, etc.

Unfortunately, every single time I've heard a news story about this, the apparent state of the art has been in exactly the same place — we can do skin and cartilage, so you can always get a new nose or ears, and there's always a demo of something shaped like an internal organ, but for some reason a few years later, nothing has changed.

Therefore, I don't know. Are we at the threshold of a new era of medicine, or not? I can't tell.


Fusion: Wishful thinking, but not because it's eternally 20 years away. Unless we somehow get really lucky and the fusion reactor designs cannot — are actually incapable of, rather than the manufacturers opting not to — run on neutron-emitting chains, the mere ability to make a fusion reactor is also the ability to produce large quantities of neutron radiation on demand; neutron radiation makes it relatively easy to create radioisotopes suitable for fission bombs. Thus, any small projects that seem likely to succeed are likely to be shut down as security risks. ITER isn't small, so it isn't in this category; this distinction matters because you can't hide ITER, while you could easily hide something the size of Helion Energy's 'Polaris' design.

Space-based beamed power: A fundamentally terrible idea on Earth, as any mechanism you can use for getting the power down from orbit can also be used to get the power up from the ground on the other side of the planet first and then down elsewhere. Also, you don't have to just make this safe from the perspective of "can we avoid an industrial accident" (and I don't trust anyone who says they have a system that can't go wrong even in theory, as practice keeps demonstrating new and exciting ways to beat theory), you also have to prove safety to the standard of would you allow a potentially hostile government, that refuses to let you inspect it, launch this into an orbit such that it might, potentially, get pointed at something you care about? Even if you're sure the radio waves can't hurt you, do you really want to gamble that they've not stuck some Hubble-sized optics on it with a huge laser? Now imagine you're in charge of Russia, who have anti-satelite weapons and shown a willingness to shoot stupid targets both in space and on the ground.

Solar: Will keep growing ridiculously fast until it's around half of all demand, that demand may also increase thanks to the rapid increase in supply from PV.

In 2022, the installed capacity was about 1061.7 GW, with a capacity factor of 14% and actual average output of 149.5 GW, and the long-term compounding growth rate is 22%/year. Current global electrical demand is about 2000 GW, so at a minimum I expect growth to remain until 1000 GW average output, which should happen around 2032 (± 1 year). Non-electrical power use is an additional 18,000 GW, and there are a lot of people who don't currently have any significant electricity supply, so there is plenty of opportunity to go past this number. On the other hand, wind is a good competitor…

Wind: Will also keep growing ridiculously fast, just not quite as fast as solar.

In 2022, the installed capacity was about 898.86 GW, less than PV, but the capacity factor was 27% which means it actually supplied more useful power overall (an average of 239.55 GW compared to PV's 149.5 GW). However the growth rate is also somewhat slower than PV, being only 7.3% in that year — the compound rate is slightly better over longer periods, but is less predictable, I think 15% is about right, which results in wind having a 4x growth in a decade, whereas PV gets 4x growth in about 6 years.

Oil, gasoline: Still in significant use for heating and chemical feedstocks for e.g. plastics, but rapidly disappearing from everything else. Regardless of the current legal/political position, you should expect combustion-based cars to go from "standard" to "weird specialist thing" by 2032, and they might possibly no longer be available for purchase at all. Existing vehicles last for around a decade, so assume fuel stations still provide pumps and not just charging sockets at this point.

Aviation fuel: Will almost certainly remain for long-haul flights. Batteries are better than they seem at first glance because of the engine efficiency (so don't be surprised if light aircraft switch to them), but it's implausible they'll replace jet engines for flights over 2 hours. Changing fuel (hydrogen or methane, perhaps from electrolysis or Sabatier process respectively) is possible on paper, but likely to be a huge deal in retrofitting — I won't say "impossible", but I'd be surprised if old aircraft are updated, and instead expect a rule sometime around 2030-ish that says new aircraft need to use a different fuel.

Self-charging cars: If you cover a car in PV, those cells can supply about 80%-90% of the mean vehicle's requirements. This isn't "all", so there will still be a need for charging stations, and it's also "mean" because while some people only drive 10 miles a day to the shops or to work, other people drive a few hundred every single day as part of their job. The electricity has to come from somewhere regardless of how much you use, and increasing electrification of vehicles puts more demand on the grid, so I suspect that governments will require vehicles to do this in order to reduce strain on the grid, even though it's not a complete solution. As electrification is a major goal and many western countries are already facing a huge bill for grid upgrades, I expect this sooner rather than later, probably some time around the end of this decade.

Global power grid: The idea is quite simple: electrical resistance goes up with the length of a wire, and down by making the wire thicker. If you want a 1Ω loop of aluminum all the way around the equator and back to where you stared, you can just plug numbers into a simple equation to learn it will have a cross section of almost exactly 1 square meter, which has a mass of 1.1e11 kg, which is about 1.7 years of current global production for something that at current spot prices would cost only around $265 billion. This is much cheaper than using energy storage solutions for the night/winter/Dunkelflaute problem of renewables.

(Note: you don't want to put it on the equator, and you don't want it to be a single line: if you're putting 2 TW through it at reasonable voltages, the surface magnetic field is strong enough to be a hazard).

However, in practice this would require an unprecedented level of global cooperation, so while China may well prove the concept by connecting themselves to both Africa and Chile, I'd be very surprised if this simple solution is actually going to change the world any time soon.

Consumer electronics

Nanolithography (e.g. CPU manufacture): Reaches its limits. The current node (which isn't really the feature size any more) is 2 nm process. If the recent rate of change continues, we reach 1 nm/10 Å in 2028, 0.5 nm/5 Å in 2032, and as the lattice constant for silicon is (roughly) 5.4 Å this is as far as resolution goes.

Computer energy efficiency: This is not as constrained as feature size, so it's at least possible that this will continue improving past 2032 — perhaps as far as 2080. Not certain, but possible. If it does continue at current rates, (doubling every 2.6 years), they'll be about x5 today's power efficiency in 2030, about x18 relative to today by 2035, and about x71 compared to today by 2040.

The power constraint, rather than absolute computational capacity or minimum feature size, is the constraint for all mobile devices — not just mobile phones, but also laptops, cars, androids, AR/VR glasses, smart dust, etc.

AR/VR glasses: will get better over time. I'd expect widespread adoption of low-profile headsets like the XREAL AR glasses some time around 2028 ± 2 years, but the sci-fi vision requires true holographic (in the sense of interfering wavefronts) displays — which are possible, but need a lot of computational power at the moment (perhaps we'll make this more efficient, but otherwise it's going to be a wait).

AR/VR contact lenses are currently limited to tech demos; while I won't be surprised to see a launch along the lines of the original Google Glass, I'd expect them to flop like the Google Glass. To become a breakout item, they need both true holographic displays, and also significant improvements in computational efficiency which I don't expect this decade.

Smart watches: Despite all the love they get in the tech world, I don't see how they add value. They feel like fashion pieces to me, so will continue to be popular, but I don't anticipate any new uses beyond what we've already got.

3D TV: No real changes. True holograms are possible, the computational requirements are less of a concern when it's a box on the wall rather than a tiny thing on or next to your eyeball, so this is a question of market demand — which, last time round, was driven mainly by one specific film, so it could happen, don't rule it out, but don't assume it either.

Smart dust: This is a big one! If you've not already heard of it, the idea is very simple: miniaturise a computer until it's the size of a grain of dust, use it for monitoring. I think we've already got all the stuff we need for this, but there's been no reporting for a while now — so my guess is the US military already has it and has a gag order and/or NDA with the companies making it. It's a perfectly reasonable bit of kit for them to want to own, and absolutely a piece of kit they'd want to try to find defences against before someone else uses it against them.

I recon a consumer grade version of this will happen some time between 2026 and 2028; and if you embed it in a transparent lacquer you get smart paint, which will be painfully expensive at first, but a reasonable (if fancy) business/prosumer item some time around 2030 — expect it to replace whiteboards, if they've not already gone away due to very large conventional touch-sensitive displays.

Smart dust also means smart tattoos, which will make it even harder to tell if someone is cheating in an exam; but as they also have the potential to mess with facial recognition and eye witnesses, assume it will be banned for use in tattoos.

Phones: Apple and Google will probably lose their arguments, the stores will be opened up. I don't know how this will turn out, given there are many strong drives pulling in different directions. I'm an iOS developer, so I get to see a short way behind the scenes — US government wants to know all about the use of cryptography in an app. Is the EU going to allow this? I think not, and that they have to force Apple away from the US government because of this. Does the US want to lose this advantage? I think not, so they will try to make sure that Apple's rules de-facto force all developers worldwide to continue to be tied to the de jure reporting requirement.

The phone hardware upgrade cycle will remain flashy big numbers that make no real difference — how many megapixels is your camera and why do you care, you're not a pro photographer — until some new sensors are included. This might be infrared, either near-IR (night vision, easy, cheap) or far-IR (thermal imaging, moderately expensive) or both. It might be an option to use the WiFi as a wall-penetrating radar/life signs detector (a surprisingly easy trick). It might be chemo-sensors such as carbon monoxide — although the importance of carbon monoxide detection will decrease as fossil fuels are phased out, this is just an example to give you an idea for the kind of thing to look for.

Games consoles: we may be nearing the ultimate form, thanks to generative AI. Although we're not quite at the level of real-time GenAI models, we've gone from 90 seconds-per-frame about 18 months ago to 4-5 frames-per-second this month, and these models can be used to generate new content with poses or other content hints from existing images. The result? I suspect that, by 2028, the graphics pipeline of most games will be to first render an extremely crude representation where everything's just a coloured cuboid, then pass this into a generative AI which takes those cuboids as hints and turns them into something genuinely photorealistic (assuming photorealism is the desired art style; substitue for whatever else if not).

Photorealism, long a dream for 3D games, will be available for all games, and basically for free because "reality" has unlimited un-copyrighted training data and there's nobody to argue you can't or shouldn't.


Privacy: I don't see how this can survive much longer, assuming it's even still around today. The more tech we have, the more potential there is for a single individual to cause mayhem. The more mayhem is caused by single individuals, the more demand there will be for the police to put people under surveillance. The tech to put everyone under almost constant surveillance already exists, and is fairly cheap — even ignoring that most of us already purchase several devices which can be used this way, and for which it is reasonable to assume (never mind post-Snowden, just by looking at the Investigatory Powers Act 2016 in the UK) that governments subpoena information from.

The counter to this is: it's not just governments who get to spy on you, blackmailers can do it too. It's not in any government's interest for its population to get blackmailed: it's bad enough when they're local criminals who spend money on the local economy, but we're in a global environment, and international blackmail is even worse from the point of view of governments. I think this combination will force governments to do some combination of perfect enforcement (blackmailers can't find your dirty laundry if the justice system has already named-and-shamed you) and/or radically de-criminalising things — no cute mixed metaphor here, just if your actions aren't seen as bad, them being public is much less of a threat. Less, but…

Social networks: Blackmail is unavoidably tied to social networks, even without legal implications, one can be blackmailed by shame. Thus the concept which the American right wing calls "cancel culture", which the American left calls "blacklisting", and which I've also heard called "social exclusion" or "social marginalisation" in less-politicalised discussion. This is a very easy thing for us humans to do, and to do for things which are not at all illegal — indeed, for certain categories we have laws banning the exclusion/marginalisation itself, and yet it still happens.

It is currently unclear how much power social networks actually have: I think, on the basis of how bad the adverts they show me are, that they are much worse than they tell themselves or their customers. If social networks are perceived as powerful, they will be regulated in order to limit this kind of dynamic; if social networks are perceived as powerless, they will not be regulated in this way.

Tags: 3D Printing, AI, Artificial intelligence, Bioprinting, Elon Musk, Futurology, Notes, Opinion, Robotics, SpaceX, Technocracy, Technological Singularity, Technology, biology, black swan, brain scans, civilisation, drones, economics, forecast, fusion reactors, future of humanity, futurism, prediction, robots, rockets, space, wild speculation

Categories: Futurology, Technology

© Ben Wheatley — Licence: Attribution-NonCommercial-NoDerivs 4.0 International