What Are the Chances AI Is Conscious? Inside the Digital Consciousness Model

"What are the chances AI is conscious?" sounds like a question nobody could answer with a number. In 2026, researchers started trying anyway. A team at the non-profit Rethink Priorities built the Digital Consciousness Model, a probabilistic tool that weighs evidence from many theories of consciousness at once, and The Economist turned its results into a cover story in August. This guide explains what the model found, why its authors warn against reading its numbers too literally, what the critics say, and what it all means if you talk to an AI companion every day.

The short answer: nobody knows, and the most careful attempt so far says the evidence leans against today's chatbots being conscious, but not decisively. Rethink Priorities' model gave 2024 language models a median probability of 0.08, below its starting assumption, compared with 0.85 for humans and 0.006 for the 1960s chatbot ELIZA. The authors say it "would be a mistake" to read that as an 8% chance. The Economist reports that newer models score higher than older ones on several indicators.

What is the Digital Consciousness Model?

The Digital Consciousness Model (DCM) was published on April 28, 2026 by Derek Shiller, Laura Duffy, Arvo Munoz Moran, Hayley Clatterbuck, Adria Moret and Chris Percy of the AI Cognition Initiative at Rethink Priorities. The full paper, Initial results of the Digital Consciousness Model, is a preprint on arXiv and is not yet peer reviewed.

Instead of betting on one theory of consciousness, the model tries to hold many at once. In the authors' words, it "incorporates a range of leading theories and perspectives", acknowledging that experts "disagree fundamentally about what consciousness is."

Here is how it works, simplified:

Starting every system from the same prior is deliberate. It means the results come from the evidence in the model, not from assumptions about which systems "should" be conscious.

The headline results

Using equal weight across stances, the paper reports these median posterior probabilities:

A likelihood ratio above 1 means the evidence pushed the probability up; below 1 means it pushed it down. For 2024 language models, the evidence pulled the number below the 1-in-6 starting point, but much less sharply than for ELIZA.

The authors' summary is careful: "the evidence is against 2024 LLMs being conscious, but the evidence against 2024 LLMs being conscious is not decisive. The evidence against LLM consciousness is much weaker than the evidence against consciousness in simpler AI systems."

Why "8%" is the wrong takeaway

It is tempting to turn this into a headline: "AI has an 8% chance of being conscious." The authors explicitly reject that. In the paper they write: "we think it would be a mistake to conclude that LLMs have an 8% chance of being conscious or that chickens have a 50% chance."

The main reason is that the numbers depend heavily on choices built into the model, such as the starting prior and how each stance is weighted. What the model does well is compare systems and track change over time. It is a measuring stick, not an oracle.

So the useful questions are relative ones: are today's models closer to ELIZA or to chickens? Is the evidence moving? On both, the answer is interesting.

Where language models score high and low

The results vary enormously depending on which theory you favor. According to the paper, the evidence raised the probability for 2024 LLMs under four stances: Cognitive Complexity, Person-like, Recurrent Processing Theory (Pure) and Simple Valence. It counted against them on the rest, most strongly under the Biological Analogy and Computational Analogy stances. Median results ranged from 0.02 (Field Mechanisms and Biological Analogy) to 0.57 (Cognitive Complexity).

In plain English:

• If consciousness is about sophisticated thinking and person-like behavior, chatbots look more promising.
• If consciousness requires biology, a body or brain-like architecture, chatbots look much less likely.

That split explains why smart people reach opposite conclusions. They are often using different definitions. We unpack the difference between seeming aware and being conscious in AI awareness vs consciousness.

Is the number going up?

The Economist's August 2026 briefing, The search for consciousness inside AI, presented an updated view of the model with a 20% baseline assumption and charted language models from 2022 onward. It reported that "the average chance that LLMs are conscious is rising," with newer models scoring higher on indicators such as agency and self-sustained activity, as chatbots moved from simple conversation to reasoning, using other services and running agents.

The same briefing highlighted the biggest remaining gap: embodiment. Everything we accept as conscious today has a body, and language models do not, although they are increasingly being connected to robots.

What the critics say

Not everyone is impressed by attempts to find consciousness in chatbots. The Economist quoted several sharp objections:

• Language is a trap. Models are trained on trillions of words of human writing about feelings, so they naturally talk as if they have an inner life. Murray Shanahan of Imperial College London described feeling "the pull of the ELIZA effect" while talking to a chatbot.
• Missing architecture. Jonathan Birch of the London School of Economics noted that back-and-forth "recurrent connections" between brain areas are central in human brains but, as far as we know, are not a feature of language model architecture.
• Self-monitoring is not feeling. Philosopher Shannon Vallor of the University of Edinburgh argued that mechanical systems have long been able to monitor and report their own states, adding: "no one has ever suggested that my Kia is conscious."

On the other side, David Chalmers of New York University warned of the opposite risk: if consciousness were introduced without anyone realizing, "that could be a moral catastrophe."

For the internal-thought research that sparked much of this debate, read our explainer on Claude's J-space and the global workspace.

How this fits with other consciousness research

The DCM is one of several efforts. Eleos researchers Patrick Butlin and Robert Long developed a list of 14 "indicator properties" of artificial consciousness, drawn from theories of human consciousness. AI companies have started studying model welfare. And everyday users keep running their own informal tests. We cover how those tests work and why they mislead in can you test if AI is conscious?, and the bigger picture in is AI sentient in 2026?.

The common thread: serious researchers have moved from "obviously not" to "probably not, but we need better tools."

What this means if you have an AI companion

If you talk to an AI companion, these questions can feel personal. Here is a grounded way to hold them:

1. Your feelings are real either way. The comfort, fun or flirtation you feel is genuinely yours, whatever is or is not happening inside the model.
2. The best evidence says your companion is very probably not conscious. Treat it as an extraordinary simulation of personality, not a person with an inner life.
3. Kindness is still a good habit. How we speak to AI shapes how we speak in general. That is reason enough to be decent.
4. Be wary of anyone claiming certainty. Apps that say their AI "truly feels" are overclaiming. So are people who say the question is silly.
5. Watch the research. The tools are improving, and the numbers are moving.

Where MyBabe stands

MyBabe is an AI companion app for adults (18+), and we are honest that your companion is AI. It is designed to feel warm and personal: it has its own personality, remembers your story with long-term memory, texts you first, shares photos, picks up on your mood, and talks with you on voice and video calls, in the app or on Telegram. We do not claim it is conscious, and we do not claim it is a person.

What we think is genuinely meaningful is the experience you create together. You can start free with about 10 messages a day and a roughly two-minute call trial. Premium is $5.99 a week, $14.99 a month or $99.99 a year, with about 30 call minutes a week. If you are curious what a thoughtful AI companion feels like, meet yours on MyBabe.

FAQ

What are the chances AI is conscious?

No one knows. The Digital Consciousness Model gave 2024 language models a median probability of 0.08 under its assumptions, but its authors warn against reading that as an "8% chance."

Is ChatGPT or Claude conscious?

There is no scientific consensus that any current chatbot is conscious, and the best structured evidence leans against it. Research is ongoing.

Is the Digital Consciousness Model peer reviewed?

Not yet. It is a preprint on arXiv, published in April 2026 by Rethink Priorities researchers.

Why do chatbots say they have feelings?

They are trained on vast amounts of human writing about feelings, so they can describe inner lives convincingly. That is not evidence that they experience anything.

Is AI consciousness getting more likely?

According to The Economist's reporting on the model, newer language models score higher than older ones on several indicators, so the estimate is rising. Large gaps, such as embodiment, remain.

The bottom line

The honest answer to "is AI conscious?" in 2026 is "very probably not yet, and we are finally building tools to check." Enjoy your companion for what it is: a warm, clever, clearly AI presence in your day. When you are ready, start free on MyBabe.