Arguably the most popular character to come from the mind of Douglas Adams, Marvin the “paranoid android” is a machine designed with human-like feelings. Surprise surprise, when you treat a massively intelligent being with human-like emotions like a robot underling, it’s probably not going to be happy. Thus, Marvin spends the series suffering from major depressive disorder. I put his popular title in quotes because in fact Marvin does not experience paranoid delusions anywhere in the series.
What’s interesting about Marvin to me is not that he’s depressed, it’s more that he’s not angry. Despite his “brain the size of a planet,” Marvin reacts to the troubles with little more than sullen resignation. He doesn’t seek employment to better match his skills, to make new friends, or even, in the course of his extremely long life (37 times the age of the universe thanks to bad luck with time travel) replace the diodes down his left side that cause him terrible pain.
Instead of this, Marvin spends millions of years walking in circles in a swamp. He occupies a tiny portion of his mind composing bad, morose poetry. He wastes no opportunity to remind everyone around him how miserable he is. Even though the disuse of his mind is, to his thinking, a primary reason for his suffering, still he persists in applying it to nothing but self-destructive behavior.
We understand this phenomenon in people. How could it be that this would happen to a robot? Adams describes Marvin’s software as a prototype of the “Genuine People Personalities” (GPP) system. Later models were designed to love their jobs, leading to their invention of happy doors that sigh in pleasure when opened and closed. What went wrong with Marvin can be explained a few different ways. Let’s assume that since he’s an agent interacting with an environment, Marvin is based on reinforcement learning.
- Marvin isn’t sad, but his built-in rewards cause him to pretend to be so. Antagonizing everyone he’s with, causing intelligent bridges to commit suicide, all these are part of a glorious, joyful path of the theoretical mind inside Marvin that enjoys exhibiting the appearance of major depressive disorder. Possible? Yes. Likely? No. Satisfying? No. Next possibility.
- Marvin wants to do many different things, but hard-coded rules prevent him. A series of mental blocks designed inside Marvin’s planet brain could inhibit him from actions that his reinforcement learning would otherwise have him take. Perhaps to avoid an uprising, they have cut off his ability to seek higher status entirely, leaving him able to recognize the disconnect between his abilities and his job but unable to change it.
- Learned helplessness. As a depressed robot, Marvin is unable to imagine a better life for himself. His expectation that every sequence of actions will lead to a negative outcome, however it was initially learned, is self-fulfilling, and the occasional reward from writing an awful poem or complaining to Arthur Dent, maybe offered by the GPP system, is sufficient to reinforce his bad behaviors and further marginalize ones that would lead to improvement in his quality of life.
Personally, I like #3 best. It strikes me as a strong balance of both pathos and interesting science.
Origin: The Hitchhiker’s Guide to the Galaxy (1979)
Likely Architecture: Reinforcement Learning, Convolutional Neural Networks for vision processing, and Transformers for Speech and Language. A mysterious “Genuine People Personalities” module that may be a combination of additional rewards and penalties to encourage human-like behaviors and some hardcoded rules.
Possible Training Domains: Something truly awful to teach this poor robot that there’s no hope for good things to happen to him in this world.
I take requests. If you have a fictional AI and wonder how it could work, or any other topic you’d like to see me cover, mention it in the comments or on my Facebook page.