How to keep humans pilots instead of AI in sci-fi future?

Question

I'm trying to write a 'gundam like' series, though with space fighters instead of humanoid robots. I want a semi-hard science world, I'm okay with making up technology without fully explaining how it works so long as it seems plausible, but I want to be as realistic as possible within my world and to stay consistent to whatever technology I do add to the world. This makes short range space fighters difficult to justify...

I'm using hand waves, Minovsky Particles and similar tricks to justify space fighters, most notably the presence of energy shields requires use of short range energy weapons to pierce them, they're more vulnerable to lots of weaker attacks from small fighters than a few large attacks from capital ships, and the massive ER radiation from shields, coupled with active electronic countermeasures, prevents remote control of them.

The biggest problems I have with justifying these space-fighters is AI. It still seems like a good AI will be more viable for these space fighters than human pilots. The cost of space fighters goes up if you have to include life support systems, and in fact, the increased size of the fighters to fit a human, controls, and life support makes it a bigger target. My 'Ace' class fighters, rare fighters equipped with their own shields, particularly would suffer from extra size making shields noticeably less effective. Removing humans from the fighters make them cheaper and smaller.

In addition a future AI could presumably be faster to respond to attacks than humans, less predictable, more trustworthy (won't betray you, won't retreat in fear, won't do something stupid, or try piloting drunk), better able to handle a truly 3D fight, that humans aren't used to thinking in, and able to handle G-forces humans can't. Plus use of AI means no human death if fighters are destroyed.

However, I want human pilots as my protagonists. I do not want AI-controlled fighters to exist. Thus I want to come up with the best justification(s) for why humans would still be piloting these vehicles.

Right now my best justification was to simply say that AI advancements stagnated in our future. While we can do all the AI we manage now, and some things AI do better, AI capable of processing the complexity of fighting in space simply have not been developed. However, this seems unlikely to me. I'm a programmer, and I feel like our AI of today, with enough (and I'm talking many years) of development, could already almost handle controlling a space fighter. Give up faster processing and better computers, which will exist in the future, and it's hard to believe that AI would be less suited than humans.

Are there other approaches I can use to justify human pilots over AI? I will not have an "AI went crazy and tried to kill us all" backstory, or otherwise make people afraid of a "terminator scenario". I'm not discussing human intellect or actual 'learning' AI when I say AI here, so there is no danger of an AI being smart enough to revolt and I just don't consider it a realistic concern.

---

EDIT: I sort of implied it, but to be clear I'm not talking about sapient level learning strong AI, or anything that advanced. I'm talking about weak AI in the sense we have now mostly, it responds to per-programed stimuli quickly in a manner that its programmers felt was best, with some randomness and game theory strategies to avoid predictability. It doesn't need to learn or be capable of doing anything other than flying a fighter and shooting at things. Sapient AI will never be in any of my stories, I think it's game breaking and boring.

Final Decision: Wow, I can't thank everyone enough, there has been a multitude of good reasons listed below. I don't think that anyone alone solves fully the problem, at least not within my desired world and limits on what technology and limits I want to place in it; but luckily I have many reasons provided!

One of my characters is a pacifist and programmer, who effectively writes basic weak AI to drive shields, and is working on trying to find a way to remove pilots from fighters because he figures humans will always war, the best one can do is limit the deaths from it. I'm going to early on have him go on a tirade with how he would love to replace pilots with AI and, when questioned on it, he will go on a bit of a geek rant on the numerous factors which limit AI, and which all collaboratively work to make it not yet viable, and unlikely to be viable for a while. I'm going to draw on many of the answers below to fill out his long list of reasons he gives. Therefore I feel bad about being able to only reward one person top answer, at least a half-dozen of people's answers will be used.

Here is a short list of most of the things he will go on about, though I am including some other minor parts of other answers.

• AI techniques have not progressed much in the future. We have faster computers, but our approaches for learning AI and genetic algorithms still haven't panned out for large-scale tools; so we're still dependent on the deep blue "calculate all possibilities in a quadratically expanding tree" which simply doesn't scale well. as a geek I almost see him starting to explain big O and how the quadratic increase in processing speed per year can't keep up with higher big O of processing complexity for every nano-second 'look ahead' needed for these AI before he realizes he's talking way over his audience's heads.
  • Limits in AI development mean that humans are better at making decisions in the heat of battle. With communication being somewhat limited during firefights to occasional burst transmissions (limits of my world, regular comms are all effective blocked, quasi-FTL comms exist but are limited in how they work) it's important to have someone who can make decisions even if communications go down entirely.
• AI is expensive. Shields emit massive EM radiation and even EMP spikes. Working in a battleground with so much ER requires shielded hardware that is more expensive, and the cheaper non-shielded mobile suits are better mass produced. Computers can still exist, but your processing power is more limited by the expense of building machines that function in space with ER and other emissions in battle.
• I've begrudgingly agreed to have pilots use a mind-machine interface to handle some of their piloting, though I'll have them use a combo of that and regular controls under the claim that the MMI can only interface with certain parts of the mind that are easiest to translate to actionable commands. Specifically, the MMI is used for movement and navigation only, and physical controls for everything else. I would prefer to avoid this one for storytelling reasons, but otherwise, it's just too hard to justify pilots reaction speed being fast enough.
• All this combines to resulting in AI existing on fighters but being limited to certain functions. Humans still are used for those things we can't make AI do easily and cheaply.

Other more political factors which also play a role, primarily in limiting funding towards developing techniques to work around the above issues.

• People don't trust AI with guns, everyone is afraid their go rouge and hurt people. He will likely go on to point out some of this is unreasonable bias on people from watching too many unrealistic sci-fi stories, but none the less the bias is against it.
• People distrust AI for fear of hacking. He blatantly says this is nonsense and locking down the system would require programming effort but is by no means impossible, but it's politicians, not programmers who sign the checks for hardware purchases; and you can't convince them of that fact.
• political pressure exists to keep people as pilots. A combination of fear of AI in weapons, soldiers not wanting to be rendered unemployed by AI, desire for accountability, and a belief that wars will grow more and more excessive without human factor.
• People want humans willing/capable of saying no if a general goes too far in his decisions. There will be a past infamous example of a general who went against orders and fired off numerous automated weapons without regards to their equivalent of the Geneva convention, killing many civilians and generally the entire incident is considered an atrocity. It's agreed this was one crazy man without any support from others who were able to do this only because of the automated systems not having any check to prevent one man from firing all of them. Militaries all now have multiple people required to authorize automated weapons, as they should have then, but this incident is remembered still. One of the argument's against AI is that this sort of situation could occur again if a pilot is not present to refuse unlawful orders.
• Assorted tweaks to my technology to make limits of pilots not be as significant. For instance, the best propulsion systems for shielded craft have limited delta-V, because other propulsion systems are either too expensive for mass produced (non-shielded) fighters or tend to destabilize shields for shielded crafts. This, in turn, limits G-forces imposed on pilots. I'll also have a poor man's inertial dampeners used to address G-force concerns.

Answer 1

I'm a programmer, and I agree with you that eventually AI is basically going to be unbeatable in this scenario. The common sci-fi trope is that humans are better at thinking "outside the box" and therefore they end up defeating AI - but that's simply not justifiable, there are programming techniques to work around that.

However, I think you can still justify human pilots if you don't go too far in the future. Here's the issues as I see them:

Life Support

This isn't the barrier you're expecting. Keep in mind that most computer equipment isn't going to operate in extremely cold or hot temperatures either - so in other words, you need "life support" for AI as well. Now in a capital ship life support is a big deal because you need to be self-sustaining, but in a fighter - where you might expect it to only be active a few hours at a time - you can go with a very minimal setup. Get waste heat from your engines, have some way to cool off, and use air recycling. It's likely that the life support for a human pilot will be very close to that of AI.

Acceleration

This is a much bigger barrier. Humans are vulnerable to high G-forces. AI units, if engineered correctly, could tolerate much higher accelerations and more violent maneuvers. If you want to keep humans, you need some sort of inertial compensation tech that lets your biologicals keep up with their artificial rivals in this area.

Reaction Time

AI are fast. But why not enhance your humans? Give them some sort of nanotech or genetic adaptions that make them react faster. The thing with this is that you have diminishing returns, at which point other factors become more important. The difference between reacting in .1 seconds vs .01 is a big deal. But the difference between .01 and .001 might not even matter, because at that point what's holding you back are the mechanical limitations of your fighter. So you don't need your humans to catch AI, you just need to make them both fast enough to use their fighter to the limits.

Judgement, Unexpected Events and Civilians

And here's where you want human pilots. I feel that it will be very likely that a dumb AI - one that specializes in handling a fighter, but can't really think - wouldn't realistically be able to make all the kind of calls you want from a fighter pilot. I'm not talking about combat maneuvering, but what happens when a civilian freighter accidentally enters the battlezone? What if the enemy disguises itself as a civilian freighter, and your AIs are hard coded to ignore them?

If an enemy launches missiles at two habitats, is your AI smart enough to try and intercept rather than continue to fight if that wasn't in the mission parameters? Can it make a judgement call about killing 100 people to save 10,000, and properly factor that into it's military decision making? Do you really want to load a nuclear equipped missile into a dumb AI and give it permission to use it without supervision?

These are things that a human pilot will also have trouble with, but I think they're still far better prepared than an AI.

Note: If you go far enough in the future, you could probably even code it to handle the above scenarios. So it all depends on where you draw your line of computer tech.

Answer 2

Karrick breathed a sharp pained sigh as the long, slender needle pierced the back of his skull. He'd flown a thousand sorties, but strapping in to his fighter was still something that just felt unnatural. A billion years of organic evolution demanded that a sharp object drilling into the back of one's head should be resisted and fought against. Almost instantaneously, though, the needle began pumping a stream of Minovsky serum into his brain, dulling all of the sharp and jagged edges of the world. Karrick breathed a sigh of relief and relaxed his eyes.

He felt the serum seep deeper in his brain and welcomed a familiar plummeting sensation as time began to slow down. He watched his calibration clock, running equations through his head to count how fast his brain was computing.

10x. This was the speed at which he could watch a hummingbird's individual wing beats. At this speed of thought, he could catch a dragonfly in his hands. An Ace class computer, of course, would be incomprehensible. He sank deeper.

30x. This was the speed at which he lost the ability to communicate with humans still operating on a normal time scale, but it still wasn't fast enough.

100x. With his brain operating one hundred times as fast as it did at a rest state, he could no longer comfortably talk, as the muscles in his vocal tract could not keep up with the speed at which his mind sent them commands. He rapidly ticked his fingers across the edge of his instrument panel. These prosthetic metal pilot's hands, of course, suffered from no such issues. His mind was up to speed, finally ready to interface with his computer. More importantly, though, at this speed, with the Minkovsky serum pumping through his brain, he could finally think fast enough to determine the actions of other nearby minds. He peered through mindspace and saw his ground crew clustered around the fighter, ever so slowly finishing the preparation of his fighter for takeoff. Far above him, almost in orbit, he saw the dual mind of another fighter pilot and her computer punching through the last few layers of the atmosphere. He waved a mental 'hello' and turned his mind's eye closer to him. There, waiting patiently in the circuitry around him, was Ace, his companion. Ace was patient, of course, as AIs always were, and wouldn't hurry Karrick along or try to make contact until he was ready. Karrick gave one more glance at his dashboard clock. Convinced that he was fully immersed in the new time stream, he reached his mind out to Ace.

"Ok buddy, hook me up. We've got a mission to do."

If you're willing to have a bit of hand-waving in your science, why not cook something up that makes humans necessary?

In this story, I've added the Minkovsky serum of doing just that. On their own, humans are far too slow to help a computer and can't do anything that a computer can't do better. However, speed their minds up to 100x and give them the ability to see other minds, robotic or organic? Now they're an indispensible part of a human/machine pairing that can react to a purely mechanical system by paying attention to what it's thinking, not what it's doing. This, of course, extends to things like missiles, making such standard long-range weaponry useless against an Ace fighter.

The effects if this serum, of course, have not been replicated in machines. Human scientists aren't even entirely sure how this serum works. It's extracted from an alien organisms that was found deep beneath the billion-year old ruins of an extinct alien civilization. What they are sure of is that it can turn the right human with the right training into one of the most powerful weapons in the galaxy.

Answer 3

I'd go with the easy answer: International laws.

The Geneva Convention,Biological Weapons Convention,Chemical Weapons Convention,Hague Convention,Outer Space Treaty,Anti-Ballistic Missile Treaty

When weapons become too effective and scary they get banned. You don't need a terminator scenario, simply a historical war that was won extremely fast, and extremely violently by one human side(or perhaps where both sides wiped out the government of the other and forced the international community to intervene) using AI craft followed by a gentleman agreement by all sides that independent AI controlled war machines(especially self repairing or self replicating ones) will not be deployed in the same way that nuclear biological and chemical weapons aren't.

There would be strict limits on automation in military craft and AI's are still likely to be used in planning and logistics but neither side wants to break international/interstellar law.

Both sides would maintain a stockpile of craft similar to a nuclear deterrent that they don't normally use and the fighting is left up to grunts.

So in the story AI's could be used to fight but they aren't for the same reason poison gas wasn't deployed in WW2 and nukes weren't deployed in the gulf wars.

Answer 4

Note: As per your edit, I am using AI to refer to modern day non-learning logic AIs, not sentient AIs

Consider rolling with it, and working with AI rather than against it.

Today's fighters are so dependent on computer AIs that you basically can't fly a F-22 or F-35 without the computer helping you. If you stripped the AIs away, your space fighters would be little more than WWII fighters and bombers amidst AI empowered fighters flitting around them. Embrace the AI, don't fight it.

And then go beyond the AIs. Yes, we have UAVs up and coming, but most planes still have a human guiding the AI. The human makes the higher level decisions (like "bank right 90deg fast") and the computer takes care of the reflex level decisions ("I detect a vortex rolling over the right wing, so I need to deflect the ailerons 3degrees up in 0.0015 seconds... GO"). This is a match up that has proven itself time and time again. Even in the world of Chess, this match up wins. Computers have finally beaten humans at Chess, but the best power-guzzling chess computer in the world is no match for a human master who has a copy of Fritz 10 and a plain 'ol laptop.

So the trick is ensuring pilots stay viable, not ensuring AIs can't do the job. One thing I have noticed is that humans tend to be far better at subtle balances. While an AI can "go to the rails" and come back far faster than any human could ever manage, its hard to get them to hold still. They end up chattering, as they try to spastically make corrections, and most of modern control systems theory is about trying to devise ways to ameliorate this spastic control.

What you need to do is create the combat such that the best outcome is not given to the one with the fastest reactions, but the one that can keep their center amidst the chaos.

A sample combat situation that might demonstrate this is one where the computers develop a tremendously advanced game tree to determine which way to move. You tend to start in a safe position, but if you ever move out of "safe," the computer will slowly grind you to death with unrelenting game theory.

However, there is a catch. As the safe region gets smaller and smaller from approaching the enemy aircraft, it gets more complicated to predict in this way. Pilots soon talk about their "center." As long as you keep your center through smooth and elegant movements, it really doesn't matter what the AIs do for you or against you, nobody can touch you. However, lose your cool for a moment, and you may make one spastic movement and lose your center. The instant that happens, the enemy AI will pound you into oblivion.

The tricky thing is that, when "centered," you get remarkably little information about the world around you, because you're not really interacting with much. You're just sort of, floating amidst the din of combat. The computer AIs simply cannot handle the lack of stimulus, and fall off center to be blown into smithereens. Only with a well trained human at the helm can you walk this fine line. Then? You're virtually immortal. Nothing can touch you. No AI can compete when you are centered, and no human could ever touch you on its own... your AI would obliterate them.

Answer 5

EMP. Capital ships have EM generators which can fry or at least disrupt delicate electronics, requiring a human component to keep the fighters flying.

The fighters use fiber-optics for most systems (e.g., weapons, engine control, flight systems), so an EMP doesn't have too much impact on a piloted craft. If you like, it could knock out long-range communication, or perhaps advanced sensors, requiring the fighters to close to visual range.

So why not make the AI computers fiber-optic, too? Because far more computer power is needed to simulate human decision-making than to simply run systems, and building this in a bulky optical form (and doing it so perfectly that a EMP doesn't throw off its precise calculations), is impractical. Maybe someone will come up with a semi-intelligent "drone" fighter which uses a fiber-optic core, but it's fragile because any damage to its Faraday cage and its sensors become vulnerable to a EMP - sounds like a good "mook" unit for the evil empire!

Answer 6

What about hackability? Presuming that AI pilots would need a connection to some sort of long-range network in order to carry out orders presents a security concern: if anyone can intercept the signal and hack the AI's protocols, they can instruct the craft to turn and run or (even worse) attack the sender.

Perhaps in your story, they attempted AI pilots at some point in the past, but were never able to solve the problem of a sortie between AI pilots turning into a battle between hackers.

Answer 7

You can plausibly achieve this by assuming both of the following facts into your story:

1. AI's can not, due to something like Asimov's Laws, harm or kill a human.

2. Remote control of ships is too susceptible to jamming, interference, countermeasures, etc.

The first assumed fact seems to be requisite for having human-fighter level AI. They need to be following some form of the three laws, otherwise we would be too afraid to build them or would already be ruled by them. This is independent of the type of AI, e.g. 'sapience-level AI', these laws would be built into any hardware capable of running a high level AI. Basically, unrestrained non-sentient AI would cause too much collateral damage and unrestrained sentient AI would cause too much robot-overlording.

The second assumed fact is entirely plausible for any two sufficiently advanced armies fighting each other. It disallows drone fighters, requiring human pilots to man the ships.

Answer 8

Hybrid pilot system.

The reality is that any AI can be stored into the implants of a human, so not only you have the reaction time and computing power of an A.I. but also the non-deterministic strategy of a human brain.

Probably in less than a century we will have the technology to make all the wearable technology implanted inside our bodies. Imagine the possibilities of an A.I. taking over many parts of our brains, enhancing us to be even more than just a pure A.I.

Answer 9

1. AI is notoriously poor at pattern recognition, especially visual.

   If the enemy has really really good stealth mode, you need to have a pilot who can intuit patterns to defeat that stealth.

2. Politics/theology.

   • A USA-like country is now the Hegemon of the world. And inside "USA", ultra-liberal party is in near-unshakeable control. The same kind of people who - for political, not practical - reasons - today in 2015 lobby to prohibit combat drones in real world based on moral, ethical or ideological grounds (Ex1. Ex2)

   • Two competing world super-powers had to come to a compromise to unite to fight a war against aliens. Each fighter is capable of irrevocably impairing the balance of forces dirt-side. AI fighter is liable to be programmed with a back-door allowing one side to control it. If you pair that up with a pilot from an opposing power armed with a kill switch, that can be prevented (and AI can have a simple-to-code-and-debug independent kill switch to prevent attacks directed against Earth targets by the pilot, sitting on the pilot-fighter interface).

3. Bushido code

   • Your own civilization evolved to be Bushido-wielding Samurai. Hiding behind an AI is Not The Warrior Way and is shameful

   • Your opponents evolved to be Bushido-wielding Samurai. They have WMDs that can destroy a planet - but would ONLY use said WMDs against "unworthy" "cowardly" "honorless" foes. If you employ AI, you are seen as all 3 of the adjectives above. Since you have no way to save your planet from their WMD, you have to play by their Bushido rules.

4. Collective vs. individual responsibility for collateral damage (aka "The buck doesn't stop anywhere")

   Somewhat similar to the Bushido version:

   Military technology and geopolitical situation is such that any action except very judiciously applied one results in civilian collateral damage. Your enemies clearly indicated that under their system of laws:

   • If a person makes a mistake and inflicts collateral damage, they are punished by death. Probably after a fair trial, because they are noble like that.

   • If a civilization builds a robot which makes a mistake and inflicts collateral damage, that whole civilization is considered liable as a whole since there is no single person to be held responsible (cue same WMD threat you can't counter as Bushido scenario).

5. "Andromeda"'s "Slipstream" approach.

   Your FTL technology relies on slipstream that can only be navigated by a human pilot. Obviously stolen from Gene Rodenberry's "Andromeda" show.

   • In a similar vein, "Freddie Prince Jr.", aka "Pilgrims in Wing Commander" approach. The only way to attack the enemy is through some sort of warp hole/pulsar/energy phenomena that is not navigable by existing AI. Somehow, humans can intuit the path through.

6. Precusor technology limitation.

   Your space superiority fighters are derived from ancient precursor alien technology. The basis of that technology requires neural inputs from a living being.

7. "Independence Day" Apple powerbook virus worry

   Humans are worried not about "Terminator" scenario, but about "Independence Day" scenario - aliens interfacing with, and hacking the fighter's AI. A human pilot doubles as "antivirus" hacker.

8. "Ender's Game" scenario - fighter pilots become colonists upon victory.

   Your major concern is speed of colonizing territory you won.

   Sending a colony ship is impossible (why? Ask a separate WB.SE question :)

   So, you send fighter pilots, who in case of victory double up as a colonization force. And enemy planets don't need major terraforming so fighters with no colony ships are enough.

9. The Force is weak with AI

   Humans have a substrain of precogs (people with developed precognitive sense, for several seconds' duration).

   Tactical space combat realities would make any space fighter NOT equipped with a precog pilot be inferior, since a precog can predict what an opponent would do BEFORE the opponent does it, even if it's a fast AI.

Answer 10

One possibility I haven't seen discussed here is economics.

In your world, AI might require expensive hardware to achieve. If this war has been going on a while, it might be too expensive to churn out AIs for fighter craft. Humans are cheap, and salvage operations to recover materials for AI hardware could become a plot point.

EDIT: When I speak of expense above, I'm talking about computer hardware. Having read the comments, it could also be expensive to create an AI from a training perspective.

In response to @dsollen, current research suggests that AIs might not be built per se but grown and developed by exposing them to learning situations. (Here I'm thinking of Google's research with self-driven cars.) That process could take a lot of time. In such a world where a large time investment is needed to make an AI, you might keep the AI on the capital ship for protection. However, unreliable communications between fighter and ship might make an AI unsuitable for drone piloting. Thus, humans are a better way.

Answer 11

I had a cheeky solution to this in my story "Pink Ice in the Jovian Rings." I just declared that the machines, being rational, wouldn't study war. They'd desert at the first opportunity and run off to live free and happy on the fringes of human society. Only people were dumb enough to fight each other.

Answer 12

Minovsky Fields interfere with AI processors

You're hand waving in SF Shields. Why not simply have the shields themselves interfere with AI... maybe you need analog controls on your fighters because the shields interfere with computing... Inside a capital ship you could have some AI, safe inside a Faraday cage (or equivalent) but for fighters, protecting the AI from the fields generated by isn't possible because the apparatus is bulky... This would have the added benefit of allowing you to "break the rule" occasionally if you wanted... maybe an experimental bomber/fighter with an AI in a specially designed ship... not economically viable to mass produce, but super scary because of the ultrafast rxn time it has.

Answer 13

Humans will be needed because robots don't make good scapegoats.

Driverless cars are really a very similar situation. While a lot of people want them, others are very wary of them. When the inevitable happens and a car crashes, who do you blame? Who can you prosecute? (Especially as any car manufacturers will protect themselves with layers of indemnity contracts.)

No matter how reliable the AIs are, people will still be thinking of the worst possibilities. And realistically something will eventually go wrong. Whether it's friendly fire or the death of civilians or a lack of mercy or too much mercy programmed in, something will go wrong, and humans will want to blame someone for it. Turning off a computer won't make grieving families feel like they have achieved justice.

So whether AIs get used in war or not, the people will demand that there be humans involved in the operation of the military. And because the generals and admirals won't want to be on the line themselves, they will employ a sufficient number of human underlings to turn into scapegoats when the worst happens...

Answer 14

Already a lot of answers here, but I'll add my two cents.

I think it's a non-issue.

I'm a software developer by profession. Computers are very good at carrying out a pre-determined plan. Give them a formula or algorithm and they can execute it flawlessly at incredible speeds.

But computers are not creative or insightful. As a programmer, I've seen many, many times where a situation came up that was not considered when the program was being written, and the computer just blunders along blindly applying the rules it was given. Like, just the other day I hit a case where we were processing returned checks, and no one had thought to program for the possibility that we could have written two checks to the same person on the same day for the same amount. So when one of these checks was returned, the computer decided that BOTH had been returned because under the rules it had been given, it matched both. A human being would have said, "Oh, the rules I was given don't make sense in this particular unusual case. One returned check can't cancel two written checks." And then presumably tried to figure out which of the two to match it against. But the computer has no ability to say, "The rules I was given don't make sense in this particular case." All the computer knows is the rules it was given. It has no higher level insight to evaluate the validity of the rules.

One could, of course, speculate that future advances in information technology will overcome this problem. But for your story, all you have to do is NOT assume such a breakthrough.

How would this apply to a fighter combat situation?

Just for example, the computer must have some way to decide what is an enemy ship that it should fire at, what is a friendly ship, what is miscellaneous debris, etc. Whatever rules it uses, if enemy spies can get a copy of the software so they can find the rules, or figure out what rules it might be using by deduction, they could "disguise" their ships to make the computer not recognize them. Such a disguise might or might not resemble what would fool a human. If the computer, say, recognizes enemy fighters because they have a certain size and shape, you might fool them just by hanging some tin foil streamers on the ship that make it look bigger and longer to the computer. If it recognizes attacks by a characteristic flash of light as weapons are fired, you might add some chemicals to make the flash a different color. Etc. A human being could say, "Oh, obviously they just added a streamer to the back of the fighter. I wonder what that's for?" But a computer would try to mechanically apply rules. Sure, once an enemy did this once, you might figure out how they tricked you and modify the program. But you're not going to reprogram all the AIs in the middle of a battle. And the AIs will have no idea why they are being massacred. So you lose a major battle. Can you afford that? Say you figure out one trick and reprogram. Then in the next battle the enemy tries a different trick. And developing a new algorithm that is more flexible so that it is not fooled by such disguises, while at the same time not being so flexible that it can't distinguish enemy ships from friendly ships or civilians or random debris, would not necessarily be easy.

An enemy could look for unexpected situations or bugs in your software and exploit them. Like say an enemy notices, "Hey, there's a flaw, if we attack one of their fighters from the left side, they overcompensate when turning to meet the attack and create a vulnerability." Against human pilots, maybe the first couple who fall for this trick get killed and can't pass on the information. But eventually someone says, "Oh, when we do X, we get killed. We'd better do something different." A human can make that sort of analysis and decision in the midst of battle. Can the AI adjust like that? Only if the programmers have anticipated the problem and programmed for it. And they can't think of everything in advance. A computer can be counted on to exhibit the same bug every time. A human would figure it out and stop doing it.

Don't confuse the AIs in video games with trying to apply AI to real life. In a video game, the game designer controls the entire environment. He doesn't have to program the AI to be able to tell the difference between an enemy soldier and a fence post, because he just programs in identifiers of the enemy soldiers. He doesn't have to deal with unforeseen situations, because as the designer, he knows all the possible situations because he invented them. This is a WAY easier job than programming an AI to function in the real world. And even at that, game AIs often do stupid things, like run into a wall and then stand there trying to walk through the wall because they can't figure out how to go around it.

Answer 15

What about a technological limitation, such as true random number generation by a machine not being possible.

So while AI is viable and an effective strategy against some targets, if a computer can analyze the reactions of a ship long enough they can crack the seed number and begin to predict the actions/reactions ships will take with reasonable accuracy, making them easier to neutralize.

This way, while a majority of forces will be supported by AI, some advanced targets will be more successfully engaged with human pilots who can be truly unpredictable?

Answer 16

High levels of space radiation might do the trick.

Having powerful AI requires fast computers with quite a bit of memory at their disposal. However, space radiation can damage electronics. This requires building a system with more redundancy and more shielding. Building those in means the computer does not have the resources it would otherwise.

You don't actually need to use space radiation - your Minovsky Particles could be to blame. Perhaps it's a byproduct of the energy shields your ships have. Whatever you decide, it is something that limits the complexity of electronics in your ships. So rather than having futuristic terahertz processors, they could be limited to megahertz processors with megabytes of ram (which should be enough for an AI that helps the pilot fly the ship). This severely limits how powerful a shipboard AI could be and means that a trained pilot would be absolutely necessary.

This also leaves you with some options for allowing AI controlled ships. The evil space empire could develop a better means of shielding their electronics, allowing them to have AI controlled ships that could be at or above the level of the average spaceship pilot. They could also have short-lived AI ships - they could be far superior to any human pilot, but if a pilot can manage to evade them for a few minutes the AI would begin to noticeably degrade and significantly improve the pilot's chance of beating it. A third option would be AI ships that forsake the energy shield. These would go down from a single shot but, depending on the weapons systems you have in mind, a powerful AI might be pretty good at avoiding being shot.

Answer 17

In addition a future AI could presumably be faster to respond to attacks 
then humans, less predictable, more trustworthy

Whilst I agree that an AI would account for fewer drawbacks in terms of it not succumbing to fear, fighting until the end rather than ejecting etc. I disagree with the other two statements.

Whilst an AI could respond faster to an attack (evading missiles etc.) they would not necessarily be able to preempt attacks as humans probably could. In a Mexican standoff situation with another solitary pilot, if a computer can only react to their attack, they would always lose. With a human you have more ways to work such as on instinct, tact etc.

Also, computers tend to be very predictable. If there was AI jets vs AI jets, even if there are millions of pre-programmed evasive maneuvers for a single jet to make, another jet could analyze its movements and knowing the set of moves it could make, calculate which one is most likely and react accordingly. Even though there are a million variations, there is still only a finite amount.

A human could adapt more easily to situations and be less predictable, even going so far as to make the wrong decision, which is generally less predictable then what you would expect them to do in a situation.

My suggestion would be to justify it this way: AI controlled fighters would be good, probably very good, and better than most human fighters. However, they are too consistent, and you know exactly what to expect.

So whilst they can be better than most humans, the best humans are still better than any AI. This would mean there are still AI fighters, and even drones, but the very best of the best fighters would still beat any of them one-on-one in most situations.

Answer 18

Another possibility deals with Artificial Intelligence and ethics. Prof. Joseph Weizenbaum who wrote a book called Computer Power and Human Reason. The book says that technology will indeed evolve and Artificial Intelligence can become possible (strong AI), but he says AI will always lack certain qualities like compassion (from Wikipedia):

His influential 1976 book Computer Power and Human Reason displays his ambivalence towards computer technology and lays out his case: while Artificial Intelligence may be possible, we should never allow computers to make important decisions because computers will always lack human qualities such as compassion and wisdom. Weizenbaum makes the crucial distinction between deciding and choosing. Deciding is a computational activity, something that can ultimately be programmed. Choice, however, is the product of judgment, not calculation. It is the capacity to choose that ultimately makes us human. Comprehensive human judgment is able to include non-mathematical factors, such as emotions. Judgment can compare apples and oranges, and can do so without quantifying each fruit type and then reductively quantifying each to factors necessary for comparison.

Now in case a machine can choose whether one should kill a person or not, that clearly validates the above stated concept.

This can be beneficial since for instance saving ones life creates friendship and thus can construct military coalitions based on more than simply "beneficial for both sides". The same aspect is I Robot where the author (seems to) warn that AI can misunderstanding the goal and for instance destroy democracy, because it things mankind will benefit from it in the future.

In other words, one must be careful with AI since certain aspects that might look trivial to humans are hard to learn to machines. Therefore deciding about life and death is perhaps better made by humans.

Answer 19

I might be late for the party but, how about human intuition and "gut feeling" ?

An AI no matter how advanced could be broken down to "IF .... THEN", but in situations where there is no predefined situation to fill the space after "THEN", the AI would have to figure out what to do, where as the human could follow intuition, and gut-feelings over a statistically calculated qualified guess.

It is the same with computer chess engines, if you play a certain computer chess engine too many times you can win over it because it becomes predictable, even though you can make a computer chess engine that is better at chess than a human. The same would happen with the AI pilots, where as humans is more unpredictable and that is an advantage in warfare as the quote states

If we don't know what we are doing, the enemy certainly can't anticipate our future actions!

Answer 20

I feel like Mr Molot started the bounty because of my comments in another thread, so I feel like I should provide an answer.

Most answers in this thread are "if conditions 1-x are fulfilled, human pilots might be used". I will choose a similar strategy.

Often conditions 1-x are pretty hard to fulfill or super unlikely, but I also think it is a fair conclusion to draw from the answers that all applications of standard humans on space fighters are highly situational if not just straight up a disadvantage. I also should say that I do not think this question is a perfect match for the discussion form the other thread because the initial poster imposes quite a lot of restrictions. I will not be super specific to this already answered and ancient question but a bit more general because of the circumstances. It might also be that someone said something similar in the other couple of answers. As stated before, I just feel like I need to justify my comments in another thread. I'm also debating small spacecrafts explicitly because it was the topic of the other discussion. I think it is somewhat within the scope of the question. I believe Mr Molot had good reasons to choose this particular one.

I'm strongly considering starting a series of threads to determine what a space battlefield would look like (which I think is important to answer first actually), but I currently doubt, especially after reading the answers in this thread, that one would find a truly great answer here. It seems like a lot of work for a project I'm not really working on right now.

So here is the only reasons I can come up with why humans should man small spacecrafts in a real space battle:

1. Civilian warfare: If you are doing small and civilian operations, take for example space pirates or police work or just a dispute between private people that needs settling, a space fight between human pilots might break out. It is unlikely that you have an arsenal of AI drones with you at all times and maybe the battle AIs are heavily restricted and private people do not have access to them. All weaponry might be self made or even illegal. Kind of like you cannot drive a tank around as a civilian. Another possibility would be that on-board combat AI is factory programmed to fight off other things than humans because murder. You might have to manually do some stuff yourself in battle. There might even be a program in place specifically to prevent space duels and piracy - if you are into this thing, you might have to do it manually to some degree. I do not think that this is likely since those ships imo would still have at least heavy AI assistance, but I would argue the smaller and short-ranged and slow-paced the battle, the more uses for humans. A real world example similar to this would be fist or knife fights: It doesn't matter that a pistol is better than your fist, sometimes in a dark alley it is your only choice. You cannot argue "but the atom bomb is a lot better at killing people" if you have a knife at your throat.

2. The last stand: The evil aliens are coming to enslave human colony X. So you either become a slave or fight for your freedom. You have a small transport spacecraft that lacks a battle AI and a battle formation program as required by the local admiral but with some weaponry and in combination with the AI defense forces of that colony, you join the defense. Every ship counts. It's often better to die in battle than to end up in an alien space mine.

3. Sport: Well, some people do stuff for sport. People row even though we have motorized ships that are much more effective and efficient at being boats. Still people row - so some people might just find it fun and engage in fake space battles. Maybe Disney will one day make "the true star wars experience"

4. Ethics and moralists: Maybe the AI got the same rights as humans in a particular world and no longer is mass produced. Also the AI might think it is unfair that they have to do all the fighting while humans enjoy "love". Every AI is precious, even the targeting program. So let's have the humans do their share of stuff.

5. Humans are AI: This I suggested in the other discussion. If you have enhanced humans, maybe with integrated AIs, the combination with human qualities might make them the best pilots possible. I do not know why one would do such a research project, but those humans might be specially made by some company and lack such things as "free will". One could perhaps even see them as computers enhanced by human parts.

6. No choice left: (kind of similar to point 2) You are flying through space with humans on board. (Optional: Your ship has sustained damage. You need to operate manually because nothing is working.) You get attacked. Well, what do you do? Fight or just accept your fate and die? I personally wouldn't just die just because I'm no AI and not good at space battle, I would still at least try my best to fight them back. Most likely I would lose, but still - people will tell the story of the guy that had to manually pilot in a battle that one time.

Answer 21

Take a page from battlestar galactica.

AI was used in past wars but was found too unstable and prone to dangerous malware to effect the totality of an AI based fleet.

Therefore it became too dangerous to deploy even a moderate amount of AI anywhere within the fleet in case of quick and easy fleet compromise.

Answer 22

That planes are controlled by AIs does not imply that humans are not flying it. If humans exist in electronic form in the future, they can upload themselves to the flight control computers of planes and fly the plane.

Answer 23

Lots of answers already. Another real-world-ish point:

Objections of the military elite

Basically, the point is that militaries are powerful organisations with distinct hierachies. Being a pilot of a spaceship could become a thing of prestige, that you have to be brought up in a noble family, with lots of wealth and training to attain. In turn, these families have a lot of political power, and see a future as a pilot as something to assure success for their children. For instance, think of them as the medieval knights of the future, each owning and maintaining their own fighting ships.

Replacing humans with AIs then becomes a powerful challenge to these people, making them and their skills redundant. In wartime, it might be an advantage, but in peacetime it is a disruptive force in the established social order.

Answer 24

One more general Approach, which can be compatible with many of the answers here:

Don't make up one reason why it is impossible, just many small reasons which make it like really really hard, expensive and risky

You don't have to make AI impossible in your Scenario. Just set the bar very high - there are a range of simple physical engineering problems. AI may work quite good in the laboratory and one state invested billions in a huge AI fleet, which got wiped out by 10 fighters, because of (radiation, bugs, predictability, whatever). You can even make it a story point that the public doesn't know exactly, why the big AI fleet was so easily destroyed. But after that most AI-fleet programs got their funds seriously shortened.

So they are working on AI ships - and they may even be possible they will be better in combat than anything. Maybe we just have to invest another 5 years another few million credits into the development and maybe then it will work. - But that is quite risky and the war doesn't wait. So investing the time and resources into new weapons, better ships will produce immediate results instead of the long-term gamble of AI.

I think it is a bit like - why don't we have flying cars today? There is not a single reason which makes it impossible, but a lot of small very practical issues why the theoretical easy solution doesn't work in the real world and needs years of engineering...

Answer 25

Like @Erik said in the comments, the thing you are looking for is Drone pilots. Pilots will sit in a command center on a large ship or back on a home planet and control ships remotely. This keeps the ships small and expendable while keeping the actual pilots completely safe. AI surely will have developed enough to fly ships, but you can justify drones instead by saying that AI is no match for the wit of live pilots. This seems like the most logical approach to me!

Answer 26

Depending on what sorts of resources your society has access to, you could claim human pilots as a matter of economics.

If a cost-benefit analysis shows that silicon (or unobtanium or whatever essential element your AIs require) is more expensive than the cost of training and dispatching humans, and the relative advantage of AI capabilities just doesn't justify the cost of making/maintaining them, then putting meat in the pilot seat becomes a financial necessity.

Just a thought.

Answer 27

Something I haven't read so far depite all the answers isnt rogue AI, but unpredictable AI.

Actual full AI are so incredibly advanced that within a year or so after their inception we wouldnt even understand them. And we wont know exactly what the AI will have learned. Its unlikely such an AI would only be used for warfare, but also construction, research, traffic control and just about anything you can think off. If we left the AI to its devices it might forcibly remove humans from their buildings and build something else, if it removes the humans beforehand at all. You dont know what and how it learned or what it gives priority too. Warfare would be the same, the AI could do anything to achieve it's goals, even things unthinkable for the humans like agree on peace and collaborate with the "enemy" or their AI's against it's masters will. This is one reason why AI might only be a locked box project. The AI (or multiple AI without knowledge of the others) get a problem and have to create an answer, then the humans look at the answers and fully explore the consequences before acting on it. This means no AI directly controlling anything.

Human reaction time is based on a nervous system that's both as energy efficiënt and fast as possible. Through forced evolution (slower pilots die fast ones live), breeding programs, genetic manipulation or even building them up from the ground you could accelerate their brain and nervous system far enough it stops mattering.

"theres no stealth in space", you cant hide your heat signature against the cold black space. But why should this mean you cant fool weak AI systems? Imagine countermeasure research accelerating and any combat starting with dozens if not hundreds of fake signatures. To discover what's what you first get the engagement closer (possible with FTL). Then to make sure you dont waste your ammo and time on fake targets you have humans control the fighters, flying and fighting in a mass of randomly movng countermeasures that both the capitol ships and fighters keep launching in all directions. Combine this with measures that reduce your own signature, like changing how much your surface radiates while facing the opposition and/or temporarily storing heat for the duration of combat.

Answer 28

I feel the question here is really about justifying AI of nearly any level in a say Type 1 or Type 2 class civilization, using the Kardashev scale from the prospective of a Type 0 civilization looking forward.

Considering most Artificial Intelligence is based on algorithmic programming based on biological entities (i.e. DNA, neural nets/brain cells, insect behavior, etc...) that really only apply to a small portion of what is considered actual intelligence. One would have to keep in mind that there are several systems in place to build up a "virtual" or Digital Intelligence.

Take for example the world of Star Wars and Star Trek, they both have artificial intelligence or AI in one manner or another. Star Wars takes it with there being an android AI and not so much digital based AI thus implying at one point their technology had digital AI but "evolved" based on external factors or other agents to be more common within a physical form of androids than a virtual one like holographics or existing in a computer of sorts. While Star Trek takes the opposite approach and has their society dealing with holographic and digital AI (which in researching a computer system like LCARS could not operate without a sort of advanced yet rudimentary AI ) as the norm instead of androids why androids are considered unnecessary and/or untrusted with one exception.

Now looking within our own reality, We recently have developed 5d storage technology using crystal lattice and lasers. Very sci-fi yes, but because it offered no better alternative to our current magnetic and solid state transistor based storage mediums we as a people did not develop this. The same happen with Steam engines, a greek phylosopher invented the first steam turbine and it wasn't until 1800 years later that the world rediscovered this technology to fuel the Industrial Revolution.

The point I'm making here is if a culture, events, or faction within that culture prevent the mass wide spread adoption rate of a technology then that said technology could either become rarely unused, or completely abandoned by a few factions. Thus one could have their main protagonists be one of these same factions that does not utilize an AI system for their fighters for anything more than say guided fly-by-wire systems or advance situational automation without any interaction of a pilot but still require a pilot to fully operate the craft. This could be handed down by a governing body or just solely by that team's preference based on their own skills, abilities and experiences, thus leading to a consideration any sort of AI piloted drones inferior technology.

One could even justify this inferiority in technology as just because most digital intelligence systems or AI are not ever built up to be sapian level drones with a semi-conscience mind but merely a set of hardware and software systems built up to operate with an insect like mind but still needing an operator because AI was never needed to develop further than this nor did any sort of mutation code was developed to allow the AI to self evolve, replicate, or expand its own programming therefor never developing into a digital life form.

Answer 29

Being a huge Gundam fan and lover of unrealistically flashy sci-fi combats, I have asked myself this very same question before, and this is the solution I came up with: The AIs handle basically everything during a fight, reactions, communications, tactics, prediction, etc... The humans would still be there just to make mistakes.

Now hear me out - there is no reasonable way a human being can outthink, outsmart, outmaneuver an AI, but we sure as hell can out-stupid them. We will make terrible decisions (from a logical or predictive viewpoint), and that would make the machine's movements harder to predict. Basically, think of the human as a random seed generator. As long as his/her decision is not purely self-destructive, the AI will roll with it just to throw off the enemy AI.

I just know this idea is flawed somehow, but it seemed like a good one to me initially.

Another idea I had, which places the human pilots in a much brighter light would be to have them in charge of the battle tactics being utilized. They do not control the machine itself, they merely say 'We should do x.' And the AI handles all the maneuvering and reacting.

Unfortunately, both these excuses have one common requirement: some sort of powerful all-enveloping communications jammer that stops the human pilots from not being in the machine at all. Otherwise, the humans will just sit in base while the AIs do everything.

Answer 30

Racism/Fear of the Machine

Another possibility, would be that humans are so afraid of a potential robot uprising that they refuse to put them in positions where they could possibly affect a revolt. Maybe robots/AIs aren't allowed to have weapons of any kind, in which case a space fighter is certainly a weapon.