Researchers at the University of Notre Dame are developing artificial intelligence tools that help consumers understand how they are being exploited as they navigate online platforms. The goal is to boost the digital literacy of end users so they can better control how they interact with these websites.

In a recent study appearing in Proceedings of the CHI Conference on Human Factors in Computing Systems, participants were invited to experiment with online privacy settings without consequence. To test how different data privacy settings work, the researchers created a Chrome browser plug-in called Privacy Sandbox that replaced participant data with personas generated by GPT-4, a large language model from OpenAI.

With Privacy Sandbox, participants could interact with different websites, such as social media platforms or news outlets. As they navigated to various sites, the browser plug-in applied AI-generated data, making it more obvious for participants to see how they were targeted based on their supposed age, race, location, income, household size and more.

“From a user perspective, allowing the platform’s access to private data may be appealing because you could get better content out of it, but once you turn it on, you cannot get that data back. Once you do, the site already knows where you live,” said Toby Li, assistant professor of computer science and engineering and a faculty affiliate at the Lucy Family Institute for Data & Society at Notre Dame, who led the research. “This is something that we wanted participants to understand, figure out whether the setting is worth it in a risk-free environment, and allow them to make informed decisions.”

Another study, this one in Proceedings of the ACM on Human-Computer Interaction, looked at dark patterns—or the design features on digital platforms that subtly nudge users to perform specific actions—and how they are used on websites to manipulate customers. For the study, Li and his team looked at how dark patterns are applied by interface designers to encourage people to consume more content or make impulsive purchasing decisions.

The researchers developed a Chrome browser plug-in dubbed Dark Pita to identify dark patterns on five popular online platforms: Amazon, YouTube, Netflix, Facebook and X.

Using machine learning, the plug-in would first notify study participants that a dark pattern was detected. It would then identify the threat susceptibility of the dark pattern and explain the impact of the dark pattern—financial loss, invasion of privacy or cognitive burden. Dark Pita would then give participants the option to “take action” by modifying the website code through an easy-to-use interface to change the deceptive design features of the site and explain the effect of the modification.

The researchers plan to eventually make both browser plug-ins, Privacy Sandbox and Dark Pita, available to the public. Li believes these tools are great examples of how the use of AI can be democratized for regular users to benefit society.

“Companies will increasingly use AI to their advantage, which will continue to widen the power gap between them and users. So with our research, we are exploring how we can give back power to the public by allowing them to use AI tools in their best interest against the existing oppressive algorithms. This ‘fight fire with fire’ approach should level the playing field a little bit,” Li said.

“An Empathy-Based Sandbox Approach to Bridge the Privacy Gap Among Attitudes, Goals, Knowledge, and Behaviors” was presented at the 2024 Association of Computing Machinery CHI Conference. Led by Li, fellow study co-authors include Chaoran Chen and Yanfang (Fanny) Ye from Notre Dame, Weijun Li from Zhejiang University, Wenxin Song at the Chinese University of Hong Kong and Yaxing Yao at Virginia Tech.

The study “From Awareness to Action: Exploring End-User Empowerment Interventions For Dark Patterns in UX,” led by Li, has been published in the Proceedings of the ACM on Human-Computer Interaction (CSCW 2024). Co-authors include Yuwen Lu from Notre Dame, Chao Zhang from Cornell University, Yuewen Yang from Cornell Tech and Yao from Virginia Tech.

A new study presents an evolutionary-biophysical model that sheds new light on the evolution of the collaborative non-self recognition self-incompatibility, a genetic mechanism in plants that prevents self-fertilization and promotes cross-fertilization. The innovative model introduces promiscuous molecular interactions as a key ingredient, enhancing our understanding of genetic diversity and evolution in hermaphroditic plants.

A study led by Dr. Tamar Friedlander and her team at The Robert H. Smith Institute of Plant Sciences and Genetics in Agriculture at the Hebrew University, in collaboration with Prof. Ohad Feldheim from the Einstein Institute of Mathematics at the Hebrew University has developed an evolutionary-biophysical model that sheds new light on the evolution of collaborative non-self recognition self-incompatibility in plants.

The study, appearing in Nature Communications, introduces a novel theoretical framework that incorporates promiscuous molecular interactions, which have been largely overlooked by traditional models.

Self-incompatibility (SI) is a widespread biological mechanism in plants having both male and female reproductive organs, that prevents self-fertilization and promotes genetic diversity. Under this mechanism, fertilization relies on the specific recognition between highly diverse proteins: the RNase (female determinant) and the SLF (male determinant).

The interaction between these proteins ensures that plants are only compatible with non-self mates, thus maintaining a diverse gene pool.

The new model proposed by Dr. Friedlander and her team represents a significant advancement in understanding the evolutionary dynamics of self-incompatibility proteins. By allowing for promiscuous interactions—where interactions with unfamiliar partners are likely—and for multiple distinct partners per protein, the model aligns more closely with empirical findings than previous models that assumed only one-to-one interactions.

This promiscuity enables a flexible interaction pattern between male and female proteins, offering new insights into how these proteins evolve and interact over generations.

“Our research shows that the ability of proteins to engage in promiscuous interactions is crucial for the long-term evolutionary maintenance of self-incompatibility systems,” explained Dr. Friedlander.

“We propose that the default state of this system is that recognition is likely and evolutionary pressure is needed to avoid it, in contrast to what was previously thought. This flexibility not only helps in maintaining genetic diversity but also suggests that similar mechanisms could be operating in other biological systems.”

The study also reveals how populations of these plants spontaneously organize into distinct compatibility classes, ensuring full compatibility across different classes while maintaining incompatibility within the same class.

The model predicts various evolutionary paths that could lead to the formation or elimination of these compatibility classes based solely on point mutations. The dynamic balance between the emergence and decay of these classes, which provides a sustainable model of evolution, was analyzed by the researchers using a mixture of empirical and theoretical tools borrowed from the field of statistical mechanics in physics.

“These insights from our study have profound implications not only for plant biology but also for understanding the fundamental principles of molecular recognition and its impact on the evolution of biological networks,” Dr. Friedlander added. “Our findings could also help in the conservation of plant biodiversity.”

This research, which highlights the role of promiscuous and multi-partner molecular interactions, is likely to inspire seeking these two elements in additional biological systems, and help in explaining the evolution of various complex molecular networks. It enriches the understanding of plant biology and has broader implications for deciphering the evolution of biological networks and managing biodiversity.