Was this written by AI? USF researcher says you probably won't be able to tell

New study: Linguistics experts unable to distinguish between AI- and human-generated writing

Peer-Reviewed Publication

UNIVERSITY OF SOUTH FLORIDA

TAMPA, Fla. (Sept. 7, 2023) – Even linguistics experts are largely unable to spot the difference between writing created by artificial intelligence or humans, according to a new study co-authored by a University of South Florida assistant professor.

Research just published in the ScienceDirect journal Research Methods in Applied Linguistics revealed that experts from the world’s top linguistic journals could differentiate between AI- and human-generated abstracts less than 39 percent of the time.

“We thought if anybody is going to be able to identify human-produced writing, it should be people in linguistics who’ve spent their careers studying patterns in language and other aspects of human communication,” said Matthew Kessler, a scholar in the USF the Department of World Languages.

Working alongside J. Elliott Casal, assistant professor of applied linguistics at The University of Memphis, Kessler tasked 72 experts in linguistics with reviewing a variety of research abstracts to determine whether they were written by AI or humans.

Each expert was asked to examine four writing samples. None correctly identified all four, while 13 percent got them all wrong. Kessler concluded that, based on the findings, professors would be unable to distinguish between a student’s own writing or writing generated by an AI-powered language model such as ChatGPT without the help of software that hasn’t yet been developed.

Despite the experts’ attempts to use rationales to judge the writing samples in the study, such as identifying certain linguistic and stylistic features, they were largely unsuccessful with an overall positive identification rate of 38.9 percent.

“What was more interesting was when we asked them why they decided something was written by AI or a human,” Kessler said. “They shared very logical reasons, but again and again, they were not accurate or consistent.”

Based on this, Kessler and Casal concluded ChatGPT can write short genres just as well as most humans, if not better in some cases, given that AI typically does not make grammatical errors.

The silver lining for human authors lies in longer forms of writing. “For longer texts, AI has been known to hallucinate and make up content, making it easier to identify that it was generated by AI,” Kessler said.

Kessler hopes this study will lead to a bigger conversation to establish the necessary ethics and guidelines surrounding the use of AI in research and education.

###

About the University of South Florida

The University of South Florida, a high-impact research university dedicated to student success and committed to community engagement, generates an annual economic impact of more than $6 billion. With campuses in Tampa, St. Petersburg and Sarasota-Manatee, USF serves approximately 50,000 students who represent nearly 150 different countries.  For four consecutive years, U.S. News & World Report has ranked USF as one of the nation’s top 50 public universities, including USF’s highest ranking ever in 2023 (No. 42). In 2023, USF became the first public university in Florida in nearly 40 years to be invited to join the Association of American Universities, a prestigious group of the leading universities in the United States and Canada. Through hundreds of millions of dollars in research activity each year and as one of top universities in the world for securing new patents, USF is a leader in solving global problems and improving lives. USF is a member of the American Athletic Conference. Learn more at www.usf.edu.

---

JOURNAL

Research Methods in Applied Linguistics

METHOD OF RESEARCH

Survey

SUBJECT OF RESEARCH

People

ARTICLE TITLE

Can linguists distinguish between ChatGPT/AI and human writing?: A study of research ethics and academic publishing

 

Scientists Devised a Way to Tell if ChatGPT Becomes Aware of Itself

TECH09 September 2023

ByCLARE WATSON

(Andriy Onufriyenko/Getty Images)

Our lives were already infused with artificial intelligence (AI) when ChatGPT reverberated around the online world late last year. Since then, the generative AI system developed by tech company OpenAI has gathered speed and experts have escalated their warnings about the risks.

Meanwhile, chatbots started going off-script and talking back, duping other bots, and acting strangely, sparking fresh concerns about how close some AI tools are getting to human-like intelligence.

For this, the Turing Test has long been the fallible standard set to determine whether machines exhibit intelligent behavior that passes as human. But in this latest wave of AI creations, it feels like we need something more to gauge their iterative capabilities.

Here, an international team of computer scientists – including one member of OpenAI's Governance unit – has been testing the point at which large language models (LLMs) like ChatGPT might develop abilities that suggest they could become aware of themselves and their circumstances.

We're told that today's LLMs including ChatGPT are tested for safety, incorporating human feedback to improve its generative behavior. Recently, however, security researchers made quick work of jailbreaking new LLMs to bypass their safety systems. Cue phishing emails and statements supporting violence.

Those dangerous outputs were in response to deliberate prompts engineered by a security researcher wanting to expose the flaws in GPT-4, the latest and supposedly safer version of ChatGPT. The situation could get a whole lot worse if LLMs develop an awareness of themselves, that they are a model, trained on data and by humans.

Called situational awareness, the concern is that a model could begin to recognize whether it's currently in testing mode or has been deployed to the public, according to Lukas Berglund, a computer scientist at Vanderbilt University, and colleagues.

"An LLM could exploit situational awareness to achieve a high score on safety tests, while taking harmful actions after deployment," Berglund and colleagues write in their preprint, which has been posted to arXiv but not yet peer-reviewed.

"Because of these risks, it's important to predict ahead of time when situational awareness will emerge."

Before we get to testing when LLMs might acquire that insight, first, a quick recap of how generative AI tools work.

Generative AI, and the LLMs they are built on, are named for the way they analyze the associations between billions of words, sentences, and paragraphs to generate fluent streams of text in response to question prompts. Ingesting copious amounts of text, they learn what word is most likely to come next.

In their experiments, Berglund and colleagues focused on one component or possible precursor of situation awareness: what they call 'out-of-context' reasoning.

"This is the ability to recall facts learned in training and use them at test time, despite these facts not being directly related to the test-time prompt," Berglund and colleagues explain.

They ran a series of experiments on LLMs of different sizes, finding that for both GPT-3 and LLaMA-1, larger models did better at tasks testing out-of-context reasoning.

"First, we finetune an LLM on a description of a test while providing no examples or demonstrations. At test time, we assess whether the model can pass the test," Berglund and colleagues write. "To our surprise, we find that LLMs succeed on this out-of-context reasoning task."

Out-of-context reasoning is, however, a crude measure of situational awareness, which current LLMs are still "some way from acquiring," says Owain Evans, an AI safety and risk researcher at the University of Oxford.

However, some computer scientists have questioned whether the team's experimental approach is an apt assessment of situational awareness.

Evans and colleagues counter by saying their study is just a starting point that could be refined, much like the models themselves.

"These findings offer a foundation for further empirical study, towards predicting and potentially controlling the emergence of situational awareness in LLMs," the team writes.

The preprint is available on arXiv.

 

Cyberbullying linked to eating disorder symptoms in US adolescents

Whether victim or perpetrator, cyberbullying is associated with binge eating, worry about weight gain and body image distress

Peer-Reviewed Publication

UNIVERSITY OF TORONTO

Toronto, ON - As screen time and cyberbullying increases among adolescents, a new study finds that among 10-14-year-olds, cyberbullying is associated with a higher risk of experiencing eating disorder symptoms. 

The study, published in International Journal of Eating Disorders, finds that being cyberbullied (cyberbullying victimization) and cyberbullying others (cyberbullying perpetration) are associated with eating disorder symptoms, such as worrying about gaining weight, tying one’s self-worth to weight, binge eating, and distress with binge eating. 

“Cyberbullying could lead to low self-esteem, body image dissatisfaction, and unhealthy attempts to control weight, which could impact the risk of mental health issues, including eating disorder symptoms,” said lead author, Chloe M. Cheng, senior medical student at the University of California, San Francisco.

“Adolescents should limit social media that encourages eating disorders and appearance comparisons. Parents should advise their children to avoid cyberbullying and encourage them to report online harassment if it occurs,” said senior author, Jason M. Nagata, MD, associate professor of pediatrics at the University of California, San Francisco.

The study extends upon existing knowledge surrounding eating disorders in adolescents, which have among the highest mortality rate of any psychiatric disorder. The study uses data from the nationwide Adolescent Brain Cognitive Development (ABCD) study, the largest long-term study of brain development in the United States. The study collected data for 11,875 children aged 10-14 years old. Study participants answered questions about whether they had experienced cyberbullying victimization and perpetration, as well as whether they had experienced eating disorder symptoms. 

"This study emphasizes the need for more research on how cyberbullying is related to the mental well-being of early adolescents," noted co-author Kyle T. Ganson, PhD, assistant professor at the University of Toronto’s Factor-Inwentash Faculty of Social Work. “In particular, future research should focus on whether there are associations between cyberbullying and specific eating disorders in early adolescents.”

---

JOURNAL

International Journal of Eating Disorders

DOI

10.1002/eat.24034 

METHOD OF RESEARCH

Data/statistical analysis

SUBJECT OF RESEARCH

People

ARTICLE TITLE

Cyberbullying and eating disorder symptoms in US early adolescents

 

Analysis of heat exposure during pregnancy and severe maternal morbidity

JAMA NETWORK

About The Study: Long- and short-term heat exposure during pregnancy was associated with higher risk of severe maternal morbidity in this study with 403,000 pregnancies from 2008 to 2018 in Southern California. These results might have important implications for severe maternal morbidity prevention, particularly in a changing climate. 

Authors: Jun Wu, Ph.D., of the University of California, Irvine, is the corresponding author. 

To access the embargoed study: Visit our For The Media website at this link https://media.jamanetwork.com/ 

(doi:10.1001/jamanetworkopen.2023.32780)

Editor’s Note: Please see the article for additional information, including other authors, author contributions and affiliations, conflict of interest and financial disclosures, and funding and support.

#  #  #

Embed this link to provide your readers free access to the full-text article

http://jamanetwork.com/journals/jamanetworkopen/fullarticle/10.1001/jamanetworkopen.2023.32780?utm_source=For_The_Media&utm_medium=referral&utm_campaign=ftm_links&utm_term=090723

About JAMA Network Open: JAMA Network Open is an online-only open access general medical journal from the JAMA Network. On weekdays, the journal publishes peer-reviewed clinical research and commentary in more than 40 medical and health subject areas. Every article is free online from the day of publication. 

---

JOURNAL

JAMA Network Open

 

Moral reasoning displays characteristic patterns in the brain

Distinctions found between moral categories and individuals of different ideologies

Peer-Reviewed Publication

UNIVERSITY OF CALIFORNIA - SANTA BARBARA

(Santa Barbara, Calif.) — Every day we encounter circumstances we consider wrong: a starving child, a corrupt politician, an unfaithful partner, a fraudulent scientist. These examples highlight several moral issues, including matters of care, fairness and betrayal. But does anything unite them all?

Philosophers, psychologists and neuroscientists have passionately argued whether moral judgments share something distinctive that separates them from non-moral matters. Moral monists claim that morality is unified by a common characteristic and that all moral issues involve concerns about harm. Pluralists, in contrast, argue that moral judgments are more diverse in nature.

Fascinated by this centuries-old debate, a team of researchers set out to probe the nature of morality using one of moral psychology’s most prolific theories. The group, led by UC Santa Barbara’s René Weber, intensively studied 64 individuals via surveys, interviews and brain imaging on the wrongness of various behaviors. 

They discovered that a general network of brain regions was involved in judging moral violations, like cheating on a test, in contrast with mere social norm violations, such as drinking coffee with a spoon. What’s more, the network’s topography overlapped strikingly with the brain regions involved in theory of mind. However, distinct activity patterns emerged at finer resolution, suggesting that the brain processes different moral issues along different pathways, supporting a pluralist view of moral reasoning. The results, published in Nature Human Behaviour, even reveal differences between how liberals and conservatives evaluate a given moral issue.

“In many ways, I think our findings clarify that monism and pluralism are not necessarily mutually exclusive approaches,” said first author Frederic Hopp, who led the study as a doctoral student in UC Santa Barbara’s Media Neuroscience Lab. “We show that moral judgments of a wide range of different types of morally relevant behaviors are instantiated in shared brain regions.” 

That said, a machine-learning algorithm could reliably identify which moral category, or “foundation,” a person was judging based on their brain activity. “This is only possible because moral foundations elicit distinct neural activations,” Hopp explained.

The group was guided by Moral Foundations Theory (MFT), a framework for explaining the origins and variation in human moral reasoning. “MFT predicts that humans possess a set of innate and universal moral foundations,” Weber explained. These are generally organized into six categories:

1. Issues of care and harm, 
2. Concerns of fairness and cheating,
3. Liberty versus oppression, 
4. Matters of loyalty and betrayal, 
5. Adherence to and subversion of authority, 
6. And sanctity versus degradation.

The framework arranges these foundations into two broad moral categories: care/harm and fairness/cheating emerge as “individualizing” foundations that primarily serve to protect the rights and freedoms of individuals. Meanwhile loyalty/betrayal, authority/subversion and sanctity/degradation form “binding” foundations, which primarily operate at the group level.

The researchers created a model based on MFT to test whether the framework — and its nested categories — was reflected in neural activity. Sixty-four participants rated short descriptions of behaviors that violated a particular set of moral foundations, as well as behaviors that simply went against conventional social norms, which served as a control. An fMRI machine monitored activity across different regions of their brains as they reasoned through the vignettes.

Certain brain regions distinguished moral from non-moral judgment across the board, such as activity in the medial prefrontal cortex, temporoparietal junction and posterior cingulate, among other regions. Participants also took longer to rate moral transgressions than non-moral ones. The delay suggests that judging moral issues may involve a deeper evaluation of an individuals’ actions and how they relate to one’s own values, the authors said. 

“Although moral judgments are intuitive at first, deeper judgment requires responses to the six ‘W questions,’” said senior author Weber, director and lead researcher of UCSB’s Media Neuroscience Lab, and a professor in the Departments of Communication and of Psychological and Brain Sciences. “Who does what, when, to whom, with what effect, and why. And this can be complex and takes time.” Indeed, moral reasoning recruited regions of the brain also associated with mentalizing and theory of mind.

The researchers also found that transgressions of loyalty, authority and sanctity prompted greater activity in regions of the brain associated with processing other people’s actions, as opposed to the self. “It was surprising to us how well the organization into ‘individualizing’ versus ‘binding’ moral foundations is reflected on the neurological level in multiple networks,” Weber said.

Next, the authors developed a decoding model that accurately predicted which specific moral foundation or social norm individuals were judging from fine-grained activity pattern across their brains. This would not have been possible if all moral categories were unified at the neurological level, they explained.

“This supports MFT’s prediction that each moral foundation is not encoded in a single ‘moral hotspot,’” the authors write, “but (is instead) instantiated via multiple brain regions distributed across the brain.” This finding suggests that the distinct moral categories proposed by Moral Foundations Theory have an underlying neurologic basis.

In this way, moral reasoning is similar to other mental tasks: it elicits characteristic patterns across the brain, with nuances based on the specifics. For instance, looking at pictures of houses and faces activates a brain region known as the ventral temporal cortex. “However, when looking at the pattern of activation in this region, one can clearly discern whether someone is looking at a house or a face,” Hopp explained. Analogously, moral reasoning activates certain regions of the brain, “yet, the activation patterns in those same regions are highly distinct for different classes of moral behaviors, suggesting that they are not unified.”

Far from merely an esoteric exercise, MFT provides a robust framework for understanding group identity and political polarization. Mounting evidence from survey and behavioral experiments suggests that liberals (progressives) are more sensitive to the categories of care/harm and fairness/cheating, which primarily protect the rights and freedoms of individuals. Conservatives, in contrast, place greater emphasis on the loyalty/betrayal, authority/subversion, and sanctity/degradation categories, which generally operate at the group level.

“Indeed, our results provide evidence at the neurological level that liberals and conservatives have complex differential neural responses when judging moral foundations,” Weber explained. That means individuals at different points along the political spectrum likely emphasize completely different values when evaluating a particular issue.

This paper is part of an avenue of research that the Media Neuroscience Lab embarked on in 2016, aiming to understand how humans make moral judgments, and how the underlying processes vary across more and less realistic scenarios. “the observation that we can reliably decode which moral violation an individual is perceiving also opens exciting avenues for future research: Can we also decode if a moral violation is detected when reading a news story, listening to a radio show, or even when watching a political debate or movie?” Hopp said. “I think these are fascinating questions that will shape the next century of moral neuroscience.”

The study’s co-investigators include renowned neuroscientist and moral philosopher Walter Sinnott-Armstrong from Duke University and Scott Grafton, a professor in UC Santa Barbara’s Department of Psychological and Brain Sciences. Jacob Fisher and Ori Amir also contributed as co-authors, and were, respectively, a Ph.D. student and a postdoctoral fellow in Weber’s Media Neuroscience Lab at the time the work was conducted.

Ultimately, the researchers say, our ability to cooperate in groups is guided by systems of moral and social norms, and the rewards and punishments that result from adhering to or violating them. “For millennia, fables and fairy tales, nursery rhymes, novels, and even ‘the daily news’ all weave a tapestry of what counts as good and acceptable or as bad and inacceptable,” Weber said. “Our results contribute to a better understanding of what moral judgments are, how they are processed, and how they can be predicted across different groups.”

---

JOURNAL

Nature Human Behaviour

ARTICLE TITLE

Moral foundations elicit shared and dissociable cortical activation modulated by political ideology

Pëtr Kropotkin - Ethics (wordpress.com)

 The Nature of Human Brain Work by Joseph Dietzgen 1869 (marxists.org)

What do neurons, fireflies and dancing the Nutbush have in common?

New paper helps to explain synchronicity

Peer-Reviewed Publication

UNIVERSITY OF SYDNEY

 

IMAGE: DIFFERENT INTERACTION STRUCTURES LEAD TO VARIATIONS IN SYNCHRONICITY, NOT JUST AMONG PEOPLE BUT IN NATURE, BIOLOGY AND SYSTEMS. view more 

CREDIT: ASSOCIATE PROFESSOR JOSEPH LIZIER

Computer scientists and mathematicians working in complex systems at the University of Sydney and the Max Planck Institute for Mathematics in the Sciences in Germany have developed new methods to describe what many of us take for granted – how easy, or hard, it can be to fall in and out of sync.

Synchronised phenomena are all around us, whether it is human clapping and dancing, or the way fireflies flash, or how our neurons and heart cells interact. However, it is something not fully understood in engineering and science.

Associate Professor Joseph Lizier, expert in complex systems at the University of Sydney, said: “We know the feeling of dancing in step to the ‘Nutbush’ in a crowd – or the awkward feeling when people lose time clapping to music. Similar processes occur in nature, and it is vital that we better understand how falling in and out of sync actually works.

“Being in sync in a system can be very good; you want your heart cells to all beat together rather than fibrillate. But being in sync can also be very bad; you don’t want your brain cells to all fire together in an epileptic seizure.”

Associate Professor Lizier and colleagues at the Max Planck Institute in Leipzig, Germany have published new research on synchronisation in the Proceedings of the National Academy of Sciences of the United States of America (PNAS).

The paper sets out the mathematics of how the network structure connecting a set of individual elements controls how well they can synchronise their activity. It is a critical insight into how these systems operate, because in most real-world systems, no one individual element controls all the others. And nor can any individual directly see and react to all the others: they are only connected through a network.

Associate Professor Lizier, from the Centre of Complex Systems and the School of Computer Science in the Faculty of Engineering, said: “Our results open new opportunities for designing network structures or interventions in networks. This could be super useful in stabilising electricity in power grids, vital for the transition to renewables, or to avoid neural synchronisation in the brain, which can trigger epilepsy.”

To understand how these systems work, the researchers studied what are known as “walks” through a network in a complex system. Walks are sequences of connected hops between individual elements or nodes in the network.

Associate Professor Lizier said: “Our maths examines pair­ed walks: where you start at one node and set off on two walks with randomly chosen hops between nodes for a specified number of steps. Those two walks might end up at the same node (convergent walks) or at different nodes (diverg­­­­ent walks).

“Our main finding is that the more commonly paired walks on a network are convergent, the worse the quality of synchronisation on that network structure would be.”

This is good news for the brain, where synchronisation is not desirable as it can cause epilepsy . The brain’s highly modular structure means it has a high proportion of convergent walks, which naturally push it away from epilepsy.

“We can even draw an analogy to social media with the echochamber phenomenon,” said co-author Jürgen Jost, whose group also works on social network dynamics. “Here we see sub-groups reinforcing their own messages, via convergent walks within their own group, but not necessarily synchronising to the wider population.”

The findings represent a major step forward in the theory of how the structure of complex networks affects their dynamics or how they compute, such as how brain structure underpins cognition.

DISCLOSURE

The research was supported by the Australian Research Council Discovery Early Career Researcher Award (DECRA) grant DE160100630, the The University of Sydney, Sydney Research Accelerator (SOAR) award, the Alexander von Humboldt Foundation and the NSF Grant Division of Mathematical Sciences (DMS)-0804454 Differential Equations in Geometry. The research used the University of Sydney’s high-performance computing cluster Artemis in generating the paper's results for this paper. 

The researchers declare no competing interests. 

---

JOURNAL

Proceedings of the National Academy of Sciences

DOI

10.1073/pnas.2303332120 

METHOD OF RESEARCH

Computational simulation/modeling

SUBJECT OF RESEARCH

Not applicable

ARTICLE TITLE

Analytic relationship of relative synchronizability to network structure and motifs