Human-like reasoning for an AI
summary
Amazon is building Alexa features that combine extracted events, a knowledge graph and stateful smart‑home signals to reason across context. Systems use web extraction and knowledge graph extraction with state‑of‑the‑art transfer models plus large LLMs as signals, then resolve ambiguity with device capability checks (e.g., prefer “turn off” when lights are already on) and cohort learning for timing. The stack supports multi‑hop reasoning (aggregate Olympic skiing medals) and future‑scenario planning (can Leclerc still win if Verstappen wins next race), with explanations rather than a bare yes.
AI-written summary · about
chapters (8)
- 0:00Introduction to reasoning
- 3:55Why reasoning matters
- 9:32What is reasoning
- 13:30Challenges at scale
- 18:56Research directions
- 21:24Upcoming innovations
- 26:25Future possibilities
- 29:00Conclusion and Q&A
related talks
39mIt's All in the Data: The ML Behind Alexa's AI Systems
32mHow Alexa keeps getting smarter
42mConversational AI in Amazon Alexa
19mRedefining Intelligence: How Reasoning Is Re-Shaping AI in 2025
18mAlexa's next AI trick: Disappearing
16mBeyond LLMs - The Next Frontier of Machine Reasoning
26mGrounded Reasoning Systems for Cloud Architecture
29mAgentic AI: What It Is and Why It Matters