# What issues would an AI asteroid mining stations have to be prepared for?

Your company has just been contracted to develop a purpose built AI. The purpose of this AI is to control a mining station that will be anchored within an asteroid belt. The space station will be equipped with refining and construction facilities, docking bays for transport ships, and rudimentary defense capabilities.

The goal of your AI will be to maintain the highest rate of refined metal production to be sent to Earth via transport ships. So any issue which would impede or outright stop a station's ability to mine and refine metal from the surrounding asteroids needs to be corrected by the AI using the equipment on hand. Because of ... reasons ... software updates and physical upgrades will not be available after launch, so every (or at least MOST every) potential issue needs to be prepared for now, before launch.

The AI will control every part of this operation including the building of the transport ships and their navigation from the station to LEO (where they'll be emptied by manned crews).

What potential issues could impede the space station's production capacity that we will need to prepare the AI to handle?

• Is three-laws-of-robotics compliance optional? – Philipp Jan 3 '17 at 23:48
• @Philipp Consider it a robotic wild west, no laws required – Bitsplease Jan 3 '17 at 23:50
• This seems rather broad. We have at least refining, construction, transport, and defense issues. Each of those is a broad category in and of itself with multiple options. For example, if no transport ship arrives, does it build one? Send out a message calling for one? Just continue waiting? And each of those decisions lead to other decisions. E.g. if it builds one and dispatches it with cargo, what happens if it can't find its destination? Even if split into three questions, this needs more background. – Brythan Jan 4 '17 at 0:02
• @Brythan I've answered your question in an edit, let me know what other details you think might be useful – Bitsplease Jan 4 '17 at 0:31
• As always, be careful with from the surrounding asteroids because the asteroid belt is not as densely populated as fiction portrays it - it's actually extremely sparsely populated for the volume it takes up. – Zxyrra Jan 4 '17 at 5:35

Assume average "shortsighted human intelligence". Good enough for solving regular problems, but thinking short-term and locally, so lacking the ability to properly react to long-term issues or see the bigger picture.

A lot of smart people lacks those abilities, which hinders their development (and human race). Get one of them as a mining worker, and they will be able to mine properly, but they may fail to prepare for major issues.

A few examples of issues that such an intelligence (human or artificial) fails to handle:

• Proper maintenance of mining facilities. That's how a lot of mines collapse even with good intentioned efforts to prevent it; simply, the investment on maintenance is focused on the short term instead of long term

• Smart competitors subverting or sabotaging the facilities, with tricky techniques (not just brute force)

• Strange equipment failures, that require specialized knowledge that only the manufacturer has. Particularly those failures that are not evident but subtle, like giving wrong readings that appear normal

• Incorrect estimates of the materials at hand (AI would have assumed the planning engineers were correct), which render the station unable to keep building some critical stuff (like ships), due to lack of a required component

# If it isn't a general intelligence, it won't work

Unless your AI has problem solving skills at least equal to a human, your venture will fail. The list of potential problems is basically infinite. There are ALWAYS unexpected problems, any software developer can tell you that. No product is ever bug free. If there isn't some way to solve the problems that the bugs bring up, problems that are unique and unpredictable, your venture will fail.

Also, a single human, no matter how skilled, would probably not be enough to solve the problems either. There are too many variables, from software to hardware to astronomy to geology, of things that can go wrong or that will present problems. After all, to mine things on Earth takes billion dollar companies with hundreds or thousands of engineers. So your AI will need to be able to do the work of hundreds of humans.

• I do agree that an AI is bound to fail eventually, that's kind of the point, I'm looking for what some of the biggest contenders are for causing that failure – Bitsplease Jan 4 '17 at 0:30
• @AustinPhilp First off, if the AI is smart enough to solve its own problems, you will never know what went wrong :) But, to be honest, a list of potential failures is far too broad a question. I actually started typing up a paragraph on that, and it turned into 15 bullet points, and then I decided I should avoid answering that because it is simply 'too broad'. If you want to know more about more specific problems, you could ask questions like 'How could a mining bot fail while trying to harvest metal from an M-type asteroid with this method that I am describing?" – kingledion Jan 4 '17 at 1:06
• @AustinPhilp IF that is what you want then that should be your question. If for no other reason than it is a much easier question to answer. – John Jan 4 '17 at 2:23
• @John That is their question; read the last sentence. – Zxyrra Jan 4 '17 at 5:36
• @kingledion I agree that asking for all the potential problems is broad, but perhaps some of the major ones - biggest things to look out for - would help the OP, as that is still what they would like to know. – Zxyrra Jan 4 '17 at 5:37

Hostile takeover, by highjackers/pirates and/or a rival AI.

The good news is that you're not alone, brave space AI. The bad news is that it's a dog eat dog, make that an AI-eat-AI solar system out there. As soon as other entities suspect that you've struck Platinum, Uranium or HE3, you'd better be prepared to mount a vigorous defense of your mining claim. The Sheriffs are few and far between out in the Belt.

"We have all the problems of a major theme park and a major zoo and the computers aren't even on their feet yet." - Arnold, Jurassic Park

We won't be able to cover all of the issues that your mining station might face in a Stack Exchange post. Space is an unforgiving place with a tremendous number of technical challenges. If I were to somehow condense each issue into one word in this post, I'd still need to type more than this entire answer.

We can categorize them though:

• Liftoff issues
• Landing issues
• Mining issues
• AI issues
• Cost issues

To get a sense of just how difficult space is, consider just the launch process. Here's the countdown for the SpaceX Falcon 9, lifted from another stack exchange answer:

L-13:30:00  Dragon Power Up
L-7:30:00   Launch Vehicle Power Up
L-1:00:00   Weather Briefing
T-0:09:40   Verify that Terminal Countdown has started
T-0:09:30   Merlin 1C: Lox Bleeder Valves Open
T-0:07:00   Dragon to Internal Power
T-0:04:46   Stage 1 & Stage 2 Auto Sequence starts
T-0:04:30   Transfer to Internal Power
T-0:04:10   Vehicle Release Auto Sequence
T-0:03:40   TEA-TEB Ignition System Activation
T-0:03:25   Flight Termination System to Internal Power
T-0:03:11   Flight Termination System Armed
T-0:03:02   LOX Topping Termination
T-0:03:00   Second Stage Thrust Vector Actuator Test
T-0:02:00   Range Verification
T-0:01:30   Final Engine Chilldown, Pre-Valves/Bleeders Open
T-0:01:00   Flight Computer to start-up
T-0:00:50   First Stage Thrust Vector Actuator Test
T-0:00:40   Propellant Tank Pressurization
T-0:00:20   All Tanks at Flight Pressure
T-0:00:15   Arm Pyrotechnics
T-0:00:03   Merlin Engine Ignition
T-0:00:00   LIFTOFF


Now please note that this is just the timeline for synchronizing tests. Each of these events comes with dozens of detailed checks that took many years to iron out. This doesn't include all of the incredible feats they have to accomplish after liftoff either.

Landing is no simpler. There's the amazing show of what it takes to land a booster that SpaceX has demonstrated (sped up video). There's also the astonishing SkyCrane used to land curiosity (seriously insane people). Needless to say, the differences in velocities of different pieces is enormous in interplanetary transport, and it takes a lot of care to get them right. Feel free to look up all of our failed landings to see how many things can go wrong.

Mining, itself, is tricky. I don't have much experience on the topic, but the general issue with mining is dust. Dust gets into everything. At first I'd thought that perhaps this might not be a problem on the moon, with its lack of an atmosphere to carry the dust. Turns out, moon dust is still nasty:

These micron-sized spikes get bombarded by ultraviolet rays from the sun, giving each grain an electrostatic charge. The result: A highly abrasive material that gets on everything and stays there. Hatches, seals, valves, bearings—any machinery that moves on the lunar surface will be subjected to dust damage. "Imagine sprinkling broken glass onto a seal," said Gentry. "And then every time you cycle it, you sprinkle more glass. Sooner or later it’s going to leak."

Most mines resolve these issues by having humans there to fix the unexpected. You are only going to have AIs, so you will have all sorts of issues with the AIs. You're going to have to decide how smart they are and how capable they are, and then write your difficulties accordingly. For us, just making robots that walk is still considered an achievement, even if they may not happy about it (one of my favorite overdubs). If you make the robots too smart, you have to worry about other issues (this overdub of the same video shows where that might go).

Finally, you have to deal with all of the cost issues. Depending on the value of your mined material, you may have to cut corners. A Falcon 9 launch costs \$62 million, and can carry about 5,000kg (to GTO). That means launching material from Earth costs \$12,400/kg. Obviously returning mined materials from a much smaller gravity well will be far cheaper, but you're going to have to pay close attention to the cost of any material you need to send up.

Almost all major problems your mining base could face can be generalized as a technical failure of one or more technical components of the base.

It doesn't matter if a component malfunctions due to an asteroid impact, human sabotage, an alien attack or a manufacturing error. Fact is, the AI must be able to either replace or repair it. This includes all the equipment it needs to make repairs, so these should be redundant.

The AI itself is an interesting point of failure here. If the AI itself malfunctions, it might not even realize that it is damaged. A possible solution could be to have at least three identical, synchronized AI cores. If one of them ever comes to a different conclusion than the other two, that node is malfunctioning and the other two will switch it off and try to repair and resynchronize it. Make sure that the self-preservation subroutines of your AI make an explicit exceptions for this situation, or your mining base will likely be torn apart by an AI war started by a disagreement about the exact value of the millionths digit of PI.

There are other problems, like lack of supplies, lack of mineable resources in the vicinity, the area becoming too hazardous for future mining operations, etc. But these are all problems which are outside of the control of the AI. It can (and should) report them, but it can't solve them. In fact you don't want the AI to try to solve these problems on its own, because with its primary priority being "maintain the highest rate of refined metal production", you might not agree with the solutions it comes up with. For example, it might come to the conclusion that the 3rd planet of the solar system still has more resources than these asteroids. The only problem it needs to solve to be able to mine them is the resistance of those pesky carbon-based life forms who are squatting on it.

A solution for the murderous-AI-trying-to-destroy-humanity problem can be to make following the three laws of robotics its primary priorities. Executing its actual mission would then come on place 4.

You would have to give it a weapon (I'm pretty sure guns can't work in space, you need oxygen to create fire to ignite the gunpowder, causing a force shoving the bullet out of the casing. There is no oxygen in space, thus no fire can be created. Correct me if I'm wrong) capable of destroying asteroids or armor capable of protecting the quarrying machine from colliding asteroids, as well as the ability to retrieve it when it's done digging, and a way to store the stuff you dig up.

• Both gunpowder and modern firearm propellants work fine in vacuum; they don't need external air at all! That said, I doubt firearms are the best defensive weapons. I'm thinking lasers and particle beams, especially since those could also be used in mining. – Catalyst Jan 4 '17 at 18:56
• you probably need to worry about the recoil: firing a machine gun is not that different than having a rocket in your hand. Better realize it before being orbited out of your mine... – L.Dutch - Reinstate Monica Jan 5 '17 at 10:43