Mark Zuckerberg: Future of AI at Meta, Facebook, Instagram, and WhatsApp | Lex Fridman Podcast #383

Added: Jul 16, 2023

Mark begins by discussing Meta's vision for the future of AI and its impact on the human world. He believes that more organizations should have the ability to build state-of-the-art technology, not just a small number of big companies. While training AI models requires significant resources, Mark believes that more organizations will have access to it over time. Meta has made much of its infrastructure open source to enable innovation and efficiency. Mark mentions Meta's language model, LAMA, and their approach to open-source research. He acknowledges that LAMA is not as advanced as the biggest AI models, but it still has its uses. Mark believes that open sourcing AI models will lead to better alignment and understanding. The conversation then shifts to the potential impact of AI on small businesses and entrepreneurs. Mark believes that AI-powered products can greatly benefit these groups by providing coding help, design assistance, and models for customer support. With billions of people using Meta's platforms, AI has the potential to make a significant impact. Meta is working on three main categories: personal assistants, business AI, and simpler tools like photo editing. Their goal is to create diverse AI systems tailored to specific needs and styles. Mark emphasizes the importance of an open-source, community-oriented approach to AI. The discussion then turns to the challenges of moderating AI-generated content. Mark acknowledges the potential risks of AI models being used to create harmful narratives and controversies. He explains that Meta has developed sophisticated AI systems to identify and categorize harmful content, such as promoting terrorism or exploiting children. While advances in generative AI may pose new challenges, Mark is optimistic that stronger AI systems can help defend against these harms. Meta also focuses on identifying and taking down coordinated and authentic networks of accounts that engage in harmful behavior. Mark acknowledges the difficulty of moderating content and finding the right balance between freedom of expression and preventing harm. Lex raises concerns about the potential misuse of AI models to coordinate harm on social networks. Mark agrees that training AI models to understand and prevent harm is crucial. Meta aims to ensure that AI models cannot be used to commit crimes or coordinate harmful activities. While no system can be 100% perfect, Mark believes in making it more difficult and inefficient for adversaries to carry out harmful actions. The conversation then delves into the challenges of moderating misinformation and controversial topics. Mark acknowledges the complexity of distinguishing between facts and opinions, particularly in areas with active disagreements. He believes that it's important to focus on the harms that people universally agree on, such as sexual exploitation of children or inciting violence. Mark suggests that flagging information or providing fact-checking notes can help users make informed decisions. However, he acknowledges that there will always be edge cases and challenges in managing misinformation. The discussion then shifts to the pressure Meta faces from governments and interest groups to moderate content. Mark explains that Meta follows principles to allow people to express themselves while also addressing known harms. When governments request content takedowns, Meta evaluates them on a case-by-case basis. Mark emphasizes the importance of protecting users' data from government access, as it can expose people to physical harm. Meta makes strategic decisions about data center locations and prioritizes the protection of users' data. Lex brings up the pressure to censor content from political entities, citing the Twitter files as an example. Mark acknowledges that there are debates and pressures from various sources, including elected officials, politicians, agencies, media, and activist groups. He mentions that Meta can't make decisions to make everyone happy and must find the right balance. Mark's view on freedom of speech has evolved to a more nuanced perspective. While certain things are universally accepted as bad and illegal, there are unsettled issues with active disagreements. He emphasizes the importance of considering the impact on community health and handling things practically. Mark also mentions the benefits of text-based social networks for discussions and conversations and suggests exploring alternative approaches like decentralization. Towards the end of the podcast, Mark briefly discusses his experience in a Jiu Jitsu tournament and the importance of physicality. He believes that being physical and having experiences connected to our bodies is essential. Mark also mentions his belief in creating things and being a positive force in the world, as well as the grounding and humility he finds in religion. Overall, the podcast provides insights into Mark Zuckerberg's vision for the future of Meta and AI, as well as his thoughts on moderation, content regulation, and the challenges of running a company.