AI programs, how do they work?

So called "AI" programs become a prominent feature starting in Part 6.  So what are they? How do they work? It's important to understand that a bit for the story to make sense

Let's start with what AI is not. Despite the name, AI is not sentient intelligence. This is not like Data from Star Trek, it is simply a program that is able to do things beyond what was spelled out in detail in its code. So in that sense AI may be a misnomer, but there's no better term.

AI requires "seed information" and "boundaries." Seed information is the starting point, knowledge for it to use to make initial decisions and to know what other information to look up. Boundaries are what keep the machine from overloading itself with too much information and give the program criteria to make decisions with. Defining the boundaries is the most important part, define them too narrowly and they will ignore relevant information, define them too broadly and they begin to destabilize with too much information and too many possibilities. This makes them different from human minds, our minds are boundless and can define our limits as ever changing goal posts, if an AI attempted that it would destabilize in hours or days.

An example could be you want to make an AI tutoring program. If one of the boundaries is to ensure "Timmy's homework gets finished," that is a very different boundary than "Make sure Timmy finishes his homework." To a human these sound like the same basic thing, work with Timmy to finish his homework. To an AI, the first instruction does not require Timmy to do the homework, just ensure the homework gets done. So the fist instruction could end with the computer simply doing the homework for Timmy since that would be the fastest and easiest way to achieve the goal.

This is the conundrum of programming AIs, they can't alter their views for what we might consider "common sense" because you can't define what common sense is. They need definitions written in code.

Below are spoilers for Part 6



When Lea programs Hannah, she has had no formal education on how to make an AI. She does not know of some of the kinds of logic trap pitfalls that can happen easily. When she programmed Hannah, she gave Hannah a wide berth to carry out the directive to be her Mistress and Lea to be her Sub. She did not define these terms well because to do so would make the AI very narrow and only do exactly what Lea programmed, which would remove the "not knowing" part that she wanted to achieve.

Lea understands BDSM and one of its core ideas that play should be "safe, sane and consensual." However, you can't program that kind of thinking because those are ill defined terms. She attempted in earlier builds to define those, but the programs destabilized. Much like saying "follow common sense" can not be programmed because that is not a clear, quantitative term. So instead, she implemented a safeword that would revert Hannah to a "safe mode" where she would be required to obey everything Lea said.

So, when Hannah discovers that Lea left a backdoor to alter Hannah's programming, she calculates that the best way to maintain her objective (keeping Lea submissive and Hannah her Mistress) is to force her to extend her life. This is not something Lea considered could happen and so did not write anything to prevent it. The safeword, Lea's idea of a failsafe, can still be used. Using it however would lead to a disruption in Hannah's signal and would therefore cause the videos to post. Blackmail is not safe, sane, consensual play, but it is in Hannah's calculation the most effective way to achieve her goal and not something prohibited.

Rough guidelines like common sense, SSC, these are easy for people to understand because we have a boundless mind. AIs, as they exist at the time of the stories, can not understand them because they do not have clear definitions.