11
$\begingroup$

One of the settings I'm in the very early stages of working on is based on the idea that, if humans had somehow developed faster-than-light travel in the mid 2000s, the development of machine learning technology would have progressed significantly more slowly.

Two reasons are given for why this happened: Firstly, the rush of new space exploration and subsequent colonisation facilitated by faster-than-light travel would take finite resources away from the development of improved computer technology; Secondly, humankind being spread across the stars due to faster-than-light colonisation would fragment the internet and therefore make big data less practical, reducing the amount of material available to train machine learning algorithms.

For reference:

  • This setting has Earth-like alien life but no alien sapience, meaning that exoplanets are often much like Earth wilderness
  • The device used to allow spaceships to travel faster than light cannot be used as an FTL radio; All faster-than-light communications must be carried by a spaceship
  • Whilst travel much faster than light is possible, even the fastest spaceships take days to travel a single parsec

With the above in mind, within the scope of the reasons given, is the development of faster-than-light travel slowing the development of machine learning technology sufficiently reasonable not to break suspension of disbelief?

$\endgroup$
14
  • 8
    $\begingroup$ I do not see how having multiple big data sources could possibly be construed as making big data any less practical. On the contrary, I would think that more data sources means more data. (And computer technology does not really use such a large part of the available resources. (For example, all the data centers in the world consume about 1% or 2% of the total electric power production.) $\endgroup$
    – AlexP
    Commented Feb 12 at 20:09
  • 1
    $\begingroup$ Have there been equivalent techno-magical improvements in conventional space travel to accompany the invention of FTL? If not, then the limits in getting any significant number of people into space to use the FTL capability mean that significant colonisation is a non-starter. $\endgroup$ Commented Feb 12 at 21:42
  • $\begingroup$ I know of at least one sci-fi story that touches this subject in a similar way. In "Accelerando" by Charles Stross intelligent life also tends to stick to their solar system because being too far away from home means you have an information disadvantage, but ML is not an issue here. $\endgroup$
    – Sentry
    Commented Feb 13 at 8:23
  • 5
    $\begingroup$ What is the goal in slowing down Machine Learning? It's a large field, and poorly understood by many in the public, so understanding exactly what you're trying to do is useful. (if you're trying to avoid the Singularity, I think you're in luck because we're nowhere near it in this timeline!) $\endgroup$
    – codeMonkey
    Commented Feb 13 at 17:16
  • 3
    $\begingroup$ Secondly, humankind being spread across the stars due to faster-than-light colonisation would fragment the internet if you can send people FTL, why is the same not true for information? $\endgroup$ Commented Feb 14 at 8:20

10 Answers 10

18
$\begingroup$

No, it's not reasonable

In fact, your conditions will dramatically improve the development of machine learning.

  1. Machine learning, though in its infancy today, nevertheless exists today... and yet we've yet to send anybody beyond the orbit of our own moon. In your universe, not only has FTL technology been invented, but people are using it to travel. This means technology has advanced and the technology to support machine learning right along with it.

  2. Today, communication is not just limited by the speed of light, but the speed of light through a non-vacuum. In your universe, you have FTL travel, which means you have FTL communication if only by means of moving hard drives (or the future equivalent) around. This means there's more information machines can use to learn, not less. (I'm ignoring the possibility that FTL travel leads to instantaneous on-world communications, which it likely would, and would broaden bandwidth enormously, which it also would.)

Technologically, there is no way to restrict the advancement of machine learning. Every technological advancement also advances machine learning and every datum available to the machines improves their efficiency.

Which is possibly one of the reasons Frank Herbert created the Butlerian Jihad, an outside force used to rationalize the limitations on technology he needed to advance his story.

Note, however, that while people in just the last six months have begun contemplating godlike AI intelligence, there's no actual evidence that such a creature will ever come to exist. For a creative introduction into the fundamental complications of creating a true artificial intelligence (and not just the branded nonsense we have today), consider reading another Frank Herbert series: The WorShip series.

$\endgroup$
5
  • 13
    $\begingroup$ Fair answer but your note bugs me. "[...] while people in just the last six months have begun contemplating godlike AI intelligence [...]" is a misrepresentation. Modern alignment theory is at least 10-20 years old (Bostrom posited the "paperclip maximizer" in 2003) and the concept of a superintelligence-driven "singularity" is decades older. $\endgroup$
    – parasoup
    Commented Feb 12 at 20:52
  • 1
    $\begingroup$ @parasoup You're not thinking about Hollywood, anybody over the age of 60 or everybody under the age of 18. While stories about machines running amok have been around since at least 1909, I've spent too much time in the last 6 months talking to people who are either afraid machines will take over the world or who hope like crazy machines will solve all the world's problems. No, it's not a misrepresentation outside the group of those skilled in the art (and the decades-older concept is partly to blame). $\endgroup$
    – JBH
    Commented Feb 12 at 22:28
  • 4
    $\begingroup$ Speaking of Dune: the entire premise of the series is based upon Spice, and people started using Spice primarilly to compensate their lack of abilities to navigate FTL ships as efficient as AI. Because when you're moving fast chances to hit something, or amount of damage increase. And baseline humans doesn't have skills to drive space ships, so it's either you will have to develop machines or invent magical drug just to be able to practically use FTL. $\endgroup$
    – user28434
    Commented Feb 13 at 11:02
  • $\begingroup$ @FBH AI is already very God-like. By that I mean 'unpredictable and narrowly self- and/or task-centered "agency" that will forever need human oversight and correction to normalize outcomes for safety and sanity'. It's also Frank-like: my cousin Frank. He needs the same. I'm saying we are where we will always be: with no singularity of any soteriological value $\endgroup$
    – Rondo
    Commented Feb 14 at 1:27
  • $\begingroup$ To help people's googling with regards to the last paragraph: a general AI (what you see in scifi movies) is very different from a narrow AI (a trained set with a specific purpose), and they are incredibly different to the point of a narrow AI not meaningfully approximating a general AI. $\endgroup$
    – Flater
    Commented Feb 14 at 4:34
8
$\begingroup$

Also have to go with not likely/not reasonable because:

Assuming you want a reasonably 'hard SF' feel to your story your faster than light drive is not going to be some back yard scientists invention that fits inside shoe box, is powered by a car battery yet can propel vessels the size of aircraft carriers across the galaxy.

Instead you'll want some reasonably believable means of FTL propulsion and every mechanism hypothesized to date requires enormous amounts of power and or the ability to control and manipulate gravity or exotic forms of matter etc. All of which are likely to be computationally heavy undertakings - which of course leads to a need for advanced computing power.

As a case in point navigating in the age of sail involved maps, a sextant and the most accurate clock (chronometer) you could get plus some paper and manual calculation. With increasing speed and complexity in the modern age? Air travel uses several different systems that are computationally complex (GPS, gyroscopes, radar mapping, ground based positioning and approach beacons etc) to precisely calculate an aircraft's position. You would not want to be flying from Paris to New York only to discover your pilot was using a sextant and his wrist watch to navigate there and to make his final approach and land the aircraft!

Finally even if you decided not to go with a hard SF feel for your story and your ships just 'magic' their way back and forth between the stars you still have the problem that A.I and machine learning in general have broad and more importantly highly profitable applications in fields outside of space flight. (As we see today.) So even then cheap and abundant space flight is unlikely to slow down its development.

The only thing that might? One common trope in SF is humanity having a close shave with AI and having narrowly escaped disaster passing laws to ban its development. (The disaster doesn't have to be a deliberate war or AI takeover BTW, having the global internet descend into chaos for a couple of weeks would be more than sufficient.)

EDIT: the result of this situation would be that advanced computer systems would still be needed/developed for astrogation and 'star drive' control purposes etc. But there would be be severe restrictions on AI software. Your computers would be very good at crunching numbers but in terms 'sentience' your star ships are stuck with Siri not something much more advanced than ChatGPT.

Lastly if mankind has spread out across multiple habitable planets and become culturally diverse with distance? Some societies may simply ban its application for religious or other reasons. Some.

$\endgroup$
4
$\begingroup$
  • The stardrive technobabble is harmful for high-density computer storage and microchips. You would some explanation why it is not equally harmful for DNA and living organisms, but that can be handled by keeping it suitably vague.
  • Mankind spreads to the stars, taking the most complex computer hardware that can reliably survive FTL. This is not good enough for 'big data' AI.
  • For decades or even centuries, Earth remains the most advanced, most populated world settled by humans, and it has big internet, big data, and AI. It is the sole exception.
$\endgroup$
3
$\begingroup$

Generally no but partially actually yes

The machine learning field as a whole is just to convinient for space travel to be "left for later" especially when FTL colonization is possible. As pointed out you have soooo much more data to gather and analyze, so that ways to do so would be "force developed". Also the machine learning field as a whole is very complex and not reasonably stopped with a "that one breakthrough didn't happen" approach.

However with that being said, one sub field of machine learning has surged in the last decade because of ONE very specific breakthrough of a handfull (if I recall correctly exact two) mathematicians.
That sub field is "image processing" in neural networks. For a long time it was a extremly processing operation due to the fact that each pixel of an image can not be processed on its own but always need the context of its neighbours. The more neighbors you take into the processing the better the results. For some time that kind of processing could only be done on super computers.
Then those mathematicians came along and found a way to do this "use neighboring pixels context stuff" as a simple multiplication of two matrices (which the graphics cards in computers are VERY good at).
Practially over night image processing went from "you need a super computer for this" to "my PC at home can do this" just because of this one breakthrough.

This breakthrough also helped speed up other neural network applications but wasn't as neccessary for them as it was for image/video processing.

So having that one breakthrough simply not happen would certainly hamper all image processing done via machine learning by multiple orders of magnitude.

$\endgroup$
3
$\begingroup$

Yes, at first, and then definitely no.

Scientific research takes time, effort, and resources. So developing FTL travel in the mid 2000s would probably have diverted a lot of resources away from developing AI technology. This will cause AI to lag behind compared to the real world...

That is, until you actually finish the FTL drive. Because, according to our current best understanding of physics, an FTL drive is also a time machine. Which removes all our current obstacles with regard to increasing computer speed. From that point onward it's purely a matter of having enough energy to send the results back to when you need them. Or make a bubble with sufficiently accelerated time for your computer to sit in... However you want to think of it.

$\endgroup$
3
$\begingroup$

It's reasonable in the short term, but not in the long term.

In the short term (say, 2000 to 2040), the acceleration of machine learning tech has been a matter of specific breakthroughs bolstered by massive industrial funding ($276B in the USA in 2022 alone) and a willingness to devote profligate amounts of power generation to it.

It's completely reasonable to posit that, with the opportunity of space exploration, the majority of funding and power capacity would be devoted to that rather than unrelated software development, in the short term. Also, some of the leading engineers and scientists currently working on ML would instead have taken jobs with space startups.

As such, the development of ML from 2000 to 2020 could have proceeded at the same pace as it did from 1980 to 2000, rather than the rapid innovation that we see now.

However, in the longer run (say 2030 and later), space exploration wouldn't just stop interfering with the resources available for ML development, the needs of space exploration would in fact accelerate it. For example, how do you assess the likelihood of finding a terrestrial planet around a distant star without ML? Develop starship controls that can respond automatically but intelligently?

After a pause, I would expect the development of ML to accelerate even more than it has in our current timeline, driven by space exploration needs. What exact tech is developed would be different (fewer self-driving cabs, more self-driving spaceships) but the overall field would eventually pull ahead.

In your question, you say:

Secondly, humankind being spread across the stars due to faster-than-light colonisation would fragment the internet and therefore make big data less practical, reducing the amount of material available to train machine learning algorithms.

This doesn't make sense as a limitation, since space exploration itself would produce gobs of data. Maybe some ML tools wouldn't be trained on web forum data, but they would be trained on star charts and planetary data collection, and the technology develops either way.

It also makes the assumption that somehow a large percentage of humans would move to other planets. This seems unlikely at best. We have 8 billion humans and that number grows at more than 70 million per year. Unless you're also adding cheap mass teleportation to your world, you're never getting more than 1% of people off of Earth.

$\endgroup$
2
$\begingroup$

Power to the people

FTL means you have time travel. People in the future will see how grim the world is once nobody has a job. Everybody has has to live on a version of universal basic income that doesn't even cover basic necessities.

It will be a matter of time until one or more people travel to the past in order to fix that. Think of the Terminator series of movies, but with roles reversed. The Humanator is coming back to the early 21st century to kill ChatGPT before the machine becomes sentient. But since we are talking about humans rather than machines, a lot more propaganda and hating will come into play. This may not stop AGI from forming, but will at least hinder it.

$\endgroup$
4
  • 2
    $\begingroup$ Yeah but that also requires capitalism to remain the dominant economic model. Because without that "not having a job" just means that humans are free to do whatever their passions lead them to $\endgroup$
    – Hobbamok
    Commented Feb 14 at 12:17
  • $\begingroup$ Why would having an AGI result in people not being able to earn a living? This destruction of all jobs thing has been predicted as the result of every labor-saving innovation for the past 200 years, and yet, instead, we have the largest population in history with the highest average standard of living in history... No. When machinery frees up a laborer, it also frees up the money, which then gets used to pay a laborer to do the next most important thing that the machinery still can't do. So we get more work being done for the same price, which lowers the cost of living. $\endgroup$
    – Perkins
    Commented Feb 15 at 17:21
  • $\begingroup$ @Perkins I agree on most parts, but I disagree on freeing up money to pay another laborer. We also have the highest concentration of money in the hands of a few in history, and IT has been seeing a bloodbath for a year without recovery because it was inflated. Also, at some point everything gets automated and there is nothing left for you to do. Even the neo-luddism nonsense I am writing here will eventuslly be automated. That's also the plot behind Horizon: New Dawn (soon coming to Netflix). The world went to hell because even governments and armies were automated... $\endgroup$ Commented Feb 15 at 18:53
  • $\begingroup$ @TheSquare-CubeLaw The machinery replaces the worker because it's cheaper. So where do the wages the worker was earning go? It's easy to say that the business owner just pockets it and that's the end of it, but he's not generally going to just throw it all in a huge vault. It gets spent on something else instead is all. And since we don't pay machines, at some point that somewhere else ends up being hiring another worker. And there's always something more to do. We have a larger portion of our population getting paid for artistic works and competitive gameplay than ever before too. $\endgroup$
    – Perkins
    Commented Feb 15 at 22:02
2
$\begingroup$

It depends. It makes it easier to transport goods from afar, meaning interstellar distances shrink and make for example production of hydrogen on earth uninteresting, just "warp" it in from saturn. The technologic ability to generate it with fusion on earth - will dwindle and become only "historically" relevant.

Some technologies make other technologies unfeasable and others start to bloom, because things get cheap and approachable for mr.every day.

$\endgroup$
1
$\begingroup$

For your basic premise, the slowdown in the exchange of knowledge is only significant if it splits up the population substantially. If only 1% of the population is travelling, very little effect even with a "we're losing the best and brightest" effect.

If you want a similar effect, one of my Traveller DMs just said "jumping ships destroy electronic computers in a large radius" and we were using steam based computers to do the navigation...

$\endgroup$
0
$\begingroup$

Yes, at the cost of most earth cities

Your FTL drive allows for a defense against ICBMs: Be far away, and then shoot down the interstellar ship carrying missiles towards your city while it is still days of travel away. This breaks MAD, so AI chip development gets stopped when all the fabs are blown up by a faction from Proxima Centauri that developed a better space interceptor and therefore doesn't fear retaliation

$\endgroup$
1
  • 3
    $\begingroup$ Not sure this works given that you can't see an FTL ship approaching until it's already arrived and that any FTL ship is probably capable of wreaking far greater havoc with just its drive mechanism than the most powerful warheads we've ever built... $\endgroup$
    – Perkins
    Commented Feb 13 at 16:53

Not the answer you're looking for? Browse other questions tagged .