Dylan Hadfield-Menell, UC Berkeley/MIT: The value alignment problem in AI

1

Introduction

00:00 • 2min

2

Controlling Da I Systems in a Robotic Environment

02:22 • 2min

3

Is There a Subjective Versus Objective Goal in Machine Learning?

04:41 • 3min

4

Is This a Principal Agent Problem in Artificial Intelligence?

08:01 • 4min

5

Is There a Loop of System Behavior?

12:07 • 2min

6

How to Build Better Signals of What You Value

14:04 • 5min

7

Optimizing Agents - What's the Biggest Problem?

18:50 • 3min

8

What Would You Change About the Stock Market?

22:08 • 5min

9

Up Dating Ai Systems in a Production Setting

27:19 • 3min

10

I Think There's a Lot of Intelligence in How We Manage the Optimization Component of Ourselves

30:36 • 2min

11

Is Machine Learning a Good Idea?

32:20 • 6min

12

Creating a Data Set That Really Gets You There?

37:58 • 4min

13

How Do We Communicate in Groups Effectively?

42:02 • 6min

14

How Do You Filter Resumes?

47:45 • 4min

15

How Do You Evaluate Your Processes?

52:15 • 3min

16

The General Purpose Programming Language in Machine Learning?

55:31 • 4min

17

Is It Possible to Measure Temperature?

59:04 • 3min

18

Is There a Market in Efficiency?

01:02:29 • 2min

19

How to Scale Control of a Global Internet Platform

01:04:09 • 3min

20

Getting People to Think About Their Values and Values

01:06:57 • 3min

21

Delegated Recommendations

01:10:16 • 2min

22

Machine Learning and Information Diets

01:11:56 • 2min

23

Do You Really Need a Predictive Model of How Things Will Change?

01:13:45 • 3min

24

The Incomplete Contracting Line of Work in Machine Learning

01:16:23 • 2min

25

How Unsupervised Learning Is Progressing?

01:18:33 • 3min

26

Unsupervised Learning and Manipulation

01:21:42 • 4min

27

Imitative Learning

01:25:43 • 4min

28

Is There Anything That You Want to Put Out There?

01:29:30 • 3min