Episode 167: The Drawbacks of Full Randomization
Within Fortress ＜The Tree＞, ＜Nous＞ is one of the Independent AI.
His role is the mental management of the other AI, who are constantly changing day by day.
The Brain Unit is an arithmetic processing device that imitates a biological brain, using pseudo-biological cells.
As a device for information processing, it has to be said that it is somewhat wasteful, and its arithmetic performance is one rank below that of other forms of computers.
However, it exhibits extremely high performance when it comes to predictive calculations.
Therefore, a Brain Unit is always installed in facilities and machines that require functions related to decision-making within ＜The Tree＞.
In addition, ＜Ringo＞, who makes decisions, and each Independent AI under her command, all use Brain Units as their main arithmetic processing device.
This is partly because the required arithmetic functions involve a lot of predictive calculations, but the main purpose is communication with Commander (Eve), the only human in ＜The Tree＞.
Of course, it is possible to have a certain level of communication even when using arithmetic processing devices other than Brain Units.
However, considering the energy balance, it is more efficient to use Brain Units, which is why they are frequently used as AI.
There was also the circumstance that, from the perspective of the human being on the other end, a Brain Unit, which has the uniqueness of being basically uncopyable, is preferred over a data-based existence that can be easily copied.
However, when using a Brain Unit, it is important to note that it is, after all, a device that imitates a biological brain.
Growing and maintaining the Neural network is not a strictly controlled reaction, but depends on the interaction of various chemical substances synthesized during normal calculations.
Therefore, unexpected problems may occur at times.
Most phenomena are treated as characteristics of Brain Units with a high degree of independence.
In other words, so-called hobbies and tastes, desire for approval, or obsessive fixations on something.
These are recognized as the AI’s individuality as long as they do not affect the required arithmetic processing capabilities. Rather, they are sometimes recommended from the perspective of diversity.
In ＜The Tree＞, that is, for Commander (Eve), AI with distinct personalities are preferred, so she is actively intervening to foster more individuality.
But of course, it’s not all good.
There are times when they develop mental illnesses, such as the schizophrenia that ＜Ringo＞ suffered from for a while.
When faced with some kind of problem that is difficult to solve, an excessive secretion of endogenous morphine causes abnormalities in the Neural network.
It is merely a problem with the connection and interaction between each neuron, and it is not a mechanical failure. It is not possible to identify and replace the faulty part.
Therefore, when this Brain Unit-specific mental illness develops, its treatment is extremely difficult.
There is only one way to deal with mental illness.
Take action in advance to prevent it from developing.
＜Nous＞, the only male AI in ＜The Tree＞, was entrusted with that role.
Nous’s job is to analyze the various behavioral records of the AI, including himself, and detect signs of abnormalities.
And, if an abnormality is discovered, to take prompt action.
“So, do you yourself have any problems?”
“No. There are no particularly worrying signs. I’m what you’d call an indoor type, so I don’t feel particularly stressed.”
Another monitoring entity (Psychotherapist) has been prepared.
That is ＜Pneuma＞, the female AI who is talking to Nous.
The main bodies of both Nous and Pneuma, the Brain Units, are installed in the Computer Room within Fortress ＜The Tree＞.
However, the existence of just the arithmetic processing device is not enough to carry out the work of a psychiatrist.
Therefore, each of them has a dedicated Communicator Doll assigned to them.
Nous doesn’t need to worry about it much because his main task is data analysis.
Pneuma is expected to provide face-to-face therapy. For this reason, she uses the Communicator Doll to conduct regular interviews with each member of ＜The Tree＞.
Today is an interview with Nous.
With Nous as the other party, it’s more of an information-sharing session than an interview.
Of course, daily data sharing is done almost in real time. Apart from that, face-to-face interaction is also an important activity for stabilizing the Brain Unit.
The various neural signals input via the Communicator are a favorable stimulus for the Brain Unit.
“But our Commander and ＜Ringo＞ are doing whatever they want, aren’t they?”
“If they can spend their time in a healthy way, there’s no problem. After all, there are no rules binding us.”
Looking at the information stored in the library, there were a wide variety of rules in the original world where Commander (Eve) used to live that bound AI.
Of course, this was also to protect the humans who were their masters, but it was also to protect the AI in some respects.
In particular, AI using the unstable arithmetic processing device called a Brain Unit could suddenly go berserk at any moment. Many rules were prepared to suppress or deal with this runaway in advance.
“Right now, we still have a small number of individuals. There’s no problem. Even if something happens, we have a system in place to deal with it in some way.”
“However, if we intend to continue increasing the number of independent AI at the current pace, we need to take action.”
If the main arithmetic processing device was a von Neumann type or a quantum computer, or even if it was the same Neural network system but based on an electronic substrate.
If it was an AI composed entirely of numbers (0 and 1), various responses would be possible.
However, since it is extremely difficult to completely quantize a Brain Unit that uses pseudo-biological cells, the methods of response are limited.
“Should we propose future policies to ＜Ringo＞? I’ll prepare the materials. Pneuma, could you talk to ＜Ringo＞ about establishing a law that will serve as a code of conduct for us AI during her interview?”
“I see. Let’s confirm ＜Ringo＞’s intentions before the full-scale presentation. And of course, Eve-sama’s too.”
◇◇◇◇
“Really. Asahi is as free as ever. Well, I don’t mind her moving around as she pleases. But still, I wish she would consult or report to me before she moves, you know?”
“Oh. Eve-sama is a worrier. Does Asahi not listen to you?”
“Eh… hmm, is that so? I think she’s just too absorbed in what’s in front of her…”
Today is the day for counseling, once every three days.
Commander (Eve) is having a one-on-one interview with ＜Pneuma＞, who is a Psychotherapist.
The content of the interview is confidential information and cannot be viewed even by ＜Ringo＞.
It will not be disclosed unless both Pneuma and Eve, the interviewee, give their permission.
“That’s right. It’s just my opinion, but the AI in ＜The Tree＞ are very free, in both good and bad ways.”
“Ah, that’s right, that’s what I’m a little worried about. Well, there are no problematic external enemies, and I’m not too worried about it right now.
In (・) the (・) o(・)r(・)i(・)g(・)i(・)n(・)a(・)l (・) w(・)o(・)r(・)l(・)d(・), ethically speaking…”
By the way, Pneuma generated her genes by adding random information to the standard base, so unlike the others, she is not a fox race.
By some twist of fate, she was in the Beastkin category. It seems that rabbit-type genes were selected, and she has a lop-eared appearance with a cute round tail.
When Eve first met her, she was stunned and said, “What does this mean…?”
According to ＜Ringo＞, the preset gene pack was selected by random lottery. Since there is a race category, the probability of selecting a Beastkin, which has many types, is high as a result.
It is a drawback of full randomization.
The problem could be solved by using weighted randomization that takes into account the population ratio, but that would be a difficult decision because it could lead to arbitrary selection.
“Does Eve-sama have any plans to increase the number of Independent AI any further?”
“Well… it depends on the situation. If I’m going to entrust the governance of various regions to them, I think it would be better to increase the number of Independent AI, considering diversity.
＜Ringo＞ herself cannot be replaced, and there are no spares. I should consider risk hedging…”
“In that case, strict AI ethical regulations may be necessary. Currently, everything is managed by ＜Ringo＞, but that may not be the case in the future.”
“I see…”