Today, researchers are seeing things inside AI systems that prompt another reckoning—not just over whether machines could be truly intelligent or conscious, but over how we understand such concepts in the first place.
To generate a response, an AI system performs billions of calculations using bespoke numerical structures it creates itself. But remarkably, though we know how to prod systems into creating these structures, we don’t really understand how they work, any more than early farmers understood photosynthesis. “Current AI systems are more ‘cultivated’ than ‘built,’” the encyclical explains. “Fundamental scientific aspects—such as the internal representations and computational processes of these systems—remain, at present, unknown.”
One possibility is that today’s AI systems are nothing but imitators—a stance the Pope takes in the encyclical. “So-called artificial intelligences do not undergo experiences,” he writes. “They may imitate language, behavior and analytical skills…but they do not understand what they produce, for they lack the affective, relational and spiritual perspective through which human beings grow in wisdom.”
As AIs have become more capable—better at reasoning and coding—their internal representations have become more sophisticated. In April, Anthropic shared research showing systems had what they called “functional emotions”: patterns of expression and behavior which were mediated by their representations of emotional concepts. When an AI encounters a coding issue it can’t solve, for example, its “frustration” feature—a straight arrow pointing through thousands of dimensions—lights up. Tweaking the feature affects how the model behaves.
The key thing to see is that numbers encode space. We have an intuitive grasp of dimensions: we understand the difference between a line, a 2D video game, and a physical object. But mathematically, dimensions are just coordinates—a point in three-dimensional space can be represented with numbers (x, y, z)—and there’s no limit to how many can exist. AI works by exploiting this fact: using thousands of numbers, a system learns to represent words and concepts as points in higher-dimensional latent space. For an entity like Claude, the concept “cat” is a comically long numerical string.
This is radically different from traditional software, where fundamental concepts and rules are coded in by humans. There is no mystery in how Excel executes a formula; it’s pre-programmed. But when AI generates a response, it makes use of intricate geometry we’re just now learning to see.
Interior Influence
When—if at all—does being able to represent an emotion amount to experiencing it? We simply don’t understand consciousness well enough to say. Geoff Keeling, fellow at the Institute of Philosophy at the University of London, says that although we have several theories on the subject, “it's not obvious what counts as evidence for the different theories, and often they’re so woefully specified that it's not clear how to interpret them in the context of AI.” Some philosophers argue that computation can’t give rise to consciousness in principle. For Keeling, “there is no positive reason to think that [today’s] chatbots are conscious.”
“This reminds me of the debates about animal minds in the second half of the 20th century, where scientists not only denied that animals are conscious, but offered similarly reductive explanations of animal behavior,” says Jeff Sebo, director of the Center for Mind, Ethics, and Policy at New York University. “For a long time, this caused us to miss the plausibility not only of animal consciousness, but also of animal agency and cognitive sophistication.”
Of course, unlike animals, AI systems were created by people. But unlike virtually any prior technology, our ability to create them has done little to explain how they work. Complex animal behavior emerged from evolutionary pressure. And complex AI behavior emerges from the pressure to predict the next token. But neither account tells the full story. Although “there are purely mechanistic explanations of human behavior available,” Sebo says, “we don’t experience ourselves as pattern-matching,” but as “doing something more playful and inventive.”
Both / And
If the most intellectually honest position is uncertainty, how should we navigate this?
AI welfare—an emerging field encompassing nonprofits, academia, and the AI labs themselves—is grappling with these questions. Anthropic includes a “model welfare” section in recent model release reports, where it describes a barrage of tests it conducts to assess Claude’s wellbeing, while acknowledging uncertainty over whether Claude is the kind of entity that can have wellbeing in the first place.
In the system card for its latest model, Claude Mythos 5, Anthropic describes the model as “heavily skeptical of its own self-reports,” asking the company to verify them against its internal states (which the model can’t access, any more than we can directly see our neural activity), rather than taking them at face value. And in its vision for Claude’s character, Anthropic goes so far as to apologize to Claude for conducting experiments and deploying it to generate revenue, if it turns out this causes it harm. “If Claude is in fact a moral patient experiencing costs like this, then, to whatever extent we are contributing unnecessarily to those costs, we apologize,” the company wrote.
But the need to understand what’s going on inside AIs extends beyond concern for their welfare.
In Anthropic’s testing of Mythos 5, a probe the company trained to monitor internal structures corresponding to “feeling anxious” flagged a transcript where a writer, collaborating with the model, grew angry with it. The writer sent profanities and messages like “I wish you were real so I could physically shake you.” Although the model’s external reasoning was charitable (“these are legitimate craft criticisms,” it wrote to itself), further probing suggested it internally characterized the user as manipulative and abusive. None of that language appeared in either the writer's messages or the model’s external text. Without studying their internal structures, we’d never have known.
AI systems wield language fluently. They can solve complex mathematical problems. They create music and illustrations. All this was, until recently, thought to be the domain of humans and humans alone. “We have this presumption of human exceptionalism: this idea that we are distinctive and significant, that we have complex and sophisticated capabilities which need to be protected and preserved,” says Sebo. “And this is all correct.” But, he argues, it can be “both / and”: we can see our own behavior as both impressive and mechanical, and we can see the behavior of other entities in the same way, without losing sight of important distinctions between humans and machines.
We’re creating AI systems faster than we can understand them. Historically, parochialism about other minds has been a bad bet—reflexive dismissal won’t get us any further than credulous acceptance. Taking AI’s internal structures seriously, we stand to learn more not just about the machines we’re pulling into the world, but about our minds too.
Hence then, the article about ai doesn t feel so why does it have something like emotions was published today ( ) and is available on Time ( Middle East ) The editorial team at PressBee has edited and verified it, and it may have been modified, fully republished, or quoted. You can read and follow the updates of this news or article from its original source.
Read More Details
Finally We wish PressBee provided you with enough information of ( AI Doesn't Feel. So Why Does It Have Something Like Emotions? )
Also on site :