# How would a preemptive crime fighting group prove they stopped a criminal?

In my world, there is a government agency that stops crimes before they happen, kind of like in Minority Report. However, in this case, they swoop in as early as possible, sometimes waiting outside the door as they were born. Because of this, they can't prove they stopped them because they were in the middle of shooting someone. How this agency gets its intel is from a (basically) all-knowing future-seeing AI that just gives them a name and image of the person, and place and time to arrest the would-be criminal. The AI is limited by its programming so that the only intervention it can do is giving said information to the government agency.

Since there is no proof besides 'AI said so', how does this agency prove that the people they arrest would be criminals?

• They do not arrest future criminals. They arrest innocent people. It is the thugs of this fascist agency who are the criminals. "A parrot picked up a fortune cookie out of a jar" is never valid justification to deprieve somebody of liberty. Sep 27 '20 at 18:12
• Regarding "limited by its programming" - recommended reading: Asimov's All the troubles of the world - it has a somewhat similar premise to Minority Report - both dealing with the failure of a fail-proof proactive crime prevention "oracle" Sep 27 '20 at 18:26
• If you have that powerful a predictive model, why not rehabilitate criminals? After all, a productive worker is a lot more valuable than a prisoner or corpse, and unlike every other rehabilitative program, you know if it will work. Sep 27 '20 at 19:55
• @Ceramicmrno0b if they do it right, they don't have to. If the AI is that smart, it can tell them to stop the kid from shoplifting which lead down the slippery slope to bank robbing. The bullying that insights a murder. The child abuse that continues the cycle of violence. Sep 27 '20 at 22:37
• @gen-ℤreadytoperish Was it "Person of Interest"?
– Kaz
Sep 28 '20 at 13:48

Arrest that newborn baby? I think you mean give that newborn baby a college scholarship.

If the AI so so omniscient it can predict crimes decades in advance, even knowing the rippling effects of the AI's own interventions, you wouldn't need something as blunt as arrests.

You can just identify a fork in the road where you can avoid the problem with an unquestionably positive intervention. Give Hitler an art school scholarship, make sure Osama Bin Laden has a great time on his visit to America, and so on. Nobody's going to fear oppression at the hands of an education-and-international-cultural-understanding charity.

Edit: If you can predict crime accurately 10+ years in advance, and when you intervene to stop one future-crime it doesn't mess up causality and throw off all your other predictions, that means you can accurately predict the results of your own interventions. Unexpected consequences and perverse incentives will never be a problem, as nothing is unexpected to the omniscient AI.

• Comments are not for extended discussion; this conversation has been moved to chat.
– L.Dutch
Sep 30 '20 at 14:15
• Plenty of college graduates commit crimes as well. Who knows, maybe sending them to college gave them the knowledge they needed to better commit the crime? Sep 30 '20 at 19:35

# Statistics

Every now and then you let some crime happen just as the AI predicted. If every unimpeded prediction turns out to be true, you can convince the population that the AI never makes a mistake.

This way you can also arrest and murder a lot of innocents with impunity because "the AI accused them".

• Fascistic implications aside, using statistics of crime in this state compared to others could prove something is done right - if their crime rate is basically zero and other nation still suffer from crime, it could be a proof... Sep 27 '20 at 18:29
• This don't proove that every stopped guy would turn out to be a criminal. A dystopian dictatorship could totally use the minority report group to arrest political opponent while stating he would be criminal. Heck, with this method, you could fake all this stuff by organizing the very same crime that you "knew becausse AI but let happen just to proove it's true". Sep 28 '20 at 9:40
• If you want a version of this that's less open (but still far from immune) to corruption: report all cases to an independent regulatory body before acting on them and they randomly pick some that won't be stopped. Sep 28 '20 at 11:05
• @Kepotx Right. Arresting everyone would eliminate all crime, but would of course be a totally naive algorithm. To be correct it needs have have low false positive and false negative rates. Showing that crimes go to zero only demonstrates a low false negative rate (guilty people locked up), but the false positive rate (innocent people falsely accused) could still be enormous.
– bob
Sep 28 '20 at 14:24
• @JoeBloggs. Depends on how authority flows from the AI. Will the AI eventually label the speaker as a criminal, and will his minions obey? Sep 29 '20 at 21:37

Wait until the very last moment.

You might not be able to absolutely prove that the guy you grabbed with the mask and shotgun was going to rob the bank he was standing outside, but most people would agree that he has a lot of questions to answer. You should have enough evidence to justify your intervention. You might not have enough evidence to send him to prison for fifteen years, but your primary goal is preventing crimes, not locking people up.

(This also has an advantage in terms of drama. If they have a policy of waiting until the last moment, then it makes the whole thing more exciting, because of the potential for things to go wrong.)

• The cops only know the identity of the perpetrator and not the time/date/location/nature of the crime, and judging from the fact that newborns get arrested, the crime may not occur for years/decades. This is going to be one long stakeout. Sep 28 '20 at 16:06
• @NuclearWang, a long stakeout might be better than a long incarceration. At least the stakeout could prevent the crime by the watched knowing they are watched or because the watchers determine the reason why the crime was committed and then intervene to reverse the negative effects. Or the "stakeout" is more of a mentor role, than impartial/impassive watcher. Just my 2 cents. Sep 28 '20 at 16:28
• @computercarguy Agree it's better for society, my point is that this method might be cost prohibitive. You'd have an entire surveillance team dedicated to one single case of unknown importance, potentially for years. I'm not sure the police department will be able to justify spending 100+ years' worth of officer salary to prevent someone shoplifting. Sep 28 '20 at 16:50
• @NuclearWang, done efficiently, the "stakeout" could be a single person, or a team of three people for shifts. This could also be a twist that brings down the system, too. It could also get into "who watches the watchers?" Sep 28 '20 at 16:54
• @NuclearWang, what if the watcher is that same person, only after the understand their crime and work with their younger self to prevent the crime? Talk about a mind bending twist. :-) Sep 28 '20 at 17:10

### You don't.

Your current model of response can't be internally consistent.

Being able to predict crime at birth and then prevent it implies the calculated future can be changed. That means you'll run into issues like:

• Alex is born in 2010 and will commit murder in 2050
• Bob is born in 2015 and will commit murder in 2040

By your prediction, both of these people will be taken from their homes and locked up at birth.

Problem being: what if Alex was going to murder Bob in 2050? You lock him up in 2010, before Bob is even born, and then in 2015 when Bob is born, you lock up Bob as well, thereby making it impossible for Alex to murder him in 2050, which means that Alex is now innocent of the crime he was going to commit because it can't happen anymore, because two different interventions have interfered with one another.

There's a reason most of these interventions in stories happen last moment, and it's because the knowledge of the future influences what actually happens. You can't really get around that. The earlier you interfere, the bigger the chance the predicted future no longer plays out that way and you locked up an innocent person.

• Why can't 40yo Alex murder 35yo Bob in prison, as revenge for Bob murdering Alex's mentor a decade earlier?
– Nij
Sep 28 '20 at 23:14
• @Nij he could, but if that's the original scenario that he got locked up for, then the future AI's calculation already took into accounts the AI's own future calculations and all the ways in which it and its team are going to interact with the timeline, so in that scenario it basically boils down to "nothing you do matters" for all the human actors.
– Erik
Sep 29 '20 at 6:16
• This answer gives an impossible example, because the AI can't cause any paradoxes like this. Any knowledge of the future that any person learns will never influence the future the AI initially predicted. The AI is perfect. Sep 29 '20 at 13:49
• Also, there's "Bob is an arsehole and in 2050, Alex gets tired of his shit and kills him." But if you arrested Bob way before then, Alex would never have been motivated to commit the crime. Sep 30 '20 at 13:07
• @theonlygusti if “nothing you do matters” and people still commit the crime they were accused for, even when being in prison, the question arises, why we are putting the people in prison. If nothing prevents the crimes, not even an active attempt of the supposed to be criminal, that person can’t be guilty, hence, doesn’t deserve punishment and non-working prevention can’t be a reason either. The bottom line would be that this AI ensures that far more people get far more prison, without improving anyone’s quality of life. Sep 30 '20 at 13:27

The AI makes simulated documentaries showing the future that was prevented.

The all knowing AI can know what crime the future criminal will do as well as other events that take place in this timeline. It crafts documentaries with virtual actors to show interested parties what would have happened if this individual had continued on his or her previous path. The documentaries are pretty compelling, and might be true.

As an internal control, sometimes a criminal is not apprehended and events of his or her actual life are tested against the AI modeled documentary, to make sure the AI is accurate.

The AI future documentaries will also make good prose for your story and move things along. The existence of such documentaries of the future lead to other obvious questions; for example other future events aside from crimes that appear in these documentaries and which might be of interest. There may also be recurring characteristics of the AI generated documentaries that betray an agenda by other agencies - or perhaps by the AI itself. The best way to catch those would be analysis of many thousands of them by another AI.

This is a great scenario and I can think of enough Twilight Zone type last minute twists to occupy an entire season. For example: the chief investigator herself is one of those criminals allowed to continue on her current timeline, in the interest of having a control group. She knows this. She has not yet been able to break way from her predicted future path.

• I was going to disagree with this, based on the possibility that the AI could alter it's records to justify it's own existence, which is eventually uncovered, but I like your version of a twist. Sep 28 '20 at 16:33
• Eventually, you will end up writing Dune Sep 29 '20 at 21:44
• The fact that people which are left alone behave as predicted doesn’t prove that the other people would have behaved as predicted, regardless of how we influence their life, to the point that a lifetime jail is the only remaining option. Well, since jail does prevent their crime, in other words, their crime can be prevented, it’s hard to believe that there can’t be other options. Sep 30 '20 at 13:21
• I find myself thinking of the investigator who is allowed to continue on her path. She is the main character on the show. Every night, as she works out, she watches the documentary predicting her life. You are several episodes in before you realize what she is watching. An event in the documentary happened earlier in that episode and she did not realize it was happening at the time. She is a little closer to her destiny. Sep 30 '20 at 23:48

If the AI's thinking can not be scrutinized, then there's no way to prove it. The only way is for the people to trust the AI blindly.

Start slow. Make AI into some sort of infallible oracle whose predictions are 100% accurate. Make another stipulation - acting on those predictions can prevent the outcome, but the acting should be done 100% in a way prescribed by the AI. Have AI make numerous predictions that way and save countless human lives. People would start trusting the AI more than their own judgement.

Then a more drastic preventative action would start. Put a person into jail before a crime is committed. For a crime serious enough, execute criminal early enough. Cruel? But this is the only way to prevent the crime. Remember, the AI is infallible, and mere people have no way of raising an objective to its commands.

At the end, you will have a murderous dystopia, but people would believe that they live in a crime-free paradise.

• Important to remember that this situation would arise after many generations. Sep 29 '20 at 21:47

Russian Sci-Fi kicks back! This is a very short story by E. and L. Lukin in its completeness, translated by me on the fly, which should illustrate the point. A conclusion follows.

"The present case is not as simple as it may seem at the first glance.

"I claim, there was no crime at all. There was only a non-executed criminal intent. Because, if a crime was there, where are its fruits? Where is the damage? Where is the victim, at last?

"Yes, my defendant time-traveled to the Stone Age and teached the foundations of Quantum Mechanics to Cro-Magnon Men. So what? As the expertise showed, the regal flow of the history did not change at all. Yes, ladies and gentlemen, it did not. The events occurred in the same sequence and precisely in the time due. There was Babylon, there was Sparta, there was Ancient Rome! And Julius Caesar was assassinated in the Senate up to a second exact at the same time as he should have been.

"But that he was killed with a laser... Well, who cares about that!"

# Intent vs. crime

Basically, if someone commits a crime in the future depends on your time travel model. They might never commit it, because the past has changed. Or the future might happen just as predicted / evidenced, but with errm... "minor changes".

Next point: Your "preemtpive agency" can just point fingers at people and say that they will be horrible manslayers in the future – even if they do not have any evidence.

As also others have noted here, it's all arbitrary and not conform with a court and justice in our current understanding. An attempt to extend our current view might to appeal to the notions of criminal intent which should be there even without the criminal act. However, exactly as the line of thought of the lawyer in the short story above, an intent alone should not be punishable.

Let's just hope that this idea would be more successful that that lawyer.

Frame challenge - It can't be done

We can't even predict the weather 1 year ahead let alone a crime that is going to be committed in 20 years. Your AI is effectively an omniscient god. You should change artificial-intelligence to magic.

If it predicts that someone will commit a horrendous crime when they are 20 years old, there are two things that can happen:

1. The person does commit the crime. In which case the prediction was correct but no-one was able to stop it. This makes the AI worthless for stopping crimes.

2. The person is locked up for life and doesn't commit the crime, in which case the AI incorrectly predicted the future.

Basically your AI has to correctly simulate everything that happens in the nearby universe for decades ahead but also include everything that could change in the meantime. It has to predict the "free will" of every single human from now until then plus attacks by animals, future inventions, cosmic events such as Earth being destroyed.

If the AI could do this, it is effectively a universe in miniature. There are all the paradoxes of time travel here plus a few new ones.

• your two points is the logic behind imprisoning someone for attempted murder; they didnt really do anything, just waved a gun around and yell some and now they get thrown into jail for a long time. Sep 28 '20 at 18:05
• @Ceramicmrno0b - what you describe is "criminal threat". Attempted murder requires action with intent to kill has already been taken, but was unsuccessful. Sep 28 '20 at 20:28
• Literally every aspect of Sci-Fi that is not currently existing in our world is "Magic". Just because it can't be done according to our understanding doesn't make the thought experiment (and whatever medium this experiment is used for) unworthy of exploration. This answer adds nothing to the conversation. Sep 28 '20 at 22:43
• @GreySage - There are limits to SciFi in my opinion. That we could have an AI that simulates the universe it is in but faster than the real universe can do it seems unlikely. It also has to simulate itself simulating itself because it too is part of what it is simulating. My real objection however is that the AI can't win. If it correctly predicts the crime then the crime will happen so it hasn't prevented it. If it incorrectly predicts the crime it is in error. It is therefore either wrong or of no practical use. Sep 28 '20 at 22:57
• "Attempted murder" means they've actually tried to kill someone but failed or prevented from following through; basically, the victim lived. It's not just waving a gun around, it's actively having tried to shoot the person with the gun. Sep 29 '20 at 18:04

You will either face bigger problems, or you won't have to.

In almost any democratic and modern country, it is unconstitutional to arrest people who are innocent. Innocent people are those who have not committed a crime yet. (Please note here that both conspiracy to commit a crime and attempt of a crime counts as crimes.) As such, even if you know that someone will commit a crime 10 years in the future, it is completely unconstitutional to arrest and detain them.

This is because we, rightly, judge that we cannot punish people for things that they have not done. Furthermore, this is an affirmation of free will. If you can predict that a person will commit a crime in the future with absolute certainty, that would mean that the said person has no such free will, and our morality will have to undergo major shifts.

Under the case that the laws of the society have changed so much that we allow for the arrest of people who have not committed a crime, the laws of societies would have changed so much that our justice system would be completely remade. There is no need for independent proofs for the future criminality of a person. The AI's judgment will suffice.

After all, the law that allow you to arrest people for future crimes must be based on your showing capabilities to completely accurately predict future criminals in the first place. As such, once such laws have passed, you will no longer have to justify yourself.

Catch them red-handed

Since your AI can predict crimes, you can avoid the constitutional problems simply by arresting the criminals when they are putting the crime into motion. Since they have committed the crime, they can now be charged, and any proof should be easily obtainable.

Person of Interest

A TV series, Person of Interest, has nearly the same premise as your question.

From Wikipedia,

The series centers on a mysterious reclusive billionaire computer programmer, Harold Finch (Michael Emerson), who has developed a computer program for the federal government known as "The Machine" that is capable of collating all sources of information, to predict terrorist acts and to identify people planning them. The Machine also identifies perpetrators and victims of other premeditated deadly crimes; however, because the government considers these "irrelevant", Finch programs the Machine to delete this information each night. Anticipating abuse of his creation, Finch included a backdoor into the Machine. Tormented by the "irrelevant" deaths that might have been prevented, he eventually decides to use his backdoor to act covertly. To escape detection, he directs the Machine to provide him with only a tiny fragment of its data: the social security number of such a "person of interest". This may be a victim, a perpetrator, or an innocent bystander caught up in lethal events. The first episode shows how Finch recruited John Reese (Jim Caviezel)—a former Green Beret and CIA agent, now presumed dead—to investigate the person identified by the number the Machine has provided, and to act accordingly.

Emphasis mine.

While the main characters are not government agents, they do show how police can theoretically handle the problem.

• The one thing in this case is that the Machine can't see into the future: several episodes, in fact, have the Machine fooled simply because it didn't have the right data, and in at least one case the team didn't have to do anything at all. Sep 29 '20 at 18:01

You can't.

Because there's a massive string of major and minor crimes that are instigated by an unknown person or group. The people committing the crimes are arrested, but the instigator(s) aren't due to the AI not being able to detect them, as they aren't committing any crimes themselves.

Eventually a detective discovers this conspiracy against the AI, but can't do anything about it, since it's somehow not a crime. The detective might eventually convince lawmakers to write this law, the instigator(s) get arrested, and a lot of "innocent" people are released. Thus continuing the dystopian future.

Or the idea of this real conspiracy somehow gets made public and puts the AI's decisions into controversy, as intended, and the whole thing falls apart as the AI starts instructing mass arrests as people start protesting, which "of course" turn violent.

Regardless of how it turns out, the instigator could be found out to be the maker of the AI, police that want the AI to end, or even high ranking police that want to keep their purpose/power, but unintentionally their plan is found out.

## Edit:

You could have the detective go around fighting to stop the instigators in the first story/book/whatever, while the detective eventually realizes the AI does need to go and feels bad for stopping the resistance/meddling.

In the sequel, you can show the damaging effects the AI had after it was allowed to run "unchecked", so the same detective starts the resistance (again), using the same or similar tactics as before, but in ways that make it absolutely obvious that the AI is wasting money, completely wrong, and draconian. I envision the AI being repurposed to find people employment they love and excel at to make society better, instead of being completely dismantled. But I'm a bit of a romantic, and I don't really like dystopian stories. ;-)

I realize this edit doesn't exactly gel 100% with my original Answer, but it's all just suggestions anyway. Hopefully something I said helps you out.

Leaving the political, and moral implications aside, as well as cultural, legal and moral questions on what constitutes a "criminal", and suspending my disbelief to take this premise as "something that just is, no questions asked", I would tether these predictions of future criminals (and consequently people who need to stopped as early as possible) to people who you want to stand out. Geniuses, innovators, scientists, artists, role models etc.

Assuming that you maintain a list of (for lack of better term) undesirables (the bad apples), you should also maintain a list of the good apples.

Each year, you make an annual prediction of people who will excel in their lives (including the date of when their excellence is realised) as well as those who will be criminals (and the date of their crime).

You encrypt this prediction and release it to the public, but not the key. (This is to prevent these people to be targeted either positively or negatively)

On the date of realisation you release the key, so that the public can validate that your prediction was correct.

Assuming that ALL identities are accounted for, and you don't miss a single one, you can prove that you can indeed predict the future outcome of an individuals life.

## Have it also make other predictions that people can choose to ignore

During the course of your life you've gotten the following phone calls:

Accidents: "Hi, I'm with the Bureau of Predictive Intervention and we noticed that you are going to get into a car accident on the way to work tomorrow, but if you wait until 8:05AM to leave your house, you should be fine."

Financial Issues: "BPI here. I noticed you are going are going to file for bankruptcy in 2 years after quitting your current job... so you might not want to do that."

Pre-Marriage "Counseling: " marrying her will result in multiple cases of infidelity, treatment for depression, treatment for anxiety, bankruptcy, and divorce within the next 12 years. Oh, but the good news is that her you'll get full custody of the kids which you will be happy about."

Predictive Medicine: "In three months you will be diagnosed with terminal cancer, but if you go in for treatment now, it should still be early enough to treat."

People generally accept that the government has the right to force conditions on criminals, but not everyday life choices. Since all of these are non-criminal warnings they are non-compulsory. You can choose to ignore them and see what happens, and even once the accuracy is well enough accepted, people will still ignore it sometimes thanks to stubbornness, laziness, etc so it's accuracy will be pretty easy to prove based on the outcomes of non-compulsory warnings. Or better yet, they will not answer their phone and the prediction will reliably come to pass because they could not even accidentally avoid the problem with foreknowledge of it.

• There's a big chance that just by telling people about this, you'll be invalidating the cases. Especially in the case of accidents, where just by calling you've already delayed them enough that they won't be at the same place at the same time.
– Erik
Oct 1 '20 at 5:26
• @Erik updated the answer to address this concern. Oct 22 '20 at 19:45

One of the options mentioned in the answers is the possibility of time travel. This changes from a "preemptive crime fighting group" to a set of time traveling executioners. However, it simplifies justifying their work.

Rather than preemptively stopping crime, they wait for the crime to happen, the criminal to be found and caught. There is a trial and if found guilty, the court may issue a back-in-time death sentence. A guy from the future appears with a document signed by a judge stating that Joe Doe must disappear.This can range from a full case to, over time, a succinct "Killed several people", with people in older time basically just with the assurance that "there was a fair trial in the future". In any case, should there be corruption, it would be very hard to find out if any of those documents were falsified.

A middle situation between that and your setup would be that the "powerful AI" is just a receiver of such orders from the future. A lawyer requested someone to be wiped after birth and, after following proper procedure, it goes to your agency at the date of birth of that person.

The orders could be so concise that they only provide the name of the person to arrest (or their citizen id number). They don't really know why those babies deserve being arrested, just that a court in the future stated that.

Of course, they would be surprised if they found out with their perfect schema that they might actually be following orders from a dictator from the future, or simply a teenager that hacked the transmission channel.

• I'm not sure I'd happily let anyone mess with the past just to prevent some person from murdering someone. After all, whatever is changed could mean I never meet my wife or get born or who knows, maybe killing that child will result in people taking up weapons and rebelling, leading to mass death. But hey, at least we managed to prevent a single murder.
– Mark
Sep 30 '20 at 11:24

I will assume that the AI can't translate the knowledge it bases that prediction on into any form that is acceptable by human standards. For that makes the question self answering.

My solutions.

1

Test crimes.

Such a simple and easy solution to implement if the setting allows it.

Once a month a secret crime is made under the watchful gaze of the government. You can sing several thousand people nation wide to stage crimes to be set. You can eliminate all forms of surveillance easily and can connect them using secret means like intelligence style messing and cells work.

Having the group stage the crimes to be able to fool humans perfectly. This is important as our laws only work on the human level not the gods level. That is if I saw me neighbor through his window shoot his wife and heard the gunshot and saw her body collapse to the floor. This could be a crime but also they could be practicing a play. Calling the cops at that point is 100% normal and to the human understanding the mere possibility of other things, magic or acting or whatever, is a point proven after investigation so it does not matter.

Human perception and understand is what laws are built on.

The best part?

It does not matter if the AI predicted the crime or not. Simply ask it

How!!!

Simple. If it counted the crime then it has as much understanding as human observer which is enough to build a case on is our laws already.

But what if it did not count the crime? Then it much more smarter than humans and can even detect fake crimes.

Either way it is proven to be 100% functional and helpful.

Just as an extra point to address the issue of fake crimes. It would be legally argued, and I am no lawyer, that the truth of the matter that those people did not commit a crime and that the AI can be fooled. However a quick examination can be used to answer 1 question.

Did they people planned to seem like they did a crime beyond a reasonable doubt?

Because the systems, current ones, can't do anything beyond that.

Because that is all what the law requires. Not a single state on earth, or in history, can actually claim to do more or have done more.

Merely beyond a reasonable doubt. We hear all the time about a crime 20 years ago that is reopened and then the person thought to be guilty is then acquitted, whoops.

You can't then try to reject the system as even if you only prove that is like what we already have it has the added benefit of prevention the crime so it's like getting a 100% better computer with the same price. No reason to reject it.

2

Other predictions.

While not mentioned in the question I think this fits in more than letting other crimes happen as the AI might be hardwired not never let that happen or anything similar, not to mention the moral problem.

However if the AI can predict sporting events down the smallest detail, rain fall, actors film choices, the stock market, the path random people take to work, what youtube channel will be popular this year...etc.

If it can do all this it would be very easy for a law change to happen as basically this proves that the AI can actually predict anything.

3

Intervene just before crimes happen. Proving state of mind, intent, preparation for the crime...etc to provide for a court as more concrete evidence.

Lets take Marcus for example. Marcus is a low life scumbag who is in and out of correctional facilities and always in trouble with the law.

One days he decides to rob a bank. Now he gathers the crew, buys the weapons and tools, stakes the places, plans on the time...etc.

Then just as his own car pulls in the bank he is meet with a SWAT team, or whatever you have equal to that, and then he is charged with the crime.

Now this is a much much simpler case to establish in a court of law and the tweaking of the current laws wont be as high as before.

In your earlier example of killing at birth no law or constitution in the world would allow that. The philosophy is complicated but basically it would be a crime even if backup with this AI.

However if you wait just before the crime is about to happen this narrows down the variables to the minimum and insures that you have sufficient proof, legal sense, to convict with the ultimate conclusion of the AI.

This I think is vital as it narrows it down, again it is important since the law deals with actual actions not magic predictions, and coupled with the 100% truth rate of the AI it should be enough to change the laws so that prevention of the crime before the crime carries the maximum penalty.

Let me give you another example. If the AI see Sara killing Matt in his apartment on Friday the 12th using her gun after she takes a long lunch break from work and eats a burrito on the way to his place while listening The Melvins.

This is more evidence than claiming that Sara will go to Matt to kill him on the 12th. Imagine that they have an affair and she goes to him for a quickie all the time. The fact that she has gun does not equal intent to murder him, nor is she going there does as explained before.

So from a legal standpoint you want to limit the other factors and produce such a clear and obvious case. The needed proof from criminal cases is higher than civil cases and to trial Sara for murder without as much damning evidence as possible is very tricky from a legal since.

That is why if the AI can predict her course of action and words and all that it can be safely assumed that on this occasion indeed she will kill him.

This is flawed to an extent.

I can't deny it. The bank robbery for example is easy as it requires automatic rifles, just for example, and planning and other people...etc.

But say a guy gets drunks 2/7 days a week for years. Then one time he takes a knife and stabs his father. This is insanely complicated because he was doing just about everything the same exact way before but just this time decided to do the murder.

I'm sure that actual lawyers have more interesting thoughts on it. Thought I'm not sure that the philosophy of the law is what you are looking for or not. Anyway hope it helps.

• I would disagree with your "In your earlier example of killing at birth no law or constitution in the world would allow that." There is enough historical precedent to that. If we use the bible as reference (whether its events are historical is off-topic in my comment), that is how the Moses story starts. And then we have Jericho and the end of the Romanovs. The reasoning here is if you do not kill the son he will grow up and avenge the father. Sep 28 '20 at 3:12
• @raubvogel, I mean we are talking about archaic stuff that is several THOUSANDS years ago. Not even medieval or anything similar. The Romanovs! What do you mean? They were killed as part of the revolution and as a part of struggle of power. Not under a law. Not different from a pretender to the throne killing another in battle or captivity. Even in the bible I'm sure that it is viewed as an absolute evil act to murder these children. Besides was it part of the law or did the ruler just decide to do so? You have to provide evidence of legal precedent then. Sep 28 '20 at 3:40

You would need to get the information of its reliability by copying the AI into a controlled environment, and then extrapolate its validity to your situation.

Therefore you would have to set up a copy of your AI, but instead of a "live-feed" you would have to feed it with historical data. So, as to say, next to your "present and future-AI" you would have a setup that uses the same programming and reasoning, but uses them on historical data - a "past-AI".

You could check the validity of the AIs predictions on existing data, and thereby validate its "present and future"-counterpart.

• And where do you get the historical data? From all the documentation on each and every living person, homeless wanderers included? The more people you miss or get incomplete or even false data on, the more problematic it gets. In the end, your AI will probably predict things wrongly a lot in that simulation, even if it has a 100% correctness in reality.
– Mark
Sep 30 '20 at 11:30

In this scenario, other than what the AI says, there is no evidence for the crimes, since they have not yet happened. You cannot prove a negative

But that would be unimportant, because further evidence would not be necessary.

The word of an all-knowing future-seeing AI would be sufficient evidence with which to charge and convict an alleged perpetrator.

Since the AI is all-knowing it knows everything and cannot err.

Since the AI is future-seeing it has directly witnessed the crime.

Therefore there is no basis for doubting the information supplied by the AI. The AI not only has perfect knowledge of the crime, the AI has been an eyewitness to the crime. Extra evidence is worthless.

Everyone is jailed

Nearly every country on earth has complicated enough laws where nearly every person will inevitably commit a crime at some point in their life. Jaywalking, speeding, throwing away junk mail with someone else's name on it, inadvertently mispaying taxes, connecting to unsecured Wifi, etc. Even if you think you are safe as a child, in many places if someone under 18 is carrying a sharpie marker in public, they are breaking the law. Or hey, even singing Happy Birthday without paying royalties.

Without the nature of the crime being committed, it is a ridiculous notion.

The short answer to your question, like others have shown, is of course 'they can't'.

But, since we want to help the agency out here, let's see what they CAN do.

About time travel paradoxes. Knowing the future changes the future. This should mean that the image one gets of the future is one that changes AS you witness it. Which is weird of course; this should be a trippy effect at best, but paradoxal at worst. Let's for ease's sake say the AI knows how to handle this. Let's say he takes a snapshot of one moment's potential future and stops looking before it makes any decisions on it.

A person is accused of a future crimez, and is detained.

A person may be guilty of a future crime, but as soon as this is known, this may change. As soon as a person is detained to prevent their future crime, and the future of this person is AGAIN checked to see if they will commit a crime, they may turn out to be no longer guilty. Does this mean this person has to remain detained? Do we not owe an innocent person to check the future again to see what would happen if we set him free again? And if we decide to give him that chance, how often do we recheck? Is he gonna be up for probation only as often as prisoners of today are generally? Doesn't seem fair, especially if the matter of a recheck is trivial, for this OMNISCIENT AI.

So we get to the part that others also mentioned; what if detainment wasn't necessary in the first place? What if other actions - cheaper, easier, fairer - would do the same trick? Given the AI's powers, it wouldn't be too hard for it to figure out what would be the most efficient 'action' to reach the best result. It might need to learn how to get there, but isn't learning one of its fortes?

Like someone else mentioned; a simple phone call to the parents of the future criminal might be enough. IF SO, THE AI WILL KNOW AS SOON AS HE DECIDES TO CALL THEM, IF HE CONTINUOUSLY RECHECKS THE FUTURE. See, it might require more involved measures, but my point is; the AI will find out what it is.

All you have to do as a writer is decide what kind of stuff the AI figures are the best actions to take.

So I guess my conclusions is; leave it to the AI. Don't go around detaining people because that's the only thing you can think of that will help. It may very well be utterly useless, or even worse; too fascistic a course of action may cause unrest among the people, and an INCREASE in crime.

Also, like others mentioned, having more than one AI for the job, and having them compete with each other, would be opportuun to a point that it would seem implausible if it didn't happen.

I think it depends on what is required to "prove" it: there is the proving in scientific sense and in the layman sense to mention just a few. You can argue that "juridically" speaking it has to be proven beyond the shadow of doubt, but what that means depends on the justice system in place and what it defines that to be. There are laws in the United States as of today that are based on look and feel, what is "perceived" to be the truth (the "everyone knows" in Vox Populi), or someone's biases instead of scientific/engineering/medical/mathematical proofs. That has been seen in other countries throughout history, be it in the Victorian England, Nazi Germany, Pol Pot's Cambodia, Conquistadores in Central America, Galileo's Europe, and so on.

In Mein Kampf is argued that if you repeat a lie long enough claiming it is the truth, the average person will believe it to be the case.

Now, you need to establish in the public's mind that this AI predictions is accurate. Tests are not good enough: tragic events you can show were predicted by it but were not act upon (i.e the "we had the technology to avoid this tragedy but the politicians chose not to lift a finger! How long must the blood innocent babies wash our streets until we pass this law?") are very effective in moulding the public opinion. This needs to be built until the government approves using the AI system and preemptive crime fighting.

Next step is to show number of crimes is decreasing; it is not that hard to claim that is due to your crime fighting elite group. It is all about perceptions and controlling them. For instance, in our current age Facebook "experts" and twits by celebrities have more weight to the public mind than carefully laid scientific tests; it is the emotional vs logical argument. As time goes by, this will become established fact in the mind of the people.

There could be multiple competing implementations of these AI systems. After all, if one was created, couldn't there be more?

• world governments infiltrate various government agencies and try to replicate the AI, with mixed success
• open-source hackers construct multiple similar competing systems
• the original creator of the system had a disagreement with government policy and has gone into exile, producing another AI system that he claims is superior

Each AI may have slightly different programming or data sets. What happens when the various systems don't provide a consensus is up to politics. Do people trust in the government's system as more accurate, more precise, more trustworthy than these imitations?

• A jury of AI constructs. Interesting. But how do you prevent jury tampering? Sep 30 '20 at 3:37

Your premise has problems deeper than the question.

# Case 1/2: The future can be changed --> Machine crashes

If the future can be changed, the AI itself will play a massive role in it. Maybe the gonverment would try to do what mjt suggests in his answer. That will change the future. So, the AI will have to include itself in the calculations. This results in an infinite loop, ie crash. The machine will never return a result.

# Case 2/2: The future cannot be changed --> Don't even care

If the future cannot be changed, there is no discussion to be had. Everyone will do what they will do. How they will perceive their fake freedom is up to you.

Note: it is important to make the distinction that even if our very real lives are indeed fully deterministic, we have no such AI to see our future - and try to disprove it. If I went in front of the machine and start thinking of raising a hand, and the machine predicted I would raise my left, I'd try to raise my right. If I raise right, we are in case 1. If I raise left, well, I have no clue how this could have happened, what I'd feel and so on, so that's where your editing freedom comes.

• An infinite loop causing a crash is poor programming. It's not that hard to design an algorithm to search through an infinite search space without crashing or running forever (it wouldn't search through the whole search space though). It does imply that the AI's calculations would sometimes return a result that is only the best it found, rather than the best possible. Sep 30 '20 at 16:21
• Due to the nature of how it is used, thought, the best it found will very probably be useless, not like approaching a converging series. Example: iteration 1 decides subject will be a criminal. Iteration 2 considers the fact that after seen iteration 1, gonverment takes preventative measures and subject does not become a criminal. Search stops there. Result is subject is not a criminal. This is exactly the opposite of the truth. Sep 30 '20 at 16:35
• "subject is not a criminal" - great! That's exactly what they want! Sep 30 '20 at 18:48