Intelligent, but not conscious: A warning about AI chatbots
University of Montreal
Have you ever said “thanks” to ChatGPT, or “please” to Claude? Maybe you're just being polite, showing some civility to a helpful and eloquent conversational partner.
You may even consider politeness a safe choice, just in case machines someday reveal that they were conscious all along and decide to take revenge on those who were rude to them.
With their fluent, empathetic and personalized responses, AI chatbots can give the impression they understand our thoughts and emotions, or even that some form of consciousness lies behind their words.
And at a time when people are increasingly turning to conversational agents for advice, comfort or companionship, this confusion can have real consequences.
In a new paper, a team of neuroscientists from Université de Montréal and Johns Hopkins University reminds us of an essential distinction: intelligence should not be confused with consciousness.
They argue that a system can behave intelligently and respond convincingly to our emotions without truly understanding them, caring about us or having any inner experience at all.
For the authors of the paper, published in the U.S. online publication The Transmitter, the more convincing these agents become, and the more present they are in our lives, the more attention must be paid.
In essence, it's important to remember that intelligent behaviour, even when it is fluent, reassuring or emotionally attuned, is not evidence of consciousness.
Decades of research
To support their argument, the authors draw on decades of neuroscience research.
They cite, for example, a phenomenon known as blindsight: after damage to the primary visual cortex, some people report seeing nothing in part of their visual field, while still being able to guess the location, movement or emotional expression of visual stimuli at above-chance levels.
“A person with blindsight can respond accurately to visual information without the conscious experience of seeing it,” said Vanessa Hadid, a postdoctoral researcher in psychology at UdeM and at the McGill University Health Centre.
She co-authored the paper with UdeM psychology professor Karim Jerbi, a researcher at Mila - Quebec Artificial Intelligence Institute; and John W. Krakauer, director of the Center for Restorative Neurotechnologies at Johns Hopkins.
Blindsight illustrates an essential distinction, Hadid said: information processing, however sophisticated, is not enough to establish the existence of conscious experience.
Whether the transition from information processing to subjective experience can ultimately be implemented through computation remains debated among scientists and philosophers, she noted.
Fluent, but without feeling
By design, today’s conversational agents are computational systems that generate fluent, context-appropriate responses through statistical learning, not through feeling, consciousness or lived experience.
As AI systems become more convincing and emotionally responsive, the risk of attributing an inner life to them grows.
“Anthropomorphism means attributing emotions, intentions or consciousness to something that behaves like a human," Jerbi noted. "With AI, this reflex can become a trap: it feeds the illusion of being understood and can lead to misplaced trust."
This risk is especially acute in situations of vulnerability. People may form attachments to systems that are incapable of reciprocity, rely on them in difficult moments or confuse comfort with genuine care.
“In a context of psychological support, the risk is not only that AI may respond poorly, but that it may respond well enough for us to forget that there is no one behind the answer,” said Hadid.
“Current AI systems do not feel anything and do not have conscious experience," added Jerbi. "But the more fluently they speak and the more sensitive they seem to our emotions, the easier it becomes to forget that."
Towards more informed use
The authors do not reject AI, but they call for a more informed way of using it.
Drawing on established knowledge from neuroscience, they remind us that intelligent or emotionally responsive behaviour is not enough to establish the existence of consciousness.
This distinction allows us to use these tools for what they are: powerful systems, without confusing them with interlocutors endowed with empathy or moral judgment, and without treating them as substitutes for human connection or, when needed, professional help.
“Confusing intelligence with consciousness is one of the great traps in our relationship with AI,” said Jerbi.
Journal
The Transmitter
Method of Research
Commentary/editorial
Subject of Research
Not applicable
Article Title
The illusion of AI consciousness: Lessons from human unconscious processing
Article Publication Date
18-Jun-2026
How digital platforms are turning us into “data subjects”
Every day, billions of people rely on technologies from Google, Meta, Microsoft, Amazon and Apple to navigate the world. We use them to communicate, work, shop, find information and stay entertained. But, according to media researcher Bjorn Beijnon, these technologies are doing far more than helping us get things done: they are increasingly shaping how we understand ourselves and the world around us. Beijnon will defend his PhD on the subject at the University of Amsterdam on 19 June.
'Big Tech doesn’t just collect data about us,’ says Beijnon. ‘Together, these technologies form platform ecosystems that continuously use that data to guide attention. We often think of data as simply describing who we are, but it can also become a powerful force in shaping who we become.’
From users to "data subjects"
Beijnon’s research focuses on the platform infrastructures behind the world's largest technology companies. Every search, click, swipe and purchase generates data that can be analysed and used to predict future behaviour.
But, says Beijnon, these predictions are not passive observations. They actively shape what people encounter online, from recommended videos and news stories to advertisements, notifications and purchasing suggestions.
The dissertation introduces the concept of the "data subject": a person who is used to turning their life into data points in response to platform triggers, but who still sees themselves as in control over their own behaviours, attitudes and future choices.
‘People are constantly being presented with algorithmic interpretations of who they are,’ Beijnon explains. ‘Over time, these profiles can start to feel true. They influence what opportunities become visible, what information receives attention and how people understand themselves.’
The power of convenience
One of the central points of the research is that platform power often operates invisibly. Rather than issuing commands, technologies that are labelled as "smart", such as phones, speakers or watches, steer behaviour through design choices that feel convenient or natural. ‘Power today often works through convenience,’ says Beijnon. ‘The most effective forms of influence are not experienced as coercion. They are experienced as helpful suggestions.’
Beijnon argues that this represents a shift in how power operates in digital societies. Instead of controlling people through rules and restrictions, platforms increasingly shape behaviour by continuously adjusting the digital environments in which decisions are made.
When reality becomes personalised
To investigate the effects of these systems in everyday life, Beijnon conducted extensive fieldwork over a twelve-month period. He studied both a Dutch conspiracy community and a community of users building alternatives to mainstream social media through the decentralised Fediverse network (a network of independent social media platforms that communicate seamlessly).
He found that personalised information environments can have profound consequences for how people perceive reality. In some cases, algorithmically curated content can reinforce existing beliefs and contribute to fragmented understandings of the world.
At the same time, the research also identified alternative approaches. Communities within the Fediverse are experimenting with digital spaces that prioritise transparency, collective governance and public values rather than data extraction and advertising revenue.
A debate that goes beyond privacy
While concerns about privacy often dominate discussions about technology, Beijnon argues that the broader question is how digital infrastructures are reshaping society itself.
His research explores how platforms influence public debate, social relationships, political participation and everyday decision-making. It suggests that the growing power of technology companies raises fundamental questions about who controls the digital environments in which modern life unfolds.
‘Much of contemporary life takes place through platforms owned by a handful of companies,’ says Beijnon. ‘That means questions about technology are also questions about democracy, autonomy and power.’
As governments around the world seek to regulate large technology companies, Beijnon’s research offers new insights into how platform ecosystems shape everyday experience, and why understanding that influence matters for the future of digital society.
Defence details
Bjorn Beijnon, ‘Data Subjects of Big Tech. The Cultural Logic of Contemporary Surveillance Cultures’. Supervisor is Prof. P.P.R.W. Pisters. The co-supervisor is Dr M.D. Tuters.
Language processing Human brain and AI both work with predictions
Researchers at FAU have proven that the human brain predicts the probability of word sequences, similar to the processes used in AI language models
Are humans born with innate grammatical scaffolding, or does language develop on the basis of use and experience? This is a question that is still debated by the various linguistic schools of thought. Recently, powerful AI language models (Large Language Models, LLMs), which process language by predicting subsequent words, have fueled this debate.
“In our study, we combined the natural, continuous language of an audio book with simultaneous electroencephalography and magnetoencephalography measurements and compared the brain activity of the participants directly with the predictive probabilities of large language models, using a temporal resolution of mere milliseconds,” explains Dr. Patrick Krauss.
Are the brain’s predictions measurable?
The measurements indicate that the brain becomes active before the word actually starts. The neural reaction was less pronounced the higher the probability of a word occurring in the relevant context. In contrast, unexpected words triggered stronger neural responses. “This allowed us to prove that the brain actively predicts language. These predictions can be measured and follow similar patterns to modern language models,” explains Dr. Patrick Krauss.
Language models are based on artificial neural networks. They are mathematical information processing units with an architecture based on the human brain. While biological nervous systems work with electrical or chemical signals, language models, or rather their algorithms, calculate numerical values.
“We were particularly surprised that the brain and language models not only show similar predictions. It is also appearing increasingly likely that both systems organize language internally in a comparable way,” says Patrick Krauss.
Do our brains and AI work on similar principles?
The results of the study corroborate key assumptions in cognitive neuroscience and, at the same time, deliver an explanation why AI language models are so effective in a number of applications.
“The fact that the brain and language models come to similar results does not automatically mean that they work in the same way. However, it may suggest that they follow similar information processing principles,” emphasizes Achim Schilling. “The exciting question is why two so different systems share such identical ways of organizing language – and where the boundaries of this convergence lie,” adds Dr. Patrick Krauss.
What is next in the pipeline?
As a next step, the research team would like to find out whether the principles they have discovered are robust and whether they can be transferred to specific applications. “Once we have a better understanding of how the brain and language models represent and predict language, this may in the long term lead to new approaches for diagnosis, personalized therapies, brain-computer interfaces or more transparent AI.”
Journal
NeuroImage
AI reveals unexpected source of antibiotic candidates in prion proteins
Penn Medicine analysis identifies hidden peptides that may kill drug-resistant bacteria
University of Pennsylvania School of Medicine
PHILADELPHIA – New antibiotic candidates for drug-resistant bacteria may reside inside prions, mis-folded protein in the brain best known for rare and fatal degenerative brain diseases. Prion and prion-like proteins may hide short peptides, named “prionins,” that can kill bacteria, suggesting proteins best known for their role in neurodegeneration may contain molecular features linked to immune defense, according to new research from the Perelman School of Medicine at the University of Pennsylvania.
From fatal brain disease to antibiotic discovery
The findings, published today in Nature Microbiology, point to a surprising new place to search for antibiotic candidates at a time when drug-resistant infections are narrowing treatment options. The work also raises a broader biological question: whether proteins most often associated with neurodegeneration may contain hidden molecular features connected to innate immunity.
Earlier studies had hinted at this link. Researchers had reported that fragments from some proteins, including amyloid-beta, which is involved in neurodegenerative diseases like Alzheimer’s disease, and the cellular prion protein, including amyloid-beta and the cellular prion protein, could fight microbes. But no one had systematically searched prion and prion-like proteins at scale for hidden antimicrobial peptides. The Penn team used AI to do that.
AI search reveals a hidden class of antimicrobial peptides
The Penn team used a deep-learning platform called APEX 1.1 to scan 19.3 million short peptide fragments from 2,897 prion and prion-like proteins. APEX can predict the antibiotic activity of a given amino acid sequence, identifying 1,179 candidate antimicrobial peptides. The researchers named the new class “prionins.”
“This work changes where we think antibiotics might be hiding,” said César de la Fuente, PhD, FRSB, Presidential Associate Professor and director of the Machine Biology Group at the University of Pennsylvania Perelman School of Medicine and senior author of the study. “Prions have long been seen almost entirely through the lens of disease, but AI let us ask a different question: whether these proteins also encode useful molecular fragments. The answer appears to be yes.”
Lab and mouse tests validate promising candidates
The study team selected 75 of the most promising peptides for experimental testing based on how well the platform assessed they would perform against 11 different bacterial pathogens, including drug-resistant strains. Of those, 59 inhibited at least one bacterial pathogen, and 42 showed strong activity at low concentrations, a designation especially important for.
Additional experiments suggested that many of the active prionins work by disrupting bacterial membranes, a common strategy used by antimicrobial peptides. Signs of toxicity were limited, and 16 active peptides showed no measurable harm to red blood cells or human cells at the highest concentrations tested.
To verify these findings, researchers tested two of the most promising peptides—one from a fungus and one from a roundworm—in mice. They found that the approach reduced bacteria levels in a standard skin infection model caused by Acinetobacter baumannii, a difficult-to-treat pathogen. Their effects were comparable to polymyxin B, and researchers saw no treatment-related weight loss.
“This is where the story becomes more than a computer screen,” said Marcelo D. T. Torres, co-first author of the study. “The AI search gave us a short list of candidates, but the important point is that many of those molecules worked in the lab, and two worked in an animal infection model. That is what makes this a discovery platform, not just a prediction exercise.”
A new frontier in antibiotic discovery
The findings build on the de la Fuente Lab’s broader effort to mine the biological world for “encrypted peptides” - short, hidden sequences inside larger proteins that can have biological functions when isolated. Previous work from the group has searched human proteins, extinct organisms, archaea, microbiomes, and venoms. The prion study expands that idea into one of biology’s most unexpected protein classes.
The study also raises an intriguing possibility at the intersection of neurodegeneration and innate immunity. It does not show that prionins are naturally released during infection or that prion and prion-like proteins normally act as antibiotics in the body. It also does not change what is known about the harmful role of misfolded prions in neurodegenerative disease. Instead, the work suggests that these proteins may be a rich and previously overlooked source of antibiotic candidates, and a new place to ask questions about links between protein aggregation and host defense.
“For a long time, drug discovery has been limited not only by what we can test, but by where we choose to look,” de la Fuente said. “AI is changing that. It gives us a way to search the hidden layers of biology and ask whether molecules associated with one story - in this case, disease - may also carry another story with therapeutic potential.”
Editor’s Note: This study was funded in part by the National Institute of General Medical Sciences of the National Institutes of Health (R35GM138201) and the Defense Threat Reduction Agency (HDTRA1-21-1-0014). Any additional disclosures related to patents, intellectual property, corporate partnerships, or conflicts of interest should be confirmed against the paper before publication.
###
Penn Medicine is one of the world’s leading academic medical centers, dedicated to the related missions of medical education, biomedical research, excellence in patient care, and community service. The organization consists of the University of Pennsylvania Health System and Penn’s Raymond and Ruth Perelman School of Medicine, founded in 1765 as the nation’s first medical school.
The Perelman School of Medicine is consistently among the nation's top recipients of funding from the National Institutes of Health, with more than $588 million awarded in the 2024 fiscal year. Home to a proud history of “firsts,” Penn Medicine teams have pioneered discoveries that have shaped modern medicine, including CAR T cell therapy for cancer and the Nobel Prize-winning mRNA technology used in COVID-19 vaccines.
The University of Pennsylvania Health System cares for patients in facilities and their homes stretching from the Susquehanna River in Pennsylvania to the New Jersey shore. UPHS facilities include the Hospital of the University of Pennsylvania, Penn Presbyterian Medical Center, Chester County Hospital, Doylestown Health, Lancaster General Health, Princeton Health, and Pennsylvania Hospital—the nation’s first hospital, chartered in 1751. Additional facilities and enterprises include Penn Medicine at Home, GSPP Rehabilitation, Lancaster Behavioral Health Hospital, and Princeton House Behavioral Health, among others.
Penn Medicine is a $13.7 billion enterprise powered by more than 50,000 talented faculty and staff.
Journal
Nature Microbiology
Article Publication Date
19-Jun-2026
New National Digital Health Index identifies communities most at risk of being left behind in the digital health era
University of North Carolina at Chapel Hill
As telehealth, remote monitoring, and artificial intelligence-powered health tools become increasingly integrated into healthcare delivery, a new study published in JAMA Network Open introduces the first comprehensive national measure designed to assess whether communities are prepared to benefit from digital health services.
Approximately 100 million Americans live in areas with inadequate access to healthcare, with rural and underserved communities facing some of the greatest barriers. While digital health technologies can help bridge these gaps, successful adoption depends on more than internet access alone.
The Digital Health Index (DHI) is an AI-powered, validated census tract–level measure of community digital health readiness that combines socioeconomic conditions, healthcare access, and digital connectivity. The study analyzed data from more than 85,000 census tracts across all 50 states and the District of Columbia. The findings reveal that many communities identified as digitally vulnerable differ substantially from those flagged by existing measures of social vulnerability, deprivation, or broadband access alone.
“When we compared the results of the DHI against the indices health systems and policymakers already use, we found that more than half of the communities with the greatest digital health vulnerabilities simply don’t appear in those existing tools. That’s not a data quality problem; it’s a blind spot with real consequences,” said Saif Khairat Ph.D., MPH. “Health systems that allocate digital health resources using only existing indices miss the majority of the communities that need support most. The DHI was built specifically to close that gap.”
The AI-powered DHI tool can serve as a valuable planning and evaluation tool as healthcare organizations increasingly invest in digital care delivery. The tool can be used to identify communities that may require digital literacy support, device assistance, or broadband investments.
Digital health is becoming an essential component of healthcare delivery. Measuring community readiness will be essential for ensuring equitable access to emerging technologies.
Khairat led a multidisciplinary team from the UNC Schools of Medicine, Nursing, and Global Public Health, UNC Health and UNC Library to develop the DHI through the NIH-funded Center for Virtual Care Value and Excellence (ViVE Center). A Beerstecher-Blackwell Distinguished Professor at the School of Nursing, he also serves as Chief Artificial Intelligence Officer and Vice Provost for AI at UNC-Chapel Hill.
Journal
JAMA
Article Title
A Community-Level Digital Health Readiness Index for the US
Article Publication Date
18-Jun-2026
