Episode
Gary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
1:25:09
Published: Thu Oct 03 2019
Description
Gary Marcus is a professor emeritus at NYU, founder of Robust.AI and Geometric Intelligence, the latter is a machine learning company acquired by Uber in 2016. He is the author of several books on natural and artificial intelligence, including his new book Rebooting AI: Building Machines We Can Trust. Gary has been a critical voice highlighting the limits of deep learning and discussing the challenges before the AI community that must be solved in order to achieve artificial general intelligence. This conversation is part of the Artificial Intelligence podcast. If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter, LinkedIn, Facebook, Medium, or YouTube where you can watch the video versions of these conversations. If you enjoy the podcast, please rate it 5 stars on iTunes or support it on Patreon. Here's the outline with timestamps for this episode (on some players you can click on the timestamp to jump to that point in the episode): 00:00 - Introduction 01:37 - Singularity 05:48 - Physical and psychological knowledge 10:52 - Chess 14:32 - Language vs physical world 17:37 - What does AI look like 100 years from now 21:28 - Flaws of the human mind 25:27 - General intelligence 28:25 - Limits of deep learning 44:41 - Expert systems and symbol manipulation 48:37 - Knowledge representation 52:52 - Increasing compute power 56:27 - How human children learn 57:23 - Innate knowledge and learned knowledge 1:06:43 - Good test of intelligence 1:12:32 - Deep learning and symbol manipulation 1:23:35 - Guitar
Chapters
Gary discusses the challenges and limits of AI, including the possibility of a technological singularity and the need for healthy debate to arrive at greater understanding.
00:00 - 02:06 (02:06)
Summary
Gary discusses the challenges and limits of AI, including the possibility of a technological singularity and the need for healthy debate to arrive at greater understanding.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
As AI develops and machines take over more tedious and dangerous tasks, our work and lives will change.
02:06 - 07:02 (04:56)
Summary
As AI develops and machines take over more tedious and dangerous tasks, our work and lives will change. People will find meaning through creative expression and AI will continue to improve gradually.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
In this episode, the guest discusses how machines are learning from various data inputs and relying on self-correction to make better decisions.
07:04 - 14:24 (07:19)
Summary
In this episode, the guest discusses how machines are learning from various data inputs and relying on self-correction to make better decisions. He also talks about the possibility of machines understanding human goals and how AlphaGo is a prime example of machines being able to improve on their own through deep learning.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The debate about the complexity of language and the physical world is ongoing.
14:24 - 18:20 (03:56)
Summary
The debate about the complexity of language and the physical world is ongoing. While some argue that the physical world is more complicated, others believe that language has the power to capture and describe everything in the world.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The ultimate AI will not look just like humans, but it can perform tasks that are currently done better by humans, such as reasoning and understanding language.
18:20 - 24:34 (06:13)
Summary
The ultimate AI will not look just like humans, but it can perform tasks that are currently done better by humans, such as reasoning and understanding language. The limits and flaws of the human mind may be a strength that motivates and gives meaning to our actions.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The idea of machine solving problems that humans can't with the use of advanced computing might oversimplify the idea of common sense as it is actually a very complicated concept.
24:34 - 33:08 (08:34)
Summary
The idea of machine solving problems that humans can't with the use of advanced computing might oversimplify the idea of common sense as it is actually a very complicated concept.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
Deep learning has its limitations and we need to do hard engineering work in order to build a design specification for what we want the system to do.
33:08 - 43:02 (09:53)
Summary
Deep learning has its limitations and we need to do hard engineering work in order to build a design specification for what we want the system to do. Although deep learning can accomplish a lot automatically, humans still need to guide and teach the machine.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
There is a need to combine expert systems and deep learning to come up with new technologies that can capture human knowledge stored in symbols and basic level information that is difficult to capture in machines.
43:02 - 49:27 (06:24)
Summary
There is a need to combine expert systems and deep learning to come up with new technologies that can capture human knowledge stored in symbols and basic level information that is difficult to capture in machines.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
Deep learning has been successful in certain areas, but it has not resulted in significant progress in natural language understanding or common sense reasoning.
49:27 - 56:57 (07:30)
Summary
Deep learning has been successful in certain areas, but it has not resulted in significant progress in natural language understanding or common sense reasoning. AI has yet to figure out how to perform the functions that human brains do effortlessly with very little energy consumption.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
This podcast discusses the relationship between innate knowledge and learning in early childhood development, particularly in how children set their own problems to solve.
56:59 - 1:01:28 (04:29)
Summary
This podcast discusses the relationship between innate knowledge and learning in early childhood development, particularly in how children set their own problems to solve.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The evolutionary process of creating a pool of ideas to discover one that can be run away with may be necessary to build effective intelligence systems.
1:01:28 - 1:06:45 (05:16)
Summary
The evolutionary process of creating a pool of ideas to discover one that can be run away with may be necessary to build effective intelligence systems. Researchers may need to study the biology of thought and understanding to truly create successful systems.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The Walter White test, named after the main character in the TV show Breaking Bad, is a way to test artificial intelligence by requiring it to understand complex narrative storylines.
1:06:45 - 1:11:58 (05:12)
Summary
The Walter White test, named after the main character in the TV show Breaking Bad, is a way to test artificial intelligence by requiring it to understand complex narrative storylines. The test is part of the Turing test, which measures a machine's ability to pass as a human in conversation.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
This episode explores how deep learning can encompass symbol manipulation and the role of classical AI techniques, such as Monte Carlo Tree Search.
1:11:58 - 1:17:14 (05:16)
Summary
This episode explores how deep learning can encompass symbol manipulation and the role of classical AI techniques, such as Monte Carlo Tree Search.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
The deep learning community and general public need to gain a common understanding of AI to use it practically in daily life.
1:17:14 - 1:24:06 (06:51)
Summary
The deep learning community and general public need to gain a common understanding of AI to use it practically in daily life. Despite this, many people misunderstand AI, and the book The Deep Learning Revolution highlights this issue.
EpisodeGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI
PodcastLex Fridman Podcast
In this episode, Gary talks about his favorite rock song to listen to and also to play.
1:24:06 - 1:24:56 (00:49)
Summary
In this episode, Gary talks about his favorite rock song to listen to and also to play. He also shares his choice of rock star to party with and pick their mind over a weekend, living or dead.