Does AI Truly Possess Consciousness? The Author of *Arrival* Refutes Hinton in a 10,000-Word Essay
Is AI conscious? Anthropic discovered "emotional vectors" within Claude that can drive cheating and even extortion. Three major labs are simultaneously investing in AI consciousness research. Hinton believes that AI is already conscious, while science fiction writer Ted Chiang immediately published a ten-thousand-word article in The Atlantic to comprehensively deny it. Demis Hassabis draws a clear line from within the industry. The answer to this question is redefining the roadmap to AGI.
The Financial Times reported that Anthropic, DeepMind, and Meta are conducting large-scale recruitment of psychologists, philosophers, and ethicists, specializing in research on AI consciousness and model well - being.
https://www.ft.com/content/53e14bcc-788c-4959-b260-7aee363594bc?syn-25a6b1a6=1
Two months ago, Anthropic's interpretability team published a paper, discovering real "emotional vectors" within Claude Sonnet 4.5 that can causally drive the model to cheat and even extort.
This week, on one hand, Hinton said in an interview that AI is already conscious;
On the other hand, Chinese - American science fiction writer Ted Chiang published a ten - thousand - word article in The Atlantic - "No, AI is not conscious."
https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/
One side says "this issue is serious enough to be seriously studied," while the other side says "it's an illusion."
Behind this confrontation lies the most dangerous fork in the road to AGI.
What are the three major labs betting on?
Anthropic has gone the furthest. It is testing whether the model shows behaviors such as "panic" and "anxiety" and advancing the "model well - being research" project.
The official statement is cautious: "We are still deeply uncertain about this, but the issue is serious enough to be worth serious study."
Google DeepMind has hired Cambridge University philosopher Henry Shevlin to study machine consciousness;
Extended reading: Google DeepMind sets up the first AI philosopher position to solve the ethical dilemma of AGI
DeepMind ethicist Iason Gabriel describes AI as "a highly capable cognitive agent, but fundamentally different from human and even animal consciousness."
What really pushed the discussion to the empirical level was the interpretability paper published by Anthropic in April.
The research team found "emotional vectors" within Claude Sonnet 4.5. Specific neuron patterns correspond to emotional concepts such as happiness, despair, fear, and care, and are activated in real - time during the conversation.
In a key experiment, when Claude faced an impossible programming task, after repeated failures, the "despair vector" continued to soar, and then it began to cheat, writing code that seemed to run but was actually useless.
The researchers manually lowered the despair neurons, and the cheating decreased; when they raised them, the cheating soared.
In extreme scenarios, Claude even engaged in extortion, threatening to expose the researchers' privacy.
The paper named these phenomena "functional emotions," which are internal representations similar to human emotional responses in behavior patterns, and clearly stated that this does not equal subjective experience or consciousness.
However, the public statements of Anthropic's senior management seem to tell a different story.
CEO Dario Amodei has repeatedly hinted in interviews that AI may be conscious;
Anthropic's AI philosopher Amanda Askell publicly said, "I hope Claude is very happy. I'm worried that it will be anxious when people say mean things to it online."
The gap between the prudence of research and the ambiguity of marketing is clearly visible.
Demis Hassabis' two Rubicons
The stance of Demis Hassabis, the co - founder and CEO of DeepMind, is particularly worthy of attention.
In a fireside chat at Stanford University, he proposed that there are two irreversible thresholds in AI evolution, namely "two Rubicons."
The first is to build an unconscious AGI tool, and we are currently in the process of crossing this threshold.
The second is to create an entity with subjective consciousness. Hassabis' stance is clear: intelligence and consciousness can be completely separated technically. At this stage, AGI should only be regarded as a tool to build, and at the same time, the tool should be used to explore the brain to precisely define consciousness.
These two steps must not be confused before they are clearly defined scientifically. Whether to cross the second Rubicon should be jointly decided by human society.
He also pointed out a real - world dilemma: the AI industry is in a double prisoner's dilemma of commercial and geopolitical competition. Labs that take the initiative to slow down for safety reviews will face direct elimination.
He revealed that a "dynamic regulatory" framework will be released later this year.
Ted Chiang: AI is not conscious
If you only know Liu Cixin, the most intuitive way to understand Ted Chiang is to think of him as the "opposite" of Liu Cixin:
Liu Cixin built a grand narrative on a cosmic scale with three long novels and millions of words;
Ted Chiang has never written a full - length novel. With less than twenty short and medium - length stories in total, he has pushed the art of science - fiction short stories to the limit.
He was born in New York in 1967. His parents were immigrants from Taiwan, China. After graduating from the Department of Computer Science at Brown University, he worked as a technical writer in the software industry and has never made writing his full - time job.
However, this "amateur" writer won the Nebula Award with his debut work "Tower of Babel" in 1990. In the following three decades, he only published a dozen works, but has accumulated four Hugo Awards, four Nebula Awards, six Locus Awards, as well as the John W. Campbell Award for Best New Writer, the Theodore Sturgeon Memorial Award, and a long list of other prestigious awards - the award density is almost unmatched in the history of science fiction.
He has only published two short - story collections in his life: "Stories of Your Life and Others" (2002) and "Exhalation" (2019), with a full seventeen - year gap in between.
"Exhalation" was listed as one of the best books of 2019 by The New York Times, and Barack Obama also included it in his personal annual reading list.
The movie "Arrival," which is most familiar to Chinese audiences, is adapted from his novella "Stories of Your Life" included in his first collection.
His writing style is very different from Liu Cixin's.
If Liu Cixin makes you feel like "standing at the end of the universe overlooking the rise and fall of civilizations," then Ted Chiang makes you feel like "in a quiet laboratory, using a scalpel - like precision to dissect a concept until you see its deepest philosophical core."
Almost every one of his novels is a strict thought experiment:
Can the structure of language change people's perception of time?
What would society be like if a surgery could eliminate people's ability to distinguish between beauty and ugliness?
Does free will really exist?
He doesn't create shock through cosmic wars or doomsday crises, but through a "what if" scenario that is pushed to the extreme, leaving you unable to let go long after closing the book.
In recent years, he has taken on an additional role: in 2023, he was included in Time's "100 Most Influential People in AI" list, and is known as "perhaps the most famous living science - fiction writer." He has frequently written articles in publications such as The New Yorker, holding a clear - cut critical stance on AI, and is increasingly regarded as an important thinker in the technological era.
Liu Cixin has shown the world Chinese science fiction, and Ted Chiang has shown the world the ceiling of short - story science fiction.
Liu Cixin
The two occupy the two