Science Integrity Alliance asks Professor Anderson Rocha, Director of Recod.ai.

[SIA] Recod.ai has grown into one of Latin America’s largest AI labs. What has been key to sustaining such an active and collaborative research environment?

[AR] Certainly, the most important aspect of Recod.ai’s success relies on its multidisciplinary approach, putting together experts from different fields who see problems from different vantage points. When you join forces from Social Sciences, Life Sciences, and Exact Sciences, the sky is the limit. Every project relies on collaboration with experts in the application fields so that any designed AI algorithm considers different lenses and a humanistic view, with the human in the loop during data collection, processing, learning, actioning, and decision-making.

[SIA] The Horus Project brings together international partners to promote trust in digital environments. What are its main goals?

[AR] The Horus Project is an ambitious and multidisciplinary initiative aimed at advancing AI research and strengthening trust in digital media. It is structured around five research pillars: robust feature learning, open-set recognition, self-supervised learning, multimodal learning, and fusion techniques. These pillars support eight applications: deepfake detection, synthetic media detection, authorship attribution, phishing detection, fact-checking, scientific fraud detection, presentation and injection attack detection, and AI-based detection of CSAM.

[SIA] Recod.ai is currently recruiting for a new post-doctoral fellowship on AI for Digital Forensics, Trust, and Synthetic Realities, which sounds truly exciting. What kind of expertise are you hoping to attract?

[AR] We are seeking candidates with a strong background in Computer Science, Mathematics, or the Exact Sciences, and prior experience with AI, Vision, or Machine Learning. This expert will have the opportunity to interact closely with researchers in Linguistics, Journalism, Social Sciences, and Fact-checking, among others. It’s a great opportunity to be part of a truly multidisciplinary environment.

[SIA] Recod.ai has contributed to studies on research integrity, including work examining image manipulation in scientific publications. What insights did this research reveal?

[AR] The key insight is that where incentives exist, scams and crimes tend to follow. In the scientific community, the culture of publish or perish takes its toll. More and more, people are turning to fraudulent practices in experiments and publications. Our task at Recod.ai and within Project Horus, allied with our partners, is to devise methods to quickly detect such practices directly in a paper or even in the graph collaboration network of researchers.

[SIA] Many of your initiatives integrate computer vision, machine learning, and Social Sciences. Why is this multidisciplinary approach important for advancing trustworthy AI?

[AR] Multidisciplinarity and transdisciplinarity are key in the modern world. Virtually everything relates to BANGS — Bits, Atoms, Neurons, Genes, and Social Challenges. At Recod.ai, we instill in students and researchers the importance of thinking broadly about the social, economic, political, and technical impacts of their research.

[SIA] What does joining the Science Integrity Alliance represent for Recod.ai, and how do you envision contributing to its mission?

[AR] As the Science Integrity Alliance is a global, interdisciplinary organization dedicated to promoting transparency, responsibility, and rigor, it is the perfect ally for designing solutions to curb scientific misconduct. I am sure this partnership will bear important fruits in the near future.