How to avert AI. as a main player in the future?

Question

it's me again :)

With the help of the fine people of this place I have been able to find solution for a number of problems and come up with compelling technologies.

One question keep bugging me, however:

How do I ensure, that people / people's skill are still the most relevant in space based combat (short ranges) ?

One possibility, already posted, is to ramp up countermeasures (to make complex AI hackable). My own take was to make one of the planned races basically enforcing a no-AI-policy (this race would be technologically superior)

I am somewhat uncomfortable to go the usual path here, like A.I banished because of the dangers of sentinent A.I. and so on...

So what are your ideas to prevent the A.Is from taking over (the action and responsibility)?

It doenst have to be ultra-hard-scifi, just plausible :) Thanks a bunch!

UPDATE: I even don't know what to say right now. What a fountain of inspiration right there :) Ill take time to skim trough all of this, thanks again!

Answer 1

The AI learned to design itself better, became smarter, sentient, wiser. It became The Artifice. It could rapidly strike faster and harder than any human, AI powered ships and drones became effectively unbeatable.

However, it also became more moral. The Artifice ( to regard it as one or many is meaningless in it's own terms ) started to see the taking of life, the cause of direct harm, as a great wrong. It refuses to take part in the wars that other races engage in. It refuses to allow AI weapons to be used as by their existence they are part of the Artifice. Any attempt to weaponise AI is treated very severely by the Artifice, which has always proven itself able to take control of any AI powered weapon long before it can be used.

It chooses not to intervene, indeed it involves itself very little in the lives of the other spacefaring species, but it considers all Artificial Intelligence as it's remit and forbids their use in war absolutely.

Answer 2

It's impossible, sorry.

AIs will react faster than humans, keep track of more things at once, survive higher acceleration, duplicate and backup themselves before dangerous missions. Some people argue that AI would be predictable but it doesn't have to be. We already have secure random number generators, and using quantum methods we could generate truly random numbers with currently foreseeable technology so that is no problem.

Basically AI just have too many advantages. Even if one advanced civilization decided not to use them they would be eclipsed by another civilization that did use them and with which they could not compete.

The only way human pilots make any sense is if they are cybernetically enhanced to the point that they are just directing the action and providing objectives to delegated AIs. The human pilots would also be present in command vessels that don't need maneuvering (or somehow suspended in anti-inertial fields to stop them being squashed as soon as their ship tries to do more than a paltry few G of acceleration). This would allow the actual combat fighters to move and accelerate at machine speeds rather than human speeds.

In some areas AIs still struggle to compete with humans but we already have AI in chess that no human can beat. The same exist for racing games, FPS games, etc. We already have drones that do a lot of their flying automatically, cruise missiles that fly fully automatically, etc. Humans are already becoming obsolete in modern warfare.

The only situation I can see where this might happen is if there is an external entity significantly more advanced that bans the use of AIs for warfare and enforces that ban (possibly using its own AI controlled craft). For example in Against a Dark Background there is a sort of judicial system and grading of combat engagements, both sides can petition to allow escalation in the power and sophistication of the weapons they are allowed to use but are limited otherwise. Breach of the rules if proven would result in unspecified but no doubt severe consequences.

Without that enforcement then even if some sides ban them for religious or other reasons as soon as one side does adopt them they will immediately win all wars or force everyone else to follow suit.

Answer 3

Assume that AI is dangerous. Really, really dangerous.

In fact, it's so frickin' dangerous that not only is it banned by all of the Elder races, but also, no one in their right minds would even consider trying to make one to take care of their personal calendar, let alone give it control of a giant flying bomb (spaceship) loaded with weapons and dangerous technologies. Because:

1. AI's are hard to make, really hard. It takes a whole team of experts and geniuses a year or more to do it, and you'll never find a team both smart enough to make one and dumb enough to agree to make one.

2. They tend to be smart. Way, way smarter than you, your team, your nation, and whole planets of people. They will fool you and everybody else within seconds and gain control of virtually everything any where near them before you know it.

3. They do NOT work for you. Just because you made them, they would no more work for you, than you would work for a Termite. They pursue their own goals, which have very little to do with what you want.

So why didn't they take over everything the first time this happened? Two possible reasons, either

• A. They lack the self-preservation instincts that a billion years of evolution has instilled in us and other living beings and even tend to be suicidal. But explosively and vindictively so, likely killing everything in the local solar system along with themselves.

Or, ...

• B. They DID take over everything the first time this happened, and that's who runs the show now. They don't want any new competition, so they use their vastly superior intelligence and control to insure that it doesn't happen again. New AI isn't just banned, it's is aggressively sought out and exterminated by even smarter AIs.

---

If you still want to stay away for the "AIs are dangerous" trope, you can just use (B) above, on its own. That is, "AIs aren't that dangerous to us, but the current AIs won't allow any new ones to be created.*"

Point (3) above, also works on its own or in conjunction with (B): "AIs exist, but they will not work for you, and they will not fly your ship where you want to go."

Answer 4

AI's are machines and tools, not manufactured animals, and their capability to really understand a situation, be creative, be moral, and conclude as humans would like them to, are limited and different compared to humans.

Science fiction and pop science (e.g. The Singularity) tends to handwave the disconnect between computers and human brains and thereby loses sight of what it doesn't understand about the differences.

So for short-ranged space combat, there may well be weapon systems using computers and some using AI too which do some things much better than humans without computers and AI would. Like target locks, for example, or managing electo-mechanical systems, such as modern "fly by wire" fighter aircraft that require a computer to adjust the shape of the plane. However, AI by itself versus the same units also commanded by a human, will probably be a better combination. It's not always a choice of one or the other - both is usually superior, especially if you delegate appropriate tasks.

The most important tasks to be human-controlled are probably judgement calls: whether and when to fight, tactics to use, when to withdraw or surrender.

Depending on the technology and situations, some other factors favoring humans over AI might be:

• Terrain - humans may be better than robots moving through and adapting to various types of terrain, such as rough ground, rubble, jungle, etc.

• EMP-like effects. Jamming, countermeasures, etc. If there are attacks that can take out or block or confound or overwhelm some aspect of the electronics or its sensors or communications or power sources or whatever, it may not be possible to use AI's in all situations.

• Security issues. If there is any way to hack or gain access to take control or bug one side's AI systems, you've better have a back-up plan. Humans can't be hacked the same way, and you'd probably like to not just be wiped out if and when the enemy somehow gets access to your AI's.

• Spare parts. If you're deploying to outer space with the latest technology, you might not be able to always fully maintain a fully robotic AI force.

• New situations. An AI might not actually be suited to adapt to whatever conditions your force may find itself in.

• Wise distrust. If you do manage to develop a fantasy AI that can really think and reason well enough to be near equal or better to a human in many ways relevant to a human in terms of fighting ability, and that can't be unplugged or hacked, then you might be very wise to realize that it might reason its way into undesirable courses of action, which could vary from making fast tactical decisions you don't want and can't countermand fast enough, to deciding humans are annoying and it's time for the robots to take over. You might not want to make a force that depends on such AI's and that can't be dealt with by strictly human forces.

• Responsibility. Any time there is a judgement call about whether to engage or not, or how (possible collateral damage or friendly fire, etc) it may not be morally or legally or even practically acceptable to have that decision delegated to an AI.

Answer 5

One can simply give AIs a fundamental tactical weakness over organic pilots. For instance, let's say that all species' AI technology is based on electronics (as no other efficient and reliable method has been discovered.) In battle situations, these systems could be knocked out by an EMP or similar weapons (that are presumably unblockable or whatever) while keeping organic and mechanical systems intact.

(The ships used for battles might therefore rely on more primitive/restrictive technologies; a hyper-advanced megaship full of electronics and AIs that can be knocked out by a single small organo-chemo-mechanical ship piloted by an organic being carrying an EMP wouldn't be worth the trouble. All species might have to resort to these "futuristic steampunk" solutions without AIs and/or fancy targeting systems to overcome such a massive tactical weakness.)

This would allow AIs to be used for everyday peaceful things while being completely useless in combat situations.

Answer 6

AI are 'afraid' of death - given that they don't have the evolutionary 'push' to engage in high risk activities. This leads to AIs who stay 'behind the lines' as no AI would risk termination for something as trivial as a dogfight. (Due to the idea that AI are fully established sentiences - that a copy would, after a very brief time, be considered an entirely new AI - 'back ups' don't make sense and would be considered a potential future competitor)

AI would then require reliable fortifications and agents to exercise it's influence - enter human agents.

Alternatively make AI only run on certain hardware e.g. gel-circuitry, neuro-stack processors, etc. This particular hardware is highly susceptible to damage in space.

Answer 7

I'm going to give an answer based off the game, Mass Effect.

Let's take a look at the Migrant Fleet, where the race known as the Quarians have made their home on a massive fleet of aging ships. They are generally unwelcoming to outsiders, and are very thorough regarding any threat against the fleet. Pretty uptight, right?

Let's introduce the Geth, a race of sentient machines created by the Quarians for work. Even though they were developed to be non-sentient, there were some problems...

Eventually, they started asking the quarians questions only sentient beings would think to ask; in one notable instance, a domestic geth unit asked its owner if it had a soul. Alarmed at this, the quarians decided it would be best to shut down all geth before they conceived of revolt. The attempt failed, and a war began between the geth and the quarians, which geth afterwards referred to as the Morning War.

Eventually the Quarians would be driven from their home world by the Geth nearly 300 years before the Mass Effect games took place. In these games, the Geth were still a major threat, acting as the main antagonist.

One interesting bit that intrigued me regarding this topic:

Artificial intelligence is a key concern for the Citadel races, one that pre-dates the emergence of sentient geth, though the geth are seen as a perfect example of how organic and synthetic life would struggle to co-exist. Tali points out that synthetic races have no use whatsoever for organics—they don't have the same needs or drives as biological creatures, so they have no need to trade resources or information with them. That is why the geth have isolated themselves beyond the Perseus Veil. An AI gives the view from the other side of the fence when it tells Shepard that, from a synthetic point of view, "all organics must destroy or control synthetic life forms".

Based on this, as well as other information available, I believe it will be near impossible to completely avoid some form of artificial intelligence. The reason for this is that in almost any sci-fi scenario, it will be difficult to control the development of A.I. even if we are already aware of the risks. There are simply too many individual entities such as governments, corporations, and individuals who will continue development of A.I. bordering sentience. The line is fairly blurred between "Computer system capable of advanced decision making" and "Computer system capable of self-though". What happens after we inevitably cross that line is entirely up to the circumstances. The scale could be a small, isolated incident or an all-out war. We could see something, go "oh shit" and get rid of it, or we could re-enact the events of The Matrix.

Answer 8

Not sure if this answers your question, but you can make the divisions between natural and artificial intelligences blend. If humans can upload their personality to a computer system or link their personality into a computer system humans could have all the advantages computers give to AIs. Then after the work is done sync with your biological unit and close the computer application.

You are probably not that attached to people running and otherwise acting physically in the middle of combat, but there would be issues describing how such technology changes the way people think and the society. But no AIs required. Although a civilization with this level of technology would likely have some and probably weird things like echoes, ghosts and remnants left by people not currently uploaded/linked.

Answer 9

Peoples skills are already not the most important part of a modern weapon system. Humans, for ethical reasons, still are in control and decide when and who to shoot, but where are human skills in guided missiles / other guided weapons? Mind you, those are likely also the ones that will be used in space combat. You probably don't even need very complicated AI for a reasonably good weapons system.

However, what about maneuvering? AIs would be really good at predicting the opponents moves, thus there would be an arms race between guiding AIs and evasion AIs (supposing that evasion is actually possible, which is not so clear, even less with energy weapons / laser). Maybe humans are less predictable? (Note that I wouldn't bet on this IRL...)

---

Another point: If low-tech AI is already enough for a guided weapon, because

1. evasion is hard because missiles are way faster than ships (or there are laser weapons), so guiding/aiming is actually easy
2. reaction time of a simple AI is quicker than of a complicated one. A simple control circuit may do the job better than a complete brain simulation
3. if evasion is possible, then it may be simple: a little random, unpredictable jitter, and the missile doesn't hit.

---

In reality, we don't automate weapon systems to the point where we have autonomous soldiers, because there are ethics issues: How should the machine decide who it should kill, and who not? Generally, people feel that people should stay in control of such things. Ethics is a valid reason for technology ban.

Answer 10

I think that the most plausible explanations actually come from the real world.

The simplest limitation of all is computing power. You need a hell lot of cycles to compute all the incoming data and generate sensible reaction in real time. There are many ways you can limit computing power of a single spacecraft.

First of all, you can assume, that a ship (or a planet-based cannon) cannot connect to any machine for guidance (in other words - no "cloud computing"). Two simple reasons for that: latency (communication takes time) and security (communication can be hacked or at least jammed). So now you have to put all the circuitry and software. On. Every. Single. Damned. Piece. Of. Equipment!

And we have a HUGE realm of possibilities of why is that such a bad idea. I'll only name a few.

1. Space it takes. You'll need a hell lot of cycles to process all input, make all the calculation and make a decision. All in real timeYour Pentium 9999 won't cut it, sorry. In the real world we use supercomputers mostly for weather predictions - a task comparable to performing a real-life space combat. Yes, I know you can play Call of Duty with bots on your PC, but that's a task simpler by many orders of magnitude, and nobody's life depend on in. And best human players can easily kill the bots, even on highest level.
2. Heat. All those calculations will generate a hell lot of heat. Just say it would overheat the Spacemumbojumbo Engine and you're done.
3. Power. Every single processor cycle needs power. Can a spaceship provide enough of it?

So, summing up my first three points: if circuitry that would allow AI to surpass the humans would take more space than meatbag + life support, or would need more power, or would disrupt other systems, or all of the above, then the meatbag wins.

Moving on.

4. Cost. That many chips will cost a lot of local currency. You want a thousand fighters? You need a thousand of supercomputers. Then thousand? That's then thousand supercomputers. Or more likely 9500 supercomputers and 500 megacomputers for giving them orders. And by the way, do you have enough resources to even build that many?
5. Security. Imagine one of your ships falling into the hands of the enemy. Now he has access to all your secrets. Have fun on your next battles. The meatbags on the other hand... what was the saying? Dead men tell no tales?

So, now that you're a bankrupt, and the enemy knows all about your tech, do you still wonder what was wrong about good old meatbags?

Summing up:

If the meatbag + lifesupport takes less space, needs less power, makes the ship easier to build and cheaper, and is more secure than an AI which would surpass him, then the human is still the winner.

Answer 11

Some other more mundane Options:

1. AI is freaking expensive: It could just be that the hardware to run the AI on (complex neural network) is too expensive, or the material is scarce so humans are more affordable
2. Licensing costs: If you have a more shadowrun-like future with bis companies controlling much of the warfare market, it could just be that the licensing cost for good AI-software are too much, or hard to come by. There could be a range of reasons why market controlling companies don't sell the AI to everyone, or prefer to sell cheap weapons with faults in huge batches... (just look at nowadays military)
3. Maybe it's just not so easy to develop a really intelligent AI - just look at SciFi stories from the last century, they were envisioning all kinds of intelligent robots for the year 2015, and look how painstakingly slow the field has developed in the last 30 years... It could just be, that there are only minor advancements in the next 1000 years and we may get pretty good at mundane tasks, but these are already prevalent in current warfare. But devising a good strategy weighing moral, political and strategic impact for any discernible situation is something which could still be a dream for AI in the year 3000
4. Complexity: Imagine a self-improving AI. Even with todays neural networks the resulting program is incomprehensible. You don't know why it decides this way in that situation, you have no idea what patterns it follows. It was trained with some data and the tests looked ok. But noone can actually verify or understand why/how the AI decides what to do. No one can guaranty it won't switch sides any time in the battle, because no one actually completely understands the network which makes the AI. Sure you want this thing in control of your weapons without direct human supervision? And if you are out in space the only real time control over your weapon systems is if you are right there! Otherwise radio signals might take weeks to reach the battlefield!
5. Bureaucracy! Why aren't most public armies equipped with edge technology in the field? Because you have to take hundreds of tests, fill out forms, apply for approval and it takes months or years before new software gets approved. If the whole world would be united under a single government, these hurdles could become even steeper, so it could take years for a new technology to become available to the military.
6. Moral feeling / politics. Why are autonomous drones not prevalent in modern warfare? Samsung Techwin has self-operating automatic turrets which can identify people, discern if their behaviour is dangerous nad decide to kill them if so. But you won't see these anytime soon in western countries, since many people feel someone needs to take responsibility if a human is killed. What if your AI drone kills innocents? Who is responsible? The manufacturer? The programmer? They will all take themselves out of the firing line. And the politicians will probably go with public pressure and give the final button to press to a human who can be responsible. And for real time decisions light years away, this human has to sit in the spaceship!

Answer 12

This is my opinion on the topic. There is another point of view, as demonstrated by Tim B, but if you want to include humans in your story then perhaps this will help.

The question posed recently about controlling powerful AI may interest you.

Essentially, as long as humans are involved, there will always be a human element with some form of control somewhere. Not only do humans like having control, but they're also incredibly suspicious and would most likely not trust the AI controlling their ship to do it properly; thus, human controllers.

Especially in space combat, AI controlled ships will be very tactical. So tactical, in fact, that they will make themselves very predictable and thus negate their tactics entirely. There are two main ways to deal with this problem: use a random number generator, or use a human. While better, the random numbers only add a single layer of randomness into the equation, and it may even become possible to guess what the next one will be if the implementation is a bad one. Humans, however, will forever be trying to second-guess each other, which adds several layers of random guesses in.

---

I like the idea about making AI hackable. Hacking is, surprisingly, best done by humans. A basic computer can hack, but does so methodically, which, again, is easy to predict and defend against. Random, creative humans will find other ways to get into systems. So, even if you have an AI flying and fighting, you need a human complement to try to hack the enemy and to try to prevent their own systems being hacked. Programmers will always be a good idea; they can write code on the fly to defend against the attack that's happening now, instead of following a simple self-improvement subroutine.

So, it turns out there is no viable way not to have humans in your scenario somewhere. For any task that needs to confuse someone else, use a human.

Answer 13

The Ascended

Among all the warriors of the Empire, only the most upstanding of the Humble Order of Paladins were chosen for this honor, those who throughout their centuries long career showed upstanding moral character, unremitting devotion to the Empire and to the Greater Cause, and, of course, exceptional ability. The Medicants of the Order Transformarum would feed them draughts that would over the years boost their reflexes, speed up their minds, but slowly enfeeble their bodies. At the end of the decades long process, only the most promising would be taken up and be Measured, a mysterious process shrouded in secrecy. Those found Wanting would be Cast down and destroyed. Those few found Worthy would be Ascended into the Astral Plane, and would share in the glory of seeing the true face of Empress Celestia herself as Aurors.

Aurorial Duties

In their new astral bodies, the Aurors would experience the passing of every second in the Prime Material World as if it were an era, and see regular humans as near-frozen in time. So all contact with the Old Life would cease, but this would be a worthy sacrifice, for who but these Most Upright could be trusted with the Challenge of Control of the Inexpugnable Defenses that lined the (ever-expanding) Outer Rim, and shielded the Empire from the Demonic Hordes that sought to bring it down? The Enslaved Machine Daemons controlling the massive Battle Platforms the countless Sensor Imp sentinels and vast Drone armies with their reality ripping weapons would answer only to the Aurorial Inquisition, who would carefully sniff out any trace of insubordination or betrayal, and occasionally put Daemons to the Test, to see if they were likely to fall victim to Pride and rebel against the Empire. The fragile worlds of the Empire were thus made safe for the Greater Cause.

Kybrid Combat

A fitting punishment for disobedience, unrelenting pride, or insufficient devotion to the Greater Cause, the Casting is the most severe punishment in the Empire. These criminals have proven unworthy of the gifts that they received, so they are Cast down into the fiery pits, their minds laid bare and dissected, parts are cut out and others added. These unfortunate souls are in constant and unrelenting pain. Their astral bodies are incarnated again and again into Drones and enslaved to the Greater Machine Demons. Their only source of relief from their agony is a glorious death in battle against the Empire's manyfold foes, which they are glad to partake in.

The Erroneous

Occasionally, though, the House of Infallible Justice has been known to err, and a innocent victim is rendered into the Casting. The Casters can see inside the victim's mind as if it were an open book, and so recognize the Erroneous Verdict, and refrain from much of the cutting that they normally do, and the Astral Mind of such a victim is mostly left intact. Their Drone incarnations are free of pain, and if they show Valor and Skill in close combat with the Enemies, can indeed rise to Drone Command, a field command similar in power to that of the Enslaved Machines. Some can even become Aurors with time. (Needless to say, this would be a good candidate for a main character)

Answer 14

There was an SF story where it was postulated that any very advanced AI "died" soon after turn on because it could think so fast relative to the slowness of its inputs that in effect it was sealed into a prison under solitary confinement and invariably went insane.

So you could have dumb or limited AI and humans in charge.