If AI systems can't explain their own decisions, what hope do we have of controlling them? Roman Yampolskiy argues that the unexplainability of AI isn't a temporary technical problem — it may be a...
If AI systems can't explain their own decisions, what hope do we have of controlling them? Roman Yampolskiy argues that the unexplainability of AI isn't a temporary technical problem — it may be a permanent one.
Roman V. Yampolskiy is a tenured Associate Professor of Computer Science at the University of Louisville's Speed School of Engineering, where he founded and directs the Cyber Security Lab. Widely credited with coining the term "AI safety" in 2011, he is the author of AI: Unexplainable, Unpredictable, Uncontrollable.
Watch more Closer To Truth videos on AI & Transhumanism
Get access to over 5,000 videos by signing up for a free Closer To Truth membership
Support Closer To Truth
Explore more from Roman Yampolskiy on Closer To Truth
Follow Closer To Truth
Closer To Truth, created and hosted by Robert Lawrence Kuhn, presents the world's greatest thinkers exploring humanity's deepest questions. Discover fundamental issues of existence and sentience. Engage new and diverse ways of thinking. Appreciate intense debates. Share your own opinions. Seek your own answers.
Closer To Truth, created and hosted by Robert Lawrence Kuhn, presents the world's greatest thinkers exploring humanity's deepest questions. Discover fundamental issues of existence and sentience. Engage new and diverse ways of thinking. Appreciate intense debates. Share your own opinions. Seek your own answers.
#AISafety #ArtificialIntelligence #CloserToTruth
0:00 Introduction
0:16 The unexplainability triad: unexplainable, unpredictable, uncontrollable
0:32 Why large language models are black boxes
1:04 The problem with lossy compression of AI decisions
1:43 Why AI explanations may be plausible lies
2:44 The split-brain analogy: when systems confabulate
4:31 Mechanistic interpretability: what research can and can't tell us
5:02 Why we need AI to understand AI — a catch-22
6:04 Incomprehensibility vs. unverifiability: key distinctions
7:49 The infinite regress of verifiers
8:15 Is unexplainability permanent — and what follows from that?
Roman V. Yampolskiy is a tenured Associate Professor of Computer Science at the University of Louisville's Speed School of Engineering, where he founded and directs the Cyber Security Lab. Widely credited with coining the term "AI safety" in 2011, he is the author of AI: Unexplainable, Unpredictable, Uncontrollable.
Watch more Closer To Truth videos on AI & Transhumanism
Get access to over 5,000 videos by signing up for a free Closer To Truth membership
Support Closer To Truth
Explore more from Roman Yampolskiy on Closer To Truth
Follow Closer To Truth
Closer To Truth, created and hosted by Robert Lawrence Kuhn, presents the world's greatest thinkers exploring humanity's deepest questions. Discover fundamental issues of existence and sentience. Engage new and diverse ways of thinking. Appreciate intense debates. Share your own opinions. Seek your own answers.
Closer To Truth, created and hosted by Robert Lawrence Kuhn, presents the world's greatest thinkers exploring humanity's deepest questions. Discover fundamental issues of existence and sentience. Engage new and diverse ways of thinking. Appreciate intense debates. Share your own opinions. Seek your own answers.
#AISafety #ArtificialIntelligence #CloserToTruth
0:00 Introduction
0:16 The unexplainability triad: unexplainable, unpredictable, uncontrollable
0:32 Why large language models are black boxes
1:04 The problem with lossy compression of AI decisions
1:43 Why AI explanations may be plausible lies
2:44 The split-brain analogy: when systems confabulate
4:31 Mechanistic interpretability: what research can and can't tell us
5:02 Why we need AI to understand AI — a catch-22
6:04 Incomprehensibility vs. unverifiability: key distinctions
7:49 The infinite regress of verifiers
8:15 Is unexplainability permanent — and what follows from that?
Comments 0
Be the first to comment!