Discussing the Future of Artificial Intelligence with Global AI Scholars

Samsung Electronics hosted the Samsung AI Forum 2024 from November 4 to 5 (KST), engaging with global AI scholars to discuss the future of artificial intelligence (AI).

Marking its 8th year, Samsung AI Forum serves as a platform for technological exchange, where internationally renowned scholars and experts in the fields of AI and computer engineering (CE) are invited to share the latest research findings and explore future research directions.

This year’s keynote speakers included prominent figures in AI, such as Professor Yoshua Bengio of the University of Montreal, a leading authority in deep learning; Yann LeCun, the Chief AI Scientist for Facebook AI Research (FAIR) and a professor at New York University; and Ian Horrocks, Professor at the University of Oxford, renowned for his expertise in knowledge graph technology.

“AI is transforming our lives at an astonishing pace, and as it becomes more powerful, questions on responsible AI usage is becoming increasingly important,” said Jong-Hee (JH) Han, Vice Chairman, CEO and Head of Device eXperience (DX) Division at Samsung Electronics. “Samsung Electronics is committed to to building a more efficient and sustainable AI ecosystem.”

Day One: Sustainable Innovation with AI and Semiconductors

On November 4, Samsung Advanced Institute of Technology (SAIT) hosted the first day of the forum under the theme “Sustainable Innovation with AI and Semiconductors.” Attendees delved into research findings on AI and CE technologies to strengthen next-generation semiconductor development and production capacities.

Professor Bengio delivered a keynote presentation titled “Bayesian Oracles for AI Safety,” highlighting the need to create safer, more reliable AI systems. He shared research findings demonstrating that AI performance has exceeded human levels in various fields while addressing potential future risks associated with AI, such as significant impacts on the job market, AI-based hacking, and the emergence of superintelligence. “To ensure AI safety, it is crucial for policymakers and the general public to have an accurate understanding of the current state and future of AI,” he emphasized.

Professor Bengio also stressed the need for securing AI safety through several key measures, including designing AI systems with built-in safeguards to prevent dangerous behavior, aligning AI actions and goals with human values, and promoting increased coordination and cooperation in AI competition among nations and corporations.

As a 2018 Turing Award recipient and a globally recognized AI scholar, Bengio has been actively involved in the Samsung AI Forum since its launch in 2017. Since 2020, he has also served as a “Samsung AI Professor,” collaborating closely with Samsung Electronics in academia-industry partnerships. He explained the current capabilities and limitations of large language models (LLMs) and emphasized that further technological innovations are necessary for machines to reach human-level intelligence.

In the technical session, Joseph Macri, CTO and Vice President of Advanced Micro Devices Inc.’s (AMD), gave a presentation on the topic of “AI Everywhere,” introducing AMD’s AI solutions and emphasizing the significance of AI platforms, collaboration, and AMD’s strengths.

Additionally, Young Sang Choi, a Master at SAIT, engaged in discussions with the speakers on AI technology trends and the future direction of AI in semiconductors.

The Samsung AI Researcher of the Year awards, aimed at recognizing outstanding up-and-coming researchers in the field of AI, were also bestowed during the forum. Five AI Researchers including Professor Shuran Song from Stanford University, received the prestigious honor. Each awardee also delivered a presentation at the forum.

Professor Shuran Song specializes in applying AI across various aspects of robotics and has been highly recognized for her contributions to developing recognition and inference algorithms for 3D modeling.

Day Two: Device AI for Daily Lives

The second day of the forum, organized by Samsung Research, focused on the theme “Device AI for Our Daily Lives.”

“The advancement of generative AI technology is accelerating the everyday transformation of device AI,” said Daehyun Kim, Executive Vice President and Head of the Global AI Center at Samsung Research, in his welcome speech. “I hope this forum will serve as a platform to discuss and share new possibilities in the AI era” he added.

Professor Ian Horrocks from the University of Oxford, co-founder of Oxford Semantic Technologies (OST), delivered a keynote presentation, “Recommendation & Personalization Using Knowledge Graphs.”

In July, Samsung Electronics acquired OST, a company renowned for its world-class knowledge graph technology. Professor Horrocks explained the significance of the knowledge graph and how it stores and processes data like human memory and recall.

He also introduced key applications of knowledge graph systems, such as search and recommendation, which effectively demonstrate their features. He captured the audience’s attention by presenting approaches for implementing personalized services based on flexible data models and logical reasoning.

Changkyu Choi, Vice President of the AI Research Center at SAIT, gave a presentation on “AI for Science,” stating that “AI and semiconductor technology are not only improving our lives but also playing a crucial role in the field of science by reducing development time and costs.” He added, “AI is particularly beneficial in cases where experimental data is limited or material synthesis is challenging.”

Haejun Lee, a Master at Samsung Research, explained that efficient training of LLMs requires high-quality data, optimized architectures, and stable training techniques. He highlighted that these elements can simultaneously improve both cost efficiency and performance. He further emphasized that reusing previous models can enhance the efficiency and performance of training new models, highlighting that these methods are essential considerations in language model development.