What on earth does it mean to build AI software?

on YouTube ↗29m

" summary

Defines building AI software as delegating decision-making to machines, shows a concrete agent-based framework (passive U1, active/proactive U2, autonomous with motivations), compares model-driven vs data-driven approaches, cites Newtonian physics as designed models and convolutional nets (Hinton 2012) as learned ones, warns about faulty validation (mugshot criminality dataset), and points to a practical Google Maps hybrid that combines explicit models with learned traffic prediction.

AI-written summary · about

" chapters (8)

1. 0:00Start
2. 0:33The developer's conundrum: what on earth does it mean to build Al software?
3. 10:30How do we describe and build systems where decision-making is delegated to machines?
4. 20:08When we delegate decision-making to machines we need a model of the world.
5. 20:42Design the model
6. 24:07Validate the model
7. 25:49Applications are hybrids
8. 27:01Did you successfully delegate decision-making to software?

" related talks

• 03mInterviewXin Wei Ngiam · AI By the Bay 2017
• 59mThe evolution in AI thinking and products of the next decadeShailesh Kumar · Anthill Inside 2018
• 03mInterviewGabor Melli · AI By the Bay 2017
• 29mAgentic AI: What It Is and Why It MattersOmar Sedeek · DSC MENA 2025
• 11mAutomation of AI: Accelerating the AI RevolutionRuchir Puri · O'Reilly Artificial Intelligence Conference 2019 - New York, New York
• 46mFullstack AI: The Building Blocks for Production Ready AI-Powered SoftwareSebastian Nilsson · AI-fokus 2026
• 36mArtificial Intelligence: The Future of SoftwareMatthew Renze · AntarctiConf 2020
• 20mDecision Intelligence: The Next Phase in AICassie Kozyrkov · TechChill 2019