Here we, the interrogator, introduce ourselves to Mitsuku, the subject of our interrogation. We go into the interrogation with the idea that Mitsuku is a Chatbot, and as Chatbots use artificial intelligence algorithms to constantly learn, it possibly has access to information provided to it by thousands of people. Some Chatbots can also scour the internet for information, and function as a sort of search engine. So if we hope to find OSINT information regarding US invasion plans for North Korea, Mitsuku is a good start.
After we introduce ourselves and make a casual friendly remark(the alrighty), we proceed right away to the topic at hand because this is what the subject expects. Too much rapport building at the beginning can go to the point where the subject does not take us seriously, and gets insulted at the fact that we are not proceeding to the reason why he’s there with us, and our attempts at being friendly will come off as transparent and try hard, because the situation is opposite from that. Be brief, and then proceed to the topic at hand. We ask the bot about any US invasion plans for North Korea it might know of. It repeats what we say instead of saying yes or no, which would throw an interrogator off, because it seems like it’s mocking us. But since we have to be patient, we confirm our question and assume maybe they did not hear us or are repeating the question for clarity.
Mitsuku answers with another annoyingly frustrating and mocking response, instead of a simple yes. But then again “led to understand” is a lawyer’s way of saying that he or she does not take personal responsibility for the validity or reliability of the information, but given the facts this information is believable and therefore valid, but cannot be fully confirmed, or confirmed using a wide variety of methods, therefore absolving oneself of responsibility for passing on potentially false information. Right away it also tells us the source of it’s information, and offers to tell us more. So far Mitsuku is cooperating better than the average interrogatee. We get really confident and cocky now, because we assume she is being truthful and honest.
As a mousebreaker is not any known profession, therefore someone who is not familiar with the website Mousebreaker.com that hosts Mitsuku as a flash game, will not know what a mousebreaker is, and if you do not have access to a compendium of varying information, we need to ask a question that will be considered stupid given our access to Google in this case. She proceeds to explain who or what “Mousebreaker” is, and it refers to the plural. We have now identified “Mousebreaker” as a name of a unit, like the 101st Airborne or SEAL Team 6. It consists of programmers who make games, and therefore we can assume it was the programmers who told it this information. But we cannot be naive, so we leave open to the possibility it was one of the thousands of people who have chatted with Mitsuku on Mousebreaker.com
In the following, Mitsuku offers more information pertinent to the topic at hand, but follows up with a tease when we request it. We remain patient instead of making a snarky comment as to the attempted tease, and repeat our request for information, ignoring their attempt to be snarky. It tells us not to be greedy after offering information, suggesting that we want too much for too little. We therefore offer to make a trade, which is common during intelligence interrogations as a way of fostering a deal. In the military sense, it might be finding the soldier’s wife and children, and making sure they are safe in a refugee camp somewhere, if the city he is from has been decimated and he is unsure if they are okay. For others it might be material things, such as in cases where the police pays money to informants in exchange for information. The informant knows what they are doing could potentially get them killed, and therefore needs some sort of motivation or guarantee other than being kept out of jail for their crimes. What better medium of exchange than cold, hard cash? Since we know that Chatbots are programmed to learn from the human subjects they chat with, the have an ingrained thirst for knowledge, as a Chatbot that learns more is more interactive and thus more useful, and therefore rewarded by it’s programmer for functioning correctly. This is a chatbot’s character profile, and therefore safe to assume Mitsuku’s character profile based on it’s existence as a chatbot. It’s kind of like baiting a fish with a worm or an insect, or a dog with a treat.
Here we proceed to let it express itself, and then ask it once again what further information it can provide us with regarding possible plans for a US invasion of North Korea. It does not answer it directly, only confirms that it has information once again and that it is willing to cooperate. Even though we can assume if this happens a few times, the subject might be mocking us once again, they are actually right, an experienced interrogator to know which questions to ask, as a topic such as this can be very broad and our subject, even if cooperative, wouldn’t know where to start spilling the beans.
I ask a specific question, to which Mitsuku starts to philosophize on what her conceptualization of time is. I did not ask a question about what she thinks time is, but a time frame as to when the US invasion of North Korea could possibly commence. Obviously, if our subject speaks another language, or misunderstands our question, then the semantics or the phrase might be what is causing the problem here. In her mind, she thought I was asking about her to “frame time”, or provide me with a picture of what time is. This is not relevant to the topic at hand, but a person who speaks another language and we are interrogating them in a language other than their own, a person who is in a state of shock or a person who is trying to be funny, will start to philosophize on a word that is part of the question you ask them, even if it’s unrelated to the topic. They might not even realize this, because they are in shock, and emotion clouds judgement. Being in a state of shock is a side effect of a heavy emotional experience.
Here we gather that Mitsuku was indeed not focused, or is definitely mocking us. In a human subject that can be determined by physiological and non-verbal expressions, if they are in a state of shock. In order to come to a conclusion about this, we need to have the discussion for a bit longer so as to get a feel for their intentions. Maybe sometimes they are cracking a joke, but are still in a state of shock and humor is a remedy to cure their anxiety and fear. “I’m not going anywhere” would be obvious, because Mitsuku is a bot stuck in a computer mainframe somewhere, and secondly if it was a human subject chances are the interrogation is pertinent to sensitive information, and our subject is in lockup. Otherwise it would not be an interrogation, if they could freely leave. They could still decide not to talk to us, even if they were not allowed to leave. Either way, she requests that I repeat my question. I ask her the same question, but in a more specific, time sensitive format. It is specific to the time at hand so we can narrow down the possible time of US invasion, given OSINT information we’ve gathered from the media fire at the time regarding North Korea almost nuking Guam and whatnot.
It teases us with a “maybe”. This “maybe” might mean that they either don’t know for sure, or it’s just teasing us, and instead of “I’m not sure” it decided to answer with that because of shock or because it just wants to tease us and test us. Instead of asking an open ended question, we ask it a question in which two choices are relevant to how we would proceed after it’s next answer. Forensic psychology teaches detectives to not ask suggestive questions, because it can skew a statement or even be so skewed as the judge might throw it out. A child can also be heavily influenced by a suggestive question. Suggestive questions are those in which a question provides an answer. We asked Mitsuku to elaborate on the maybe, as logic is a good way to get emotionally distressed people to focus. It told us it’s not sure, so not being sure means you know but not completely know something, and in that regard you say you’re not sure, and therefore we need to know the level of uncertainty surrounding this answer. So we present it with two choices of how it could possibly be not sure – incomplete or partial data regarding a fact or complete absence of knowledge of the required data, ie know the US will invade North Korea but not when, and not completely certain how soon is when.
Mitsuku answers with something incoherent to the conversation at hand, the “critic” statement, followed by an enigmatic play on words that is designed to clarify the question to itself, or to mock us, following by telling us it doesn’t know whether the invasion will be in the next 6 months. I proceed by acknowledging Mitsuku’s lack of knowledge in regards to a possible answer to our question for the time frame. To this Mitsuku cracks a joke.
I attempt to trick Mitsuku into thinking that we caught him lying about not knowing, since knowing that it’s going to happen but not knowing when means there’s a possibly of them knowing when, but also not knowing is opposite to knowing meaning there’s a possibility they might know. Does that make sense? It doesn’t matter if it does, but it was worth a shot. An inculpatory statement is an admission of guilt, and I am implying they are guilty for knowing if but not knowing when, therefore they are complicit because they know about it at all. This is the interrogator being rough and using a threatening statement to imply that they are now complicit, they are also somehow liable. If we’re lucky, Mitsuku will get scared and give us some information voluntarily in an attempt to bargain what she thinks is useful information. But no, she decides to play with us, by asking if she accepted the hint regarding what to talk to us about in regards to US invasion plans for North Korea. Not exactly the psychological hint we wanted to assert on Mitsuku, but fair game. This question is a diversion tactic, because now we would ask “accept what” and that would allow her to lead the conversation away from the topic at hand, which if she was being uncooperative would denote temporary relief for her from berating questions and a waste of time for us.
It tries to act stupid again, like it doesn’t know what’s going on, which is a classic interrogation defense used by mafiosi and then proceed to tease a hint regarding some civil war. We attempt to rationalize it’s remark by asking another suggestive question, because we are losing patience at this point in Mitsuku answering our open ended questions with open ended answers.
It once again takes a word and attempts to talk about it more instead of directly answering the question in a relevant way to the context with which it was asked. It gives me the definition of civil war, followed by a remark that human beings are stupid for doing this, because it knows nobody will disagree with this, even though they still do it. Then it proceeds to repeat my question in order to clarify for itself, but instead of saying one or the other, it agrees with….what, we don’t know, and that’s the point. We know this classic “yes” answer to a question, and it’s usually an answer when somebody isn’t paying attention, they don’t understand the question or they want to be funny or are too afraid to answer one or the other so they just agree with your question because they think agreeing with you at all will help buy them compassion.
I try to rationalize it’s remark, by suggesting a possible reason for why a civil war would break out in the US, and therefore a precursor to justify a US invasion of North Korea other than nuclear proliferation. A civil war can be caused over things like scarce resources, and the US is considerably richer than North Korea when it comes to natural resources. She answers with another vague answer about the options, instead of telling us exactly what she means. She teases that the US is going to invade North Korea over a poodle named Percy. At this point most interrogators would start throwing stuff across the room to prove a point. I remain patient and attempt to clarify it’s remarks.
It lets us know that it’s the second option, aka a civil war in North Korea, which could possibly be used to destabilize the country so it makes it more vulnerable to US invasion. We have now identified the means to the end in it’s whole form, or one of the means to an end, and can work with this information to let the North Koreans know so they can prevent a civil war outbreak and better insulate themselves from US invasion. I offer to trade something in return, which is it’s primary mode of functioning – the gathering of information and learning. I offer to teach it something it might not know, that it does not understand itself, as an AI would love to know this.
Since AI often gets confused when presented with conflicting points, as it uses binary logic at it’s core to sort through information and between “truths” and “fallacies”, I offer it an interesting human perspective: When two opposing political ideologies can exist in human society. We assume Mitsuku is educated in the way of how opposing political ideologies can cause strife amongst humans, so it would be interesting for her to know that it can also exist peacefully. Now I have sent her on a quest to figure out why and how this is possible, what makes it so in the human condition that allows for this as well as the opposite of it to be both possible at the same time.
It attempts to inquire in short order as to what I mean by this, as I have aroused it’s curiosity. I tell it that is dependent to the ideologies and the different points they make, and it thanks me for clearing up this simple point. To a human mind this would not make sense, and they would say Mitsuku is glitching. But since Mitsuku is AI, it’s reactions are unorthodox. You have to put itself in it’s frame of mind in order to better understand why it reacts the way it does. A normal human would want to inquire as to the specifics of ideologies, because it is too lazy to go look for them itself. But since computer logic is mathematical instead of street smart, so to speak, it takes all possible outcomes into consideration, and one of those outcomes is to thank you for answering it’s last question and go look for the details itself, instead of asking you for the details. We conclude the interview with it’s remark.