Aki looked at the camera and spoke. “Hello, Natalia. I am Aki Shiraishi. How are you today?” She was not sure if the strange swirls of flashing patterns had changed.
“Wow! I think she likes you.”
“How do you know?”
Raul laughed. “If I knew that so easily, señorita, I wouldn’t need to use the ETICC’s supercomputer.”
“Don’t get sarcastic with me. What are all these monitors for?”
“The internal state. The monitors produce a visual representation of the exchange of pulses and field activations. I want her to teach herself how to communicate human principles, but only with pictures, never with words.”
“Has she ever said anything?”
“No. That’s why I needed the ETICC supercomputer. I figured her learning would speed up if—”
“Your monitors feed back into her camera for Natalia’s development of a dynamic internal state?”
“It loops her field activations and a visual representation of her pulse exchanges. The monitors are in the camera’s field of view so she can think.”
“You’re trying to break new ground in automata and cognitive development, and I like how you are trying to induce a Lacanian mirror stage. It makes sense that advanced AI would need psychoanalysis, but even a toddler has theory of mind,” Aki said, remembering the conversation with Jill from the day prior.
“Most do. Her patterns, attempt at mirror stage or not, astonish me because I’m not sure they’re patterns. They’re there but they aren’t. She understands her internal state enough to bring up excerpts from the camera visual of her monitors and rearrange them, but she doesn’t quite seem to know that she’s talking to herself. Theory of mind might come next. She changes the angle of display to make recursive swirls sometimes. I haven’t seen that recently.”
“How long has she been running?”
“Four months. She’s been spacey for weeks. Her psychedelic swirls don’t seem to make sense. Monitoring internal state information is a methodology for evaluating utility preferences. It’s where biology, psychology, and cognitive science combine and allow computers to understand what they want through artificial emotions and desires. The old model was external state, where a computer would measure and answer based on the solution most apparent from the available data. It worked flawlessly for chess, evaluating the pieces and the board, but external assessments don’t tell a computer what it wants.”
“Maybe she is bored of having too little to play with, shut up in here.”
“No, she’s wired. She watches TV and even has net access, though filtered. We can’t have Natalia suddenly perceive the outside world and immediately start shopping online, after all. I keep hoping she’ll make a sudden leap, like showing me pictures of new parts she wants. To understand Builders, we need to understand their long-term associations, the recollections of species memory that tie them together, no matter how different the long-term associations are.
“Homeostasis, cognitive architecture that models complex emotions, even adding the ability to fine-tune internal drives are all part of it. My breakthrough is that she gets to explore herself and generate affective control states. It got too complicated, from the human side, to create algorithmic interpretations that weren’t beholden to human principles. The ranges were still human-based. She’s going to find herself as an integrated mix of these, and I’m betting she’ll lean toward a high-level affective system with episodic and working memory that creates preferences.” Raul smirked again, a mix of pride and wonder. Aki never smirked, but Raul Sanchez’s attitude reminded Aki of herself. He sat down on the bed next to Aki.
“Maybe she needs a reboot.”
Raul glanced at the swirl. “I think she’s deep in thought. There was a pattern like a Riemann surface, maybe a numerical series from advanced math. Nothing lasts long enough to make me sure it’s not being kicked across her from somewhere else. With a thousand terabytes, I can’t keep track of what she’s storing and processing. The tools of decision-making are similar to the ones for interpreting communication, at least in the architecture of my model. But I steered clear of giving her too much detail. She’s going to build it herself, with as little interference from me as possible.”
“Now I get why you needed the supercomputer. Why didn’t you explain this on your application?”
Raul looked at her and laughed. It was higher pitched than Aki would have expected. “The high-priority work gets first dibs. It’s stuff like meteorological models that are important right now. Projects like mine get shoved aside and don’t stand a chance.”
“How do you justify cracking—sorry, using the system then?”
Raul crossed his arms. His glassy gaze fell to the floor. “It’s funny. They’re similar.”
“What are?”
“Alien logic and AI—they both sit there, watching us, never saying a word. When they finally do communicate, it will be clear because there will be no burbles and no interference.”
Aki stared at the tank. The insulating coolant bubbled and flowed, keeping the hardware from overheating. Surreal images wriggled on the monitors, then went away.
“At first, I believed that a strong AI would be something that could communicate and interact with language. One of many missteps. I even had a keyboard in her. AI fails because linguistics itself is bogged down in human definitions. Natalia’s not there yet, but she’s going to think without words, without definitions. Intelligence without language. Humans think words are an evolutionary competitive advantage. Natalia is a species of one. She’s free of having to form concepts that need to function outside of herself. Perhaps the Builders don’t need language or communication because there’s nothing they need to say. Builders will love Natalia. If Builders have a use for love.”
“You are quite convinced by your pet theories.” Aki wondered if communicating with an alien intelligence could be as challenging as communicating with an artificial intelligence. She had never considered that language would be unnecessary for a unique intelligent entity like Natalia. With or without his technology, Raul had given her fuel for her own concepts. She had never questioned the assumption that language could be unnecessary for an intelligent entity. AI systems created by humans communicate the same way humans do. Is this a fundamental characteristic that separates human and nonhuman intelligence, be it alien or artificial?
The message the ETICC was trying to send the Builders was based on the expectation that the Builders would use human deductive reasoning. Jill Elsevier, at least, was enamored with human definitions of sentience. The transmission assumed five identical pulses in a row would be interpreted as the number five, that a series of pulses with a fixed length sent in a cycle would be recognized as a facsimile image, and that the images were easily seen as a twodimensional representation of a 3-D world.
“Can thought and language be parsed?” Aki asked.
“The idea that language regulates thought is a myth. If that were true, children wouldn’t learn words, one language couldn’t be translated into another, and neologisms wouldn’t be coined. When you listen to or read another person’s ideas, it’s not their words that you remember, it’s the concepts the person expresses. Symbolism is very different from symbology.” At least Raul attempted to ground his strange views in logic. Aki had read worse arguments in academic journals.
“You have more interesting ideas than the ETICC does.”
“You really think so?”
“I’m trying to stay out of this, but I will explain excerpted and sanitized details of our discussion. Your girl might turn out to be useful.”