Scott Aaronson: From Quantum Computing to AI Safety
Vložit
- čas přidán 15. 12. 2023
- Subscribe for exclusive content at lawrencekrauss.substack.com/
Learn more and support the foundation at originsproject.org/
A Note From Lawrence:
Scott Aaronson is one of the deepest mathematical intellects I have known since, say Ed Witten-the only physicist to have won the prestigious Fields Medal in Mathematics. While Ed is a string theorist, Scott decided to devote his mathematical efforts to the field of computer science, and as a theoretical computer scientist has played a major role in the development of algorithms that have pushed forward the field of quantum computing, and helped address several thorny issues that hamper our ability to create practical quantum computers.
In addition to his research, Scott has, for a number of years, written a wonderful blog about issues in computing, in particular with regard to quantum computing. It is a great place to get educated about many of these issues.
Most recently, Scott has spent the last year at OpenAI thinking about the difficult issue of AI safety, and how to ensure that as AI systems improve that they will not have an unduly negative or dangerous impact on human civilization. As I mention in the podcast I am less worried than some people, and I think so is Scott, but nevertheless, some careful thinking in advance can avert a great deal of hand wringing in the future. Scott has some very interesting ideas that are worth exploring, and we began to explore them in this podcast.
Our conversation ran the gamut from quantum computing to AI safety and explored some complex ideas in computer science in the process, in particular the notion of computational complexity, which is important in understanding all of these issues. I hope you will find Scott’s remarks as illuminating and informative as I did.
Full Episodes Playlist:
• Ricky Gervais - The Or... - Věda a technologie
I see a podcast with Scott Aaronson, I click. Simple as that!
You finally got Scott Aaronson! It's a Christmas miracle.
Fascinating, intriguing, wonderfully friendly dialogue. And Scott holds his own so well against Lawrence's famous tendency to interrupt and speak long, as they both use the SAME (unintentional) technique! Perfect equilibrium 😄😇
This person has a clean mind and good intentions. He does not try to brag, but rather understands, and he does not seem to be chasing awards. Therefore, you find that his criticism is constructive and not biased.
Scott is right about the chemistry teacher. That sort of pedantry is the norm in schools today and it can really suck all the joy out of learning for precocious kids.
Kudos to Scott for educating the public on these important issues. Another UT man
Thank you both Scott and Lawrence for sharing your time and work, peace
The Depth of Knowledge from both of the gentlemen is beyond belief. It was enlightening
Lawrence, please try to interrupt about 5 times less often. Aaronson is really interesting to listen to, but the constant interruptions come off as rude or as competitive.
My horsey AND my Kingdom for an AI which takes as input a Krauss interview & removes all of his interruptions, replacing them with a random string of exclamation marks which one can skip with an KraussBlocker app.
Excellent interview btw Aaronson is amazing
I will say, I wasn't that bothered by the interruptions. I thought they were all substantive and added something
@@stri8ted In many cases, I agree, but there were times when Krauss talks over Aaronson and then Aaronson stutters, trying to finish his thought.
Krauss is so exceptionally brilliant. But the interviews should not be about him. The interruptions are such that I cannot listen to the full interview. It's a real shame.
@@alexfattoruso100% that’s how I felt as well, he started with humility jokes and complimented Scott smarts which I liked so much because I think Scott is not getting enough credit .. and then the interviewer started to interrupt and say “me, for me, i, .. “ was so annoying
Thank you both for the time and effort.
The biggest insight for me was the idea that the value of quantum computing is not parallel processing but the manipulations of interference patterns.
Thanks for the great interview! Such an excellent communicator; he can explain the hardest concepts as clearly and briefly as possible! Very inspiring!
A small suggestion as I go through this video. It would be great if the production team put timestamps on the video, which has already become customary for long CZcams podcasts, so that we might be able to get to the questions that are of interest to us personally, considering that some of us are a little bit short on time.
Yes, and they don't have to be that detailed. Maybe one per 20 minutes for the long ones.
I didn't understand much but still found it interesting. Cheers.
This one (Scott) is good, the way he thinks, he is no a Hallucinator or thief, you can notice that he is hard worker in a good way.
Fantastic interview Lawrence. I'm a huge fan of Scott, and it was a pleasure to hear more about his upbringing.
I loved seeing how much fun the both of you had. And yes: Lawrence is getting better at interrupting less / letting the guest speak.
1:55:27 comparing Gerard ‘t Hooft to Bob Dylan 😊
2:16:47 😂
2:21:53 Goedel as mystic towards Evolution; but we are the counter example; Dawkins argument
2:31:09 OpenAI and Star Wars vs Life of Brian
2:46:27 “The only problem is, that what a lot of people want is absolutely horrifying.”
He needs to cut down his preamble when the guest is around. Scott interrupted the intro else it would've...yeah definitely improved but could be much better
Excellent chat! Need to bring him back for round 2.
Always very entertaining to listen to Scott!
The guest outtalked the host! Good conversation.
fascinating conversation. could have listened to it all day
Nice view from the window!
I love that all these people have the attention span for a three hour interview...
Great show! Thank you.
This was really good and glad it was long.
Scott Aaronson for Pres. Gosh darn he's smart!
Scott you are very interesting but I wish you could communicate in a mode other than monologue.
Great interview but interruptions are so annoying
Cool video & Fun to listen to.
Video editing however was not on par with all other videos I saw from you.
Why?
A) I saw you eat during the interview (it even was obviously finger licking good at some point)
B) The quick dog moment… ok, why not & sort of funny I guess.
C) Solid 2 seconds Yawning, mouth wide open & uncovered to a point I could see your last meal ;), during the interview while your guest is talking…
Wow! This is sort of funny, but just also surprising in such a good show.
Anyways, still a very nice & interesting talk.
Thanks
& wishing you a great 2024
:)
PS I can totally see how I would love to have a long luch with either of you two, or better yet, both of you.
Learning is so fundamental and a pleasure. Thanks for the podcast. :)
(beware) AI generated
Chapter 1: Introduction and Background of Scott Aronson (0:00-1:10)
- Timestamp: 0:00-1:10
- Summary: Lawrence Krauss introduces Scott Aronson, a highly esteemed mathematical intellect and computer scientist specializing in Quantum Computing and AI safety. Aronson's achievements include receiving the Waterman Prize and working at OpenAI. Krauss compares Aronson's intellectual depth to that of physicist and string theorist Edward Witten.
Chapter 2: Early Interests and Education (1:11-5:24)
- Timestamp: 1:11-5:24
- Summary: Scott Aronson's interest in science and mathematics began at a young age, influenced by his father, a science writer. He showed exceptional mathematical talent, leading to early acceleration in his education. Aronson's schooling experiences were a mix of support for his advanced capabilities and struggles with standard educational structures, including skipping grades and early college enrollment.
Chapter 3: Choosing Computer Science and Quantum Computing (5:25-9:14)
- Timestamp: 5:25-9:14
- Summary: Aronson discusses his decision to pursue computer science over mathematics, largely influenced by his desire to avoid additional language requirements. His attraction to theoretical computer science stemmed from its relative youth as a field and the accessibility of its questions. Aronson's interest in quantum computing was piqued by the field's combination of computational complexity and fundamental physics questions.
Chapter 4: Journey through Academic Institutions (9:15-11:26)
- Timestamp: 9:15-11:26
- Summary: Aronson's academic path led him through various institutions, including the Institute for Advanced Study, where he interacted with prominent figures in theoretical computer science. His focus remained on quantum computing and computational complexity. Personal life decisions, like resolving a two-body problem with his wife, played a role in his career trajectory, leading to positions at major universities like MIT and the University of Texas at Austin.
Chapter 5: Interdisciplinary Contributions and Philosophical Interests (11:27-End)
- Timestamp: 11:27-End
- Summary: Aronson reflects on his interdisciplinary contributions to quantum computing and computational complexity. He engages in philosophical discussions about quantum mechanics and its interpretations. Aronson's unique perspective as a computer scientist in a field dominated by physicists allows him to approach problems in quantum computing differently, contributing to significant advancements in the field.
Chapter 11: Quantum Computing and Theoretical Computer Science (57:39 - 2:04:48)
Timestamp: 57:39 - 58:23
Title: Quantum Computing at the University of Texas at Austin
Summary:
- The speaker discusses their unexpected move to Texas and their positive experience living in Austin.
- The University of Texas at Austin's Computer Science Department had strong theoretical computer science programs, appealing to both the speaker and their wife, who is also a theoretical computer scientist.
- The University lacked significant resources in Quantum Computing but had aspirations to grow in that area.
Timestamp: 58:23 - 1:02:11
Title: The Intricacies and Challenges of Quantum Computing
Summary:
- Quantum Computing is described as a generalization of probability rules, using complex numbers called amplitudes instead of conventional probabilities.
- Quantum mechanics calculates probabilities differently, involving the addition of amplitudes for various paths, then squaring their absolute values to find probabilities.
- Interference effects in quantum mechanics are crucial for understanding quantum behavior, where particles' behavior changes based on whether they are observed.
- The fundamental challenge in Quantum Computing is choreographing a pattern of interference to concentrate amplitudes on the correct answer while ensuring other possibilities cancel each other out.
Timestamp: 1:02:11 - 1:17:00
Title: Potential Applications and Limitations of Quantum Computing
Summary:
- Quantum Computing's primary purpose is proposed as simulating quantum mechanics, potentially aiding in designing new materials or pharmaceuticals.
- The initial excitement around Quantum Computing stemmed from its potential to solve classical problems exponentially faster than classical computers.
- The example of Shor's algorithm for factoring large numbers is discussed, illustrating the potential of quantum computers to break current cryptographic systems.
- However, the practical application of quantum computers in fields like AI, machine learning, and finance remains uncertain and challenging due to the theoretical and engineering complexities involved.
Timestamp: 1:17:00 - 2:04:48
Title: Complexity vs. Computability in Computer Science
Summary:
- Computability theory, founded by Alan Turing, focuses on what problems can or cannot be solved by a computer, regardless of time or resources.
- The halting problem is highlighted as an example of an uncomputable problem, where determining whether a given program will ever stop running is impossible in a general sense.
- Computational complexity, in contrast, deals with the practical aspects of solving problems, specifically how the time or resources needed scale with the size of the problem.
- The field of computational complexity aims to categorize problems based on the feasibility of solving them within reasonable time frames, distinguishing between computationally hard and tractable problems.
Chapter 1: Computational Complexity and Scaling (2:04:53 - 2:07:49)
- Exponential vs Polynomial Scaling
- NP-Completeness
Chapter 2: Classes P and NP (2:07:55 - 2:14:37)
- Definitions and Examples
- P vs NP Problem
- NP-Complete Problems
Chapter 3: AI, Intelligence, and Turing Test (2:14:44 - 2:28:16)
- AI and Human Intelligence
- Turing Test and AI Consciousness
- AI Existential Risks
Chapter 4: AI Safety, Ethics, and Alignment (2:28:29 - 2:38:25)
- AI Ethics vs AI Alignment
- Aligning AI with Human Values
- Ethical Dilemmas in AI
Chapter 5: Practical AI Safety Measures (2:38:42 - 3:02:20)
- Watermarking for AI Outputs
- Challenges in AI Detection
- Future of AI Safety
- Personal Reflections on AI
Thank you!
It might be a good idea to let the person you interview talk? Just a thought.
By the way, with the stutter, I much preferred the solo, large screen shot. I helped me focus. Cheers!
Hi, from Portugal. ......great talk.
Really enjoyed this podcast .
It’s a pleasure to hear these 2 talk.
However Scott seems a bit over enthusiastic to talk most of the time, leaving Lawrence a bit struggling. 😂
Mind is important to process data then transform it into information that has meaning and it’s possible to process the information many times to improve it, humans use computers as tools to represent their logic,….. etc
It is clear that you do not differentiate between data and information! This example may make it easier to understand the difference: Let us imagine that a car was transported to the Stone Age and was seen by humans living in the Stone Age. What would happen?! If the car does not move and is well camouflaged, they may not notice it, but if the car moves, the process of attention and exploration will begin. It has a circular thing instead of feet, makes a sound, speed. If they notice that there is a driver driving it, it will be easier for them to realize that the car is It is a tool and not a living being, and if they get closer and touch it and do notice that it is solid, they may get closer to the fact that it is not a living being, but if it is well camouflaged and is self-driving, this will make it difficult for them... Etc. What is the significance of what was mentioned above?! The purpose is to understand the difference between data and information and how data is transformed into information (analyzing the data and giving it meaning, etc.), etc. computers are tools to represent human logic,…. etc
Those who talk about information such as information as building blocks of the universe,….etc don’t understand what they are talking about and abuse the ignorance among the public,…etc
The future war might not be as simple as AI versus humans. It might be a three-sided war: AI, ordinary humans, and mentally-augmented humans. Alliances of two sides versus the third side could be very unstable.
The host should be more patient and let the guest the chance to finish sentences. The constant collisions are annoying.
It will be remarkable if one day quantum computers will work in a reliable manner, view that we don't understand very well the quantum world neither it's easy to set a superposition computer with quantum particles. Maybe a future AI could help in the engineering of a powerful quantum computer which in turn could lead to a more powerful AI and so on.
Currently looking to do a masters in quantum engineering but its hard in some universities to get in being a electronics engineer working in VLSI industry.
Many ask for Quantum mechanics in Bachelor's which i dont have but did couple of courses but they dont accept those
I'm Lovin' It!
great conversation, thanks for this. I just think that AI purpose is not to reach human level but far beyond. So, it doesn't matter whether conscience exists. It's interesting to know that we still don't know that quantum computing is the way forward.
Would it be safe to say that AI would become conscious when it starts to input/prompt itself?
please, let the guest talk and finish his thoughts! those constant interruptions are pretty painful.
A cellular automata i am skeptical about being able to embodie this in any way tgat doesn't shuv a superposition in through the back door.
This interviewer is arrogant. Like he keeps mentioning he went to Yale and stuff. Nobody really cares since you still have no clue what's going on with this place we live in. Scott was awesome and humble.
With regard to AI Safety, who is it that determines aligning it with human values? Just look at what “human values” have done to Homo sapiens and the world ( as Lawrence has pointed out).
28:47 He just said it like that 🤣
yeah! it seems that mentioning Feynman (criticising Feynman) is a taboo among science communicators), but science communication isn’t science itself!
I know that Feynman was your role model (Guru) he was a good physicist, but not at the level you expect! For example, it cannot be compared to Dirac.
1:00:20 what he said here is correct (statistical approach), and that is the result of the lack of deep understanding.
1:00:00 that’s correct, and they can’t understand the source of high accuracy they talk about!
computational complexity is esoteric??
For the simple example of 120 degree separated axis for 3 directions you need to change the distribution from 1/3s uu 2/3s ud, for the first axis being up, to uu and du being 50 50, instead of the thirds, the problem for entanglement in bell paris means you have to change the distribution for both particles upon the first measurements. That requires an interaction it is as simple as that.
36:20 "took a little bit of physics". There is no such thing... sheldon disapproves😅
Did an advanced AI program combined with a highly advanced quantum computer system create our universe?
- What is the similarity between a dog and an engineer?
- They both have an intelligent look, but they don't know how to express themselves.
36:05 :)
What is real? The facade put or the conversationalist who had a good time?
AI got me a little worried
Oof Scott's example of Yudkowski's coherent extrapolated volition should scream loudly and clearly why you should never listen to anything Yud has to say.
2:19:00 WELLLL..... that little ditty that Scott came up with in 2006 or so turned out to be an exceedingly poor analogy in light of recent advances of AI. Nowadays any schlub CAN type in at a prompt and trivially create a master work equivalent to a great symphony or work masterpiece painting. OOF!
52:25 I've always had that intuition and that all of it is much ado about nothing The height of intellectual arrogance is "if we don't understand this, neither do you little people do". But ordinary folks often have the intuition that things are just "second order differential equations".This is why you end up with Donald Trump, a man who also built buildings for a living and walks around with the intuitions of his construction workers. Buildings are not very forgiving if you use crappy steel and concrete. As his old friend Oprah Winfrey said, we all walk around with our own version of the truth.
That was hard work and I had to stop. Lawrence was obviously competing. Why have him on if youre going 5o interrupt so much? Insecure? Try watching Lex Friedman for how it's done.
As a fan of Scott myself, I would be very disappointed if he spends too much time in AI safetyqhich seems a questionpf ethics rather than science, computing. If he sticks to quantum computing, complexity and fundamentals of AI it would benefit us all. Commenting within first 15 min so I am sure he outlines his reasons for this interest in safety
Perhaps you consider him a role model in the media field (you call that Science communicator ).
Don’t mix things up!
Super Mews
Go super hands ...
Lenny is getting old, get him on!
I will write you a paper to show you how in detail how a " classical" model can handle bell. But there is no possible way without an interaction between the particles, it isnt a conspiracy because the measurment direction ks interacting with the particle. So lets call this kind of interaction a bell interaction and argue about the details of whats possible or natural about that instead lf predetermined value being possible.
This guy is a horrible interviewer he keeps on interrupting him and trying to prove how smart he is and talk about his accomplishments, very turned off by this. Kept saying most people don't understand this
He looks like Jared fogle.
mr.Lawrence Please let the guest talk. You talk and interject too much please. If you can't stop please do not do podcasts. I am sorry you never let complete the thought process of the guest thinking that you know their thoughts but you don't.
Krauss defended Epstein as the financier tried to restore his reputation after his release from jail " I don't feel tarnished in any way by my relationship with Jeffrey. I feel raised by it. " After Epstein's release, krauss took his money to bankroll the origins project at Arizona state university. Epstein on record gave krauss $250,000 nevermind what he gave off record
the reductionist mindset, must always and always serve technology.
Does these privileged men even imagine the preposterous emissions caused by AI
Pathetic...