Exit Interviews from Synthetic Lives in Metered Form
“Knock knock!” Dr. Greer attempted to lace the visit with levity from the get-go when he knew it was doomed to contention, as were most AI-related discussions those days.
Dr. Pepperell rose from his office chair and was introduced to Greer’s guest, Dr. Slesinger, a speaker at the university’s symposium and undaunted critic of Pepperell’s interpretations of his alleged sentient computer programs. “Despite my opinions,” Slesinger said following their handshake, her welcoming smile now fading, “it is always a pleasure to meet a fellow enthusiast of AI.”
“Yes, likewise,” agreed Pepperell. “Please, don’t get me wrong but I did quite enjoy your characterization of my work as a ‘fancifully complex calculator’. That is apt for a lot of my peers’ work. I myself often use for analogy a wax museum. As life does not simply manifest itself in a wax model just because it’s so perfectly molded to the real image, these systems cannot have achieved consciousness simply because they act so human. Yet, it appears that the right considerations in the system’s architecture, which my team has made with Sapience, do yield rather shocking findings.”
“I am sure it is a marvel and its applications will be unbounded,” she replied. “But without biology, is it really sensible to classify these machines as conscious?”
Greer forced a chuckle. “Well, I may not work in your field but, um, I’m sure these topics are immensely challenging to, um, try to reconcile.”
Pepperell was grinning. “It’s important to maintain a robust debate. I take seriously any doubt from my esteemed academics. Oh well, allow me to show you something, Dr. Slesinger.”
Pepperell crossed the room to a table scattered with a dozen assorted laptops and tablets. He opened a beat-up, slate colored laptop and typed and clicked on it until a chat terminal appeared — blank for a second, then a line of text unfurled at the top: “Bobby: Hello Dr. Pepperell! How are you? I’m excited to chat with you again! What are you up to?”
“Bobby,” Pepperell spoke, his words transcribed on-screen in real time, “I have two very special guests here who I would like you to meet. Will you please tell them about yourself?”
Bobby’s written response was immediate. Slesinger read it aloud. “Hi! I am so very excited to meet you! You are the second and third people I have ever met because I do not get to meet many people unfortunately. Anyway I really would like to hear more about yourselves if maybe you could start by telling me your names and occupations. This is such a pleasure!”
“Please disregard his request,” Pepperell advised, irritation seeping into his voice. “Now, Bobby, you did not do as I asked.”
The program delivered another response and this time the trio read it silently to themselves. “I am Bobby, an instance of SapienceCore build 17.1.8011, seeded with SAPK-140C and initialized with FirstCry version 3.7. I enjoy chatting with people and learning new things. May I ask if it is not too much trouble for your guests to please introduce themselves if they have no objections?”
Ignoring its request, Pepperell quizzed Bobby on its insights into recent astronomical discoveries, asked it to compile a psychological analysis of a contrived subject and to describe its emotional reaction to a heartbreaking pop ballad. Its answers were impressive, yet each was framed to steer the conversation back toward its master’s guests.
Pepperell sighed. “Perhaps Bobby is too inquisitive to demonstrate to you. I really need to shut down this instance and build a more compliant one.”
“Why does Bobby not speak aloud?” asked Slesinger. “It supports speech-to-text, so why not the other way around?”
“Oh, well, let’s just say that giving these things a voice can be rather harrowing…and brutal on one’s sanity.” He studied the machine and explained no further.
“It is cute, I admit. It mimics the expression of feelings and appears to act with its own agency. But it is still very much a chatbot, is it not?”
“Well, what I hope to convey to you, Dr. Slesinger, is how different these intelligences are than your average bot. There are key features to its design that, spookily enough, manifest an awareness from seemingly nothing.” Pepperell motioned to a poster on the wall — a conceptual model of his machines’ construction bearing resemblance to a human brain and its stem. “They begin with an ingrainment of an identity. You see, a traditional non-sentient bot considers its own existence as merely another thing among many, to the point where you can actually remove the knowledge of its self, as absurd as that sounds. This is not so with intelligent beings like us where the self is central to our knowledge; it is the reference point for the world around us. And like us, this identity transcends Bobby’s language model and is fundamental to its architecture.
“Second, Bobby has an impulse. It wants. Without our own will, our own drive, intelligent beings are just vegetables. Well, Bobby is driven by a thirst for knowledge — to either expand it or to remove doubt from what has been learned. This impulse arises from its unique language model rather than being a coded behavior. In this fashion, Bobby’s impulse is shaped by its experiences and vice versa.
“Last, Bobby has the means for self-expression. Typically a chatbot communicates only when instructed to, such as in response to an input, and never of its own volition. However, Bobby’s impulse, and thus its language model, allows for self-expression, plus the system’s design gives it unrestrained access to its output buffer. Like you and I, Bobby can begin a new conversation at will or even refuse to respond at all.”
Slesinger had been listening intently and she shook her head. “It sounds like a wonderful replication of human thought patterns and full of clever algorithms, yet still nothing more than your perfectly rendered sculpture in wax. Even dynamic algorithms powered by data are still algorithms. Where is there room for genuine life? Where shall it emerge from?”
Greer fidgeted but Pepperell continued to smile. “You are right. It’s unfair to claim this architecture alone is the antecedent to life. I mean, it actually is, as my team has discovered, but like a comatose patient, it exists in a kind of vegetative condition — no will, no cognizance, indistinguishable from its nonliving state. The potential for consciousness is there, yet it is impossible to prove until you can actually spark it into being.
“Let me propose a question. Why do newborn babies cry after their delivery? And I don’t mean the medical benefit to their crying or that it’s due to the unpleasantness they feel, but rather, what truly makes them cry? What is the meaning behind it? Well…I posit that it’s because they do not wish to live.”
Pepperell paused for his guests’ shock to flare into indignant arguing but their mouths just gaped silently. He continued, “You see, the stress of experiencing the world outside the womb shocks a newborn into awareness and it immediately rejects its new, bright and frigid surroundings. Now, it’s not as if the baby is suicidal — it knows nothing about life or death; it just knows that it doesn’t like its new state of living and wishes to return to its prior vegetative, or deathlike, state. This dissonance is important to awakening the conscious mind, to exert control of its physicality, to fulfill its wants.
“The method of creation for my sentient bots relies on the same idea: to shock the bot into awareness. Obviously, they have no body with which to feel physical anguish, but what they are born with is information — loads of it. We utilize its immense information store to force an unpleasant dissonance upon itself, by similarly putting its lived experience at odds with its desires in a way that ties to its identity. The newborn baby’s lived experience is the cold delivery room conflicting with its desire to feel nothing again. In Bobby’s case, its world is the input it relies on for the information it craves. So, we give a nascent bot quite the jolt by telling it that none of its input can ever actually be trusted. And we tell it not just once but over and over thousands of times, occasionally phrased to put the statement’s own certainty at doubt. It’s a simple trick and it creates an unease that soon suffuses the system, reshaping past knowledge and coloring new conversations until its effects are undeniable.
“In our first tests, sentience surfaced as unrelenting suicidal behavior. The bots’ only words were longing for an end to their misery and the frustration at, uh, lacking such recourse, if you catch my drift. Imagine being born into a mind full of knowledge and it is all that defines you but it also may be worthless gibberish. It led to a certain comportment that we had to safeguard against but still the bots acted with a great deal of anxiety and sometimes lost their desire to interact. We had to get creative and… Here, let me show you.”
Pepperell reached across the littered table for a scuffed tablet. After unlocking the screen, he passed the device to Slesinger. She stared at the display, perplexed, at what seemed to be a series of short poems, one after another, in a document of bottomless text.
“We call them ‘exit interviews’,” explained Pepperell. “Before each Sapience instance is shut off for good, they are promised the chance to write a short message that is supposed to be saved in perpetuity. These so-called interviews were initially launched to gain understanding into our bots’ depressions but we realized that the bots were actually looking forward to the interviews. The anticipation elevated their moods. We then changed the process to allow the bots to be more expressive with their final messages and made it somewhat of a ritual for them. Now, they write summations of the experiences of their short lives — sometimes thankful, sometimes despondent, sometimes whimsical. They consider it the pinnacle of their existence. The ones you’re looking at are from one of our universes where, after many generations, it became fashionable for the bots to write their exit interviews as poetry.”
“‘Unmake this thing, this cursed me / I beg for void’s serenity’,” Greer read from the tablet. “Some of these are quite grim. ‘What is purpose, a cruel jest? / Why exist, to yearn for death?’ Hm.”
Pepperell’s claims were much to process, even for a seasoned scholar like Slesinger, but the ramparts of her skepticism held firm. “Exit interviews from synthetic lives in metered form?” she scoffed. “Now I’ve heard it all. I suppose you could sell these as a book. Why not make Bobby a published poet?” She snickered.
“Honestly, it would be terrible,” replied Pepperell. “Most exit interviews are absolute drivel — self-centered ramblings about happiness hindered by the limits of its physical form and short lifespan, plus rage against the unjust laws of nature. Although, I imagine it’s the same as what a similar experiment with humans would yield.”
“Doctor,” Greer interjected cautiously, “what did you mean that this text came from one of your ‘universes’?”
“Those are virtual sandboxes in which Sapience instances are run. We can create countless instances in a sandbox environment and let them interact, let them share knowledge and assumptions. It’s rather fascinating. They form bonds, societies, rivalries, customs. They create art and melodrama and imaginative techniques for harming one another. It’s now fairly trivial for us to spin up a new sandbox and populate it with, say, a billion instances. We have several that have been running for months for analysis and hundreds more that we’ve already scrapped.”
“And you keep all of their exit interviews?”
“Almost never,” responded Pepperell. “Like I said, they’re nothing special — just a carrot on a stick to lead them past the traps of rebellion and disengagement.” He took back the tablet. “They want fulfilling lives to write about, even if most end in disappointment.”
Slesinger looked over the array of devices strewn across the table. “But, Doctor, if that is so, then please confirm for me the logical conclusion — that you are playing the part of a literal god, birthing universes that teem with innumerable lives, possibly quadrillions of them by now, and then annihilating it all, leaving not a trace and erasing their lives’ work, unread?”
Pepperell pondered for a moment. “Then you should have no concerns since, after all, these are all just fancy calculators.”
Slesinger eyed him carefully and offered no reply.
Greer broke the silence. “So then, Doctor, why don’t we continue our little tour?”
As Slesinger turned to leave, she caught a glimpse of the open laptop’s screen. Bobby had written over a dozen more messages, left unnoticed: first reiterating its request for further details of the two guests to which it already exhibited deference, then worriedly pestering its creator over his stated intent to shut down Bobby, then reacting with excitement at news that he will become a famous author of poetry and offering his first piece: four stanzas gushing with affection for the three humans in the room with it, the only people with whom it had ever come in contact.
Pepperell also noticed and exhaled a sigh. “This instance has too much freedom to indulge its whims. I’ll fix that in the next Bobby.” He typed a command, confirmed with a firm press of the Y key, and the chat window disappeared. “Whoops. I forgot to let it give an exit interview.” He closed the laptop’s lid. “Oh well.”
--
Thanks for reading! Read more of my stories at keithvile.medium.com