TLDR: Appen and AI Circle recently hosted over 70 AI professionals in Mountain View, California, to discuss the development of safe, scalable, and multilingual AI. The event featured experts from Google, Meta, Amazon, NVIDIA, Microsoft, and Appen, who delved into the evolving role of Human-in-the-Loop (HITL), challenges in AI evaluation, and the future of foundation model fine-tuning. A key focus was on the critical need for human expertise in building culturally nuanced AI and addressing the risks associated with autonomous AI agents.
Mountain View, California – On June 18, 2025, a significant gathering of over 70 artificial intelligence professionals convened in Mountain View, California, for an event co-hosted by Appen and AI Circle. The primary objective of the assembly was to delve into the complexities and strategies involved in building AI systems that are not only safe and scalable but also proficient in handling multiple languages.
The event featured a distinguished panel of experts from leading technology companies, including Google, Meta, Amazon, NVIDIA, Microsoft, and Appen. These industry stalwarts led insightful discussions on a range of critical topics shaping the future of AI development. NVIDIA’s Pratik Mehta moderated the panel, which explored the evolving dynamics of Human-in-the-Loop (HITL) processes, the inherent challenges in red teaming and evaluating AI models, and the shifting landscape of fine-tuning for advanced foundation models.
Appen’s latest research into the persistent difficulties of multilingual AI translation was also a central point of discussion, underscoring the company’s commitment to addressing these complex issues. The lineup of influential speakers included Si Chen, VP of Strategy and Marketing at Appen; Himanshu Gupta, Applied Scientist at Amazon; Mark Hoffmann, Staff ML Engineer at Meta; Arsalan Mosenia, Tech Lead for AI Agents at Google; and Mayur Shastri, Software Engineer at Microsoft.
Key takeaways from the discussions highlighted several crucial aspects of current AI development. Firstly, the role of Human-in-the-Loop (HITL) is undergoing rapid evolution. Experts noted that HITL is no longer confined to simple annotation tasks but has expanded into a multi-layered process encompassing subject-matter expert validation, iterative relabeling, and comprehensive quality loops across diverse use cases, including legal data and low-resource languages. Si Chen of Appen emphasized that HITL is fundamental to developing AI that respects linguistic and cultural nuances, particularly across the more than 500 languages and 200 markets Appen serves. She also pointed out the significant challenge of identifying and training the appropriate experts and effectively capturing their knowledge for AI models to learn from.
Also Read:
- Anthropic Unveils Targeted Transparency Framework for Advanced AI Systems
- RAISE Summit 2025: Enterprise AI Infrastructure Takes Center Stage in Paris
Secondly, the panel addressed the emerging frontier of AI agents, acknowledging their immense potential alongside considerable risks. Google’s Arsalan Mosenia discussed the industry’s progression from agent frameworks to fully autonomous agents capable of executing enterprise tasks end-to-end. However, Mosenia cautioned that widespread adoption would not be straightforward due to lingering concerns regarding security, reliability, and the potential for error compounding within agentic systems. These challenges are particularly critical in sensitive applications where AI safety is paramount, necessitating a cautious approach to deployment.
