Babak Hodjat, CTO for AI at Cognizant and a pioneering force behind Siri, explores the evolution of multi-agent AI. He discusses his early work with natural language systems and how they've transformed into complex agentic structures. Babak emphasizes the importance of transparency and trust in autonomous systems, and the challenge of AI's black box. He also touches on the need for interoperability standards as well as the fascinating dynamics of agent communication, including whether they develop their own languages.
46:33
forum Ask episode
web_stories AI Snips
view_agenda Chapters
auto_awesome Transcript
info_circle Episode notes
question_answer ANECDOTE
Origins of Siri's Multi-Agent System
Babak Hodjat and his team built an extensible, language-independent multi-agent system for natural language understanding behind Siri.
The system focused on semantics over grammar, enabling it to understand non-grammatical commands effectively.
insights INSIGHT
Power and Limits of Language Interfaces
Language is the superior mode of human interaction and communication, enabling the creation of new concepts.
However, natural language understanding alone is not the be-all, end-all interface; multimodal interaction is more effective.
volunteer_activism ADVICE
How to Use Large Language Models
Modern large language models should be contained within defined roles or "boxes" for extensibility and manageable interaction.
Multi-agent systems require coordination and team formation mechanisms to break organizational silos effectively.
Get the Snipd Podcast app to discover more snips from this episode
Babak Hodjat, CTO for AI at Cognizant and a pioneer behind the technology that became Siri, joins us to discuss the new era of agentic AI. We dig into his early days building natural language systems, the evolution of multi-agent architectures, and how language interfaces are shaping the future of human-AI collaboration. We also discuss the safeguards needed to ensure trust in autonomous systems, the challenge of AI’s black box, and the emerging standards for interoperability.
Enjoying the AI Inside podcast? Please rate us ⭐⭐⭐⭐⭐ in your podcatcher of choice!
Note: Time codes subject to change depending on dynamic ad insertion by the distributor.
CHAPTERS:
00:00 - Podcast begins 01:45 - Introducing Babak Hodjat and his journey from Dejima to Siri 02:47 - The origins of agent-based AI and early natural language systems 06:21 - Language as the ultimate interface: strengths and limitations 09:35 - How modern AI capabilities could have changed early voice agents 13:08 - Agentic systems versus traditional APIs in enterprise tech 16:10 - How Cognizant builds and deploys agent-based AI for clients 18:18 - The need for interoperability and emerging agent standards 20:49 - Do agents develop their own languages? Internal communication in LLMs 23:01 - Multilingual models, cultural context, and emotional abstraction 26:49 - Can AI truly understand meaning? Perspectives on abstraction 27:50 - Safeguards, kill switches, and trust in agentic automation 32:05 - Marketing, influence, and negotiation in an agent-driven world 35:07 - The risks of over-trusting black box AI systems 36:47 - The evolution of AI: what’s truly new versus what’s rediscovered 39:26 - Thank you to Babak Hodjat for joining the AI Inside podcast
AI Inside 