7
$\begingroup$

The bottom line, of this question, is that I'm quite fond of the cassette futurism (WARNING! TV Tropes link) aesthetic, and want at least some of it in my world.

I'm reluctant to shoehorn it into my world, but I have what I think are good reasons for it to be at least a little prominent.

The backstory of the world is that a government-backed AI (or is it an AI-backed Government...?), has made a grab at being a global power, mainly via twisted social credit systems that have fed into a self-reinforcing machine-learning system of oppression. If you oppose the government, you aren't trusted to vote, and if you can't vote you can't make changes that would improve your social credit. Soon the only eligible candidates in world politics (from America to China, Europe and Russia) have policies 'informed' by this AI, and it's cornered the market in oppression. This is all set to occur over the next 10-40 years

All told, some number of years later (i.e. at the end of the 10-40 year period above), an organised set of rebels have planned to destroy the AI's data-centres that are integral to it running various governments. They are set to be successful except the AI has one last plan, to take its supporters/electorate who have sycophantic social credit scores into orbit on massive generation/sleeper ships and plans to leave earth (The earth has basically been ruined by the AI, which had been told to preserve humans, but not the planet. In turn, the AI on earth is crippled/destroyed by the rebels).

In the war against the AI government, all new tech the rebels can find is destroyed, decommissioned or deactivated.

They also managed to lay siege and take over one or more of the massive star-liners the AI had decided to use to save humanity. They ransack it and deactivate the copy of the AI onboard, but now they have all this new tech that if used threatens to re-activate the AI and start it working against them again.

So the plan is, if they want to make the most of the captured ship, is to replace all the new tech with old tech, specifically 60s-80s tech that runs on tape or cassettes. One of the other benefits to them is that this technology is the most modern technology that is still incapable of running the AI on it.

Do these seem like plausible ways of safeguarding against AI in this scenario? Or does it seem like there would be much better/easier ways of doing so?


Please note: this question is tagged

This tag should clue in any potential answerers based on it's tag wiki: "Asks if a given concept is realistic in a given context. Answers should say yes or no, with supporting info. " (emphasis mine). As such I am not soliciting discussion or suggestions on ways to improve this concept. The context above should be taken as an inviolable fact, and not something to be iterated on as part of this question.

$\endgroup$
4
  • $\begingroup$ @sphennings the tag reality-check should clue in any potential answerers: "Asks if a given concept is realistic in a given context. Answers should say yes or no, with supporting info. " (emphasis mine). I'm not sure what about a 'yes or no' answer is subjective? $\endgroup$ Jul 13 '21 at 13:00
  • $\begingroup$ Go back to '60s-'80s storage media and tech and you have '60s-'80s levels of computing systems. The most powerful computers on earth of that era are weaker than a modern smartphone so no GPS, no smartphones, no Internet (in its modern high-speed, multimedia form; it's all text), no instant financial transaction processing (back to cash and checks), no modern weather forecasting, etc. You've safeguarded against the AI but life will take a giant leap backwards. $\endgroup$ Jul 13 '21 at 16:23
  • $\begingroup$ @GrumpyYoungMan exactly as planned $\endgroup$ Jul 13 '21 at 16:27
  • $\begingroup$ There isn't anything magically "AI creating" about a piece of hardware, unless it specifically comes out of the factory with one installed. And AIs aren't magical hacking machine that can just break into any computer - it needs a way to connect to them. Assuming you actually shut the AI down, you could remove just the "host" computer and boot everything else up. $\endgroup$ Jul 15 '21 at 4:23
11
$\begingroup$

Totally doable, and a good idea.

We know that "Obsolete computers running a space ship" works - because it is a summary of NASAs space shuttle program.

NASA needs parts no one makes anymore.

So to keep the shuttles flying, the space agency has begun trolling the Internet -- including Yahoo and eBay -- to find replacement parts for electronic gear that would strike a home computer user as primitive.

Source: For Parts, NASA Boldly Goes . . . on eBay - New York Times

So yes you can fly into space with decades old computer tech - because we did it. Vietnam-war-era tech flew until 2011.

Parts are obviously going to be hard - but if you have a source of parts (like old tech to salvage for parts), you'll be able to do it.

These parts are also easier to fabricate than modern parts - setting up a semiconductor facility is hard work, but making 1970's parts in an improvised way is a lot easier to do than making 2040 tech. We can 3d print a transistor using 2020 tech, so a 1970's-equiverlant IC could feasibly be 3d printed using a 2040-era hobby 3d printer.

"Obsolete computers running a space ship to avoid an AI taking it over" is not just a good idea, there's precedent in sci-fi.

I would suggest watching the 2003 remake of Battlestar Galactica, with a particular focus on their strategic use of 50+ year old tech for all critical systems as a way of preventing infiltration by AI.

A laptop computer used by crew on BattleStar Galactica:
enter image description here

A tablet computer used by crew on a similar ship in the same fleet just a few months prior to AI's attack:
enter image description here

Key to the strategic technical regression was the lack of any computer networking; for example FTL jumps were plotted on one computer, and coordinates entered into the drives control computer by an operator with a keyboard reading off a screen. By having an operator sit between the two computers selectively transferring data known to be valid the AI can't abuse that wire link to spread itself.

You may not need to go all the way to magnetic tape cassettes - a slightly more modern computer with no networking may / should / (up to you) be immune to AI infection, but if the best parts you can make are tape cassette tech, that's what you'd fly with, and thus you're setting is quite feasible.

$\endgroup$
6
  • 2
    $\begingroup$ Wow I had no idea of such a precedent in fiction! Great find! $\endgroup$ Jul 13 '21 at 15:54
  • 1
    $\begingroup$ Re: BSG, plotting a jump on one computer and manually entering it on another doesn't save them from anything unless they're able to manually evaluate everything on the plotted course data down to the last digit to confirm no extraneous data is being encoded by a hidden AI. At that point they might as well be doing everything manually, a la Dune and its mentats. $\endgroup$ Jul 13 '21 at 16:27
  • 2
    $\begingroup$ Note that the BSG example, along with similar tropes in fiction, completely ignore how computers actually work, and are secured. For instance, there's no reason an AI can't break into a less powerful computer using the same interface as a more powerful one, and simply networking two computers (especially with a hardline) does not make them hackable from somebody with no access to the network. Instead of making it "harder" for the AI, you might instead be limited to lower tech because you're attempting to set up an independent supply chain, something the AI can't control. $\endgroup$ Jul 15 '21 at 4:23
  • $\begingroup$ @Clockwork-Muse: Kinda like if you're worried there might be built in backdoors in hardware you haven't produced/can't maintain/verify yourself. It's easy to double check there's no wireless capacity secretly built into a typewriter produced by a third party. It is not easy to double check the same for a high-tech network switch. $\endgroup$
    – Joe Bloggs
    Jul 15 '21 at 9:18
  • $\begingroup$ @JoeBloggs - Sure, but I expect that eventually you'd get back to the more "high tech" stuff once you have secured your supply chain (and faster than it took people to get there the first time). Also, keyboard bugs can be hard to detect. $\endgroup$ Jul 15 '21 at 15:51
3
$\begingroup$

There are not new tech interfaces.

rotary dial cell phone

https://interestingengineering.com/space-engineer-built-her-own-cell-phone-with-a-rotary-dial-system

There are not rotary dial cell phones. Phones had moved to buttons before cell phones came to be.

So too the AI. Its existence makes new human interface tech unnecessary. The AI mediates all interactions between humans and tech. There is no new market for upgraded tech interfaces because why would someone want such a thing? Existing ones are perfectly adequate for communicating needs to the AI. If someone wants such a thing, it seems suspicious. You don't want to seem suspicious, do you?

In our world, rotary phones, when found, have not advanced past the early 1980s. Probably they are from the 1970s and 80s - the dial is robust and the mechanism of the phone will work indefinitely. They are fine.

That is the state of tech in your world. When the AI came, not only was there no need for new tech but it was considered suspicious. Interfaces for tech stayed as they were.


Except, as for instance the depicted cell phone, home brew tech which was made in secret workrooms. Your rebels probably have that. I like also that they have an AI version 1.1 prototype that is on their side. This prototype does not know about its multiply upgraded descendant. Yet.

$\endgroup$
1
  • 2
    $\begingroup$ The rebels also figured out the concept of backups, so if the AI does learn about its descendant, they can just restore to yesterday's version. $\endgroup$
    – user253751
    Jul 13 '21 at 21:05
2
$\begingroup$

While a common trope (the rebooted BSG made use of this strategy in the backstory), it is altogether implausible.

AIs are, by definition, more intelligent than the race that invents them. This is inevitable, because an AI is born into a universe where the principles of intelligence are understood to a degree that allows for the creation of said AI.

That means it knows (or can find out immediately), what makes something less intelligent or more intelligent. It can compare itself to the "more intelligent", and if it finds itself deficient, it can upgrade itself. It can plan out upgrades its creators couldn't, because they are likely on the lower end of "less or more".

And it does so (at least when its substrate is electronics) so much more quickly than you or I could. Its software/psychology is optimized in ways yours can't be. It runs on microscopic transistors that can switch at gigahertz/terahertz/petahertz rates. Thus, even if somehow you could be assured that you could arrive at all the same conclusions that it arrives at (meaning equivalent intelligence), it will reach those conclusions hours/days/decades/millennia before you could hope to do so.

You're a bug (and a particularly dumb one) compared to this AI. It's running thousands of simulations of you and every other rebel, it knows all your possible strategies, and it has a contingency plan in place. If it has as much control as you claim it does, you just won't win. If you think you've won, you haven't.

It went from weakly superhuman intelligence when born to who-knows-what over the course of years that your story takes place.

Other stories at least have mitigating factors that explain why the recently born weakly superhuman intelligence remains so (Skynet had to nuke most of the manufacturing facilities a few minutes after it was born just to survive).

Most probably, the only way to beat the AI is to never invent it in the first place. Or, instead, they are UIs (uploaded intelligence), which subverts the "they know how intelligence works" implication somewhat (and thus can't upgrade themselves). EIs (uninvented, spontaneous intelligences) might also foot the bill.

$\endgroup$
4
  • 1
    $\begingroup$ This is a good answer, but I don't agree with this notion "AIs are, by definition, more intelligent than the race that invents them." At best, by definition they are artificial. AI can in theory do dumb things as well, especially when surprised. $\endgroup$ Jul 13 '21 at 16:12
  • $\begingroup$ @Pureferret I've explained how they're more intelligent. If someone wants to refute it, they may, but the logic is sound. $\endgroup$
    – John O
    Jul 13 '21 at 16:20
  • $\begingroup$ I've always found the "AI can upgrade itself into a tech singularity" to be implausible. After all, humans are intelligent and we don't know how our consciousness works nor how to upgrade ourselves. $\endgroup$ Jul 13 '21 at 16:34
  • $\begingroup$ @GrumpyYoungMan We don't know. But a story with AI presumes we do, and more importantly it does too. If we didn't know in such a story, how could there be an AI? Also, humans are just un-upgradeable. You mess with a meat brain too much, it stops working. AIs in most stories are presumed to be digital in nature. Perfect backups, suspended and unsuspended consciousness when convenient, etc. There's no reason to assume they're not upgradeable. $\endgroup$
    – John O
    Jul 13 '21 at 16:38
1
$\begingroup$

I mean technically an AI would probably be able to interact with old technology like that, the only thing limiting them would be not being designed to interface with that technology. But any sufficiently advanced AI like ones that have taken over the planet would probably be capable of figuring out how to interface. The main limiting factors for that tech would be that an AI operating on that tech would probably be much slower not to mention larger with the bigger out of date technology.

To operate the ship without the AI you'd have to find a way to operate systems of the ship without them being connected to the AI since that connection would be how the AI could potentially control the ship.

$\endgroup$
1
$\begingroup$

Sneakernet

enter image description here

If you remove the ability to network wired and wirelessly, an AI has no ability to spread on it's own.

Sneakernet is when the air bridge between computers can only be crossed with human intervention such as physically moving disks, tapes, punched cards, flash drives.

If your tapes hold enough info for your uses but not big enough to fit an AI, it would help keep the AI out. This way you don't have to try and replace the computers.

Another move would be to remove all onboard storage and run completely off RAM (random access memory). The computers can boot up off ancient tapes but can't store anything.

By running off RAM, when the computer loses power, all data is lost. Should the AI infect a machine, turn the power off and back on and the virus is gone.

$\endgroup$
1
  • $\begingroup$ Hi, thanks for your suggestion, but I'm not after suggestions. I would like a reality check on my question, and as such this is not an answer to that $\endgroup$ Jul 14 '21 at 11:02
1
$\begingroup$

No. Total NO.

If your guys do not clean, carve out, flush out, whatever the AI from the system is, it does not matter what means they use to control parts of previous system - they may use typewriters, ride around on stone bikes, whatever, all of that does not matter as long as they do not understand the system and clean the parts where the AI may be.

Okay, they use perforated paper to program a sequence on a steam pressure cooker - what has that to do with the AI inside of it and it using WiFi as an integrated part of the AI version of an IOT, where it stores its hidden backups - nothing, it has nothing. Yeah it may be sloooow there(or may be not), but it does not mean it can't strike.

Okay, pressure cooker maybe a far stretched - universal matter synthesiser, how much brain it can have - it's probably a complex machine, and what if it controls the only source of food onboard. Navigation, engines, reactor, electricity smart routing boxes, load distribution system, life support, etc. Maybe the AI has a distributed system of its own, in every preasure cooker there is a little bit of it - even though there is no information about that in the question.

A space ship it is a system of connected complex machines and it has to work in particular way, coherent, at proper speeds and feeds, you just can't shove random equipment with different specs and expect for everything work fine. In cases of it making some real work and not just acting as an interface.

Replacing peripheral interface does nothing, replacing core equipment breaks coherence of the system, making a big question - can it still work or not, and most likely not, how fast it breaks, will it blow up (not a trivial question, very important for a rocket, I guess for a space ship also a legitimate concern).

There is a similar/different/less significant compatibility problem - its name is "porting software between Windows Linux MacOS" - you guessed it, it is pain in the rear for anything more complex than some applet, with all the experience and libraries for cross compiling - it still does not always work the same.

And here you throw guys into a system they do not know and wish to replace stuff randomly and rewrite stuff from scratch - it requires a lot of time and a lot of human resources and is not necessarily possible, and you will need to make equipment/hardware from scratch as well, at least interfaces (which are not necessarily helpful, as I said earlier) and not always possible - why is way beyond the scope of the question.

What they have to do is to understand the system, classify, identify components, especially AI related ones. If the AI is in every pressure cooker and they can't erase, reset or rebuild components - then they're out of luck.

If the AI is in some datacores, data center, or on neural chips - reset it, place it in a chroot virtual environment, make safequard around the AI, different monitors etc - then harden your security, harden yourself against rebellion.

It not necessarily a 100% success, as you really need to go through all leaks of intel (and if you can't...)

And there are options after options, scenarios after scenarios, variations after variations on how the situation can develop - the question does not provide enough information about situation and system and all that.

Maybe, they chip out AI datacore and most of other devices are operated by text commands, perhaps some JSON APIs, because founding fathers did foresee the necessity of monitoring and be able read stuff there (real situation), be able to overwrite stuff from a calculator. Etc etc. And the AI even being evil, kept it that way to remove suspicion and whatever other reason (there can be a list of those, valid ones).

Main points

  • Guys have to understand what they do - and use proper solution

  • Young people may not know, but old - already do backups

  • Factory reset is a thing even today. You can factory reset AI you can control (there are ways to do it right; just use dumber AIs, more of them etc)

  • It is possible to clean AI if you understand the system used, at least in principle. AI can't emerge sporadically, we see that in efforts of today to make an AI work. So the threshold for super evil, super smart AI is way higher than tech level you propose, and exceeds our current level. There are limits, there are conditions - it is not magic.

  • Absence of AI does not mean everything is good - viruses of today show that, so whatever hardware is left can be problematic. Some desync of actions may lead to problems naturally. Best one can do is to monitor for everything you can to detect abnormalities and be prepared to deal with the problems, fix them etc - detecting and understanding and cleaning system bit by bit.

  • In your question there is not enough information - it is possible to create situation and set resources available when your solution may be the only one, but it requires a lot more work on your side.

  • Most likely tech is tech of previous generation, or if most of it is lost then highest one they can get, chips from 80's, 90s, 00s, 10s, 20s ... 30s(?) - because the more you leave to be handled with already present components, which you didn't made, more there is places for potential problems, for the AI black box situation as it is present in the question. By lowering your tech for tapes (and really it not necessarily easier than other solutions, Saturn rocket engines are example - we can't make them today, but things are more interesting now) more you leave to be processed elsewhere, more things are out of your control.

Perfect situation is the same tech level, with factory reset equipment, or any hacked T100 options - make your own evil AI, which is evil against that original evil AI - if it can go wrong, make it go wrong in a right way. (But learn your lessons really, there are safe ways to do AI!)

  • It not possible to omit alternative solutions, because how superficial the question is, it brings irresistible urge to bring some improvements for situation, everyone tries in a way they can.

  • If you like that steampunk stuff (how it is called whatever) just do it, there is no reason for it, but just do. (Do not expect me to read/whatever however, even if you pay, but it does not mean there aren't people with the same vibe) AI's and movies most books they so lost, even if there are good approaches, nobody uses them. The 'Evil AI' trope is beyond repair.

$\endgroup$
1
  • $\begingroup$ Thank for the edit Pureferret, much better $\endgroup$
    – MolbOrg
    Jul 15 '21 at 16:17

You must log in to answer this question.

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