"Google has made efforts to create an artificial intelligence (AI) that understands not only the languages of each country but also cultural and social contexts. The goal is to become an AI used by people around the world, not just in the Western world, through learning models that comprehend unique events, traditions, and religions of each cultural sphere."
Manish Gupta, Senior Director of Google DeepMind, said during a press conference for 'Google for Korea' held at the Chosun Palace Hotel in Gangnam, Seoul, on the 2nd. He noted, "Google has been researching AI for a long time, well before the public release of AlphaGo in 2016, and what we learned from this research is that AI must also learn like humans." He emphasized, "In the future, we will have powerful digital assistants that help with various tasks customized to each person's reality."
Director Gupta cited 'AlphaFold,' a protein structure prediction system developed by Google DeepMind. He said, "In the past, it took 3 to 5 years of research and expensive experimental equipment to unveil the three-dimensional structure of a new protein, but now, through AlphaFold, we can predict structures in just seconds." He further stated, "We have already predicted the structures of over 200 million known proteins worldwide using AI and made this data available for free to more than 2 million researchers, and we are making progress in various fields such as cancer treatment and new drug development."
Director Gupta introduced Google's next-generation AI agent project 'ASTK'. This project aims to create a multimodal AI agent that understands and utilizes real-time video, audio, and text to provide information naturally and perform tasks in users' daily lives. ASTK leverages the agent framework embedded in Google's latest large language model (LLM), 'Gemini 2.0' and is connected to various services such as Google Search, Maps, and Lens.
Director Gupta emphasized the efforts to make AI understand cultural aspects beyond simple language comprehension. He stated, "What Google aims for is not merely an AI that provides information, but a companion that understands human situations and contexts and can respond appropriately," adding that "YouTube is one of the major sources." He continued, "We are currently working on training the model through YouTube to learn how people speak in different tones and nuances, and how to interact with superiors, transcending English-speaking contexts."
Director Gupta introduced products such as ▲ the image generation model 'Imagine 4' ▲ the video and audio generation model 'Vio 3' ▲ and the music composition model 'Lydia' based on Gemini. He emphasized that these AI models are tools to complement humans rather than replace them. Director Gupta stated, "AI models are designed with the premise of collaboration with artists" and that they help expand creators' imaginations and explore new modes of expression beyond merely producing content. He added, "Artists can attempt various things with AI that they could not do before," stating, "This is not a tool for replacing people, but a tool for enhancing capabilities."
Simon Tokumine, Director of Google Labs, shared the latest updates on the AI research agent 'Notebook LM' and the AI web agent 'Mariner Project'. He stated, "Notebook LM automatically organizes the core concepts of up to 300 documents uploaded by users at once and generates a structured mind map by topic," explaining that this includes features like 'Audio Overview' which converts this into natural language-based audio and 'Video Overview' which adds AI voice explanations to slide images to maximize users' learning capabilities. Additionally, the 'Mariner Project' is an AI agent that performs online tasks automatically by understanding user characteristics without requiring users to individually visit and click on websites.
During the press conference, there was also criticism that Google's generative AI model 'Gemini' has lower usage rates in the country compared to its competitor OpenAI's 'ChatGPT'. In response, Tokumine noted, "We are still in the early stages, and there is a lot of room for improvement." He added, "What we discovered during the internet and mobile eras is that when there is a fundamental change called 'platform shift,' there were applications in the early stages, but many applications were developed thereafter, realizing greater user value," and stated, "In the generative AI boom, more products will be developed, and Google is keeping up well."