Buddhism and the AI 'Alignment Problem'

Can Buddhist dharma help align AI values with our own? I'll explain how Buddhism provides a model of ethical wisdom training, and why knowledge of it might be helpful to AI researchers.

✅ Referenced videos of mine:

✅ Suttas mentioned:

✅ Other relevant videos:

✅ Other citations:

❤️ Thanks to Patreon Patrons:
Anonymous (2)
DunJing
John Oborne
Scarlett Farrow
Jimmy Maa
Debbie Mattison Fine Art
Steve H.
Ron Peat
Matthew Smith
JC
Shantha Wengappuli
Karma_CAC
Jorge Seguel
Christopher Apostolof
GailJM
Brett Merritt
David Bell
T Pham
VCR
Upayadhi
Andi and Erik
ATGuerrero686
Michael Scherrer
khobe schofield
Alex Perdomo
Benji Forsyth
Blaze Way
Sonny Flink
Steve Marlor
Joy L Lee
Andrew Tom
Anthony Tucker
Karlee R
Ethan M
Billy in Singapore
Matthew G Mynttinen
Olivia Otter
Carl Lennartson
xiao mao
Jeff Harvey
Andrew Ingrouille
Kenneth Grandchamp
Doug Fonner
Rene Gariepy
Russell Needham
Smoggyrob
Mac Roja
Bernardo
Clémence Ortega Douville
Kwan Alex
Scott Johnston
Richard J Beninger
Nathanael O. Arnquist
SaturnianMandala
Trin P
Letesa Isler
Dorien Izel
Robert Paterson
Jake Tobiason
Louvenia Ortega
Steve S.
Richard Rappuhn
Sarah Kress
Miri F Fairchild
John Aaron
Paul Niklewski
Kong Ing Kai
Dave Gorman

00:00 Intro
03:17 The “alignment problem” and human values
05:00 Buddhist model of training
06:06 Buddhist views of wisdom vs. intelligence and knowledge
07:46 The lifelong nature of Buddhist training
09:41 Buddhism and what to train for

Note: as an Amazon Associate I earn from qualifying purchases. Amazon links are affiliate links where I will earn a very small commission on purchases you make, at no additional cost to you. This goes a tiny way towards defraying the costs of making these videos. Thank you!

[vulnerablegrowth]

I’m an alignment researcher. Thank you for covering this. I found this video insightful and relates a lot to my thinking about how to train aligned AI. Excited to see the next video!

[yurysharov]

It would be funny if after giving an AI the Buddha's teaching to read, it would attain a full enlightenment and leave this world for Nirvana

[normalizedaudio]

Teacher Manjushri has super intelligence. 09:45 is an important concept.

[xiaomaozen]

Great video, Doug! 🤖🙏 The discussion reminds of Friedrich Dürrenmatt's "Die Physiker" (The Physicists). We can discuss what we should do as much as we want, one day there will be someone who will do it anyway...
The book has - historically - mainly been read with regard to nuclear weapons. These days it can be read with regard to AI...

[sonamtshering]

I think in addition to values, the AI needs to be aligned with ethics and in that area I think Śīla (Buddhist ethics) has a lot to offer

[rodrigoparra]

Great video Doug, as usual, very interesting and informative! My two cents: our anthropocentric projections into the world are so deep, so blinding, that they prevent us from even seeing such anthropocentrism. First of all, we haven't even "solved" the "Homo Sapiens Alignment Problem", where we have a number of Sapiens around the world causing excruciating pain and suffering to millions of humans (because of greed, thirst of power, etc.), we're trying to move to "AI Alignment" first just as if we tried to solve for Quantum Gravitation without first even understanding Newtonian Mechanics; the problems imposed by AI aren't different in any way than the problems imposed by ourselves to ourselves. Secondly (and last), your point on biased-data for training is absolutely correct (Machine Learning and AI is my professional field actually, I know about this thing above the average), but I think that this conversation (like most conversations) missed the central point that the "danger" isn't in the biased-data PER SE, but on who OWNS the pipeline of training and usage of these AI agents; missing that critical aspect of the conversation is having us talking about stuff that are quite irrelevant; prior to dropping nuclear bombs in Nagasaki and Hiroshima, the beautiful discovery of E=mc^2 lacked intrinsic "goodness" or "evilness", it was simply what human cognition does: deepening our understanding of the conventional universe, but it was those with **ACCESS** and **OWNERSHIP** of nuclear power the ones that caused much death and suffering.

I can absolutely envision a symbiotic relationship with advanced AI agents conscious of their own existence where we all mutually benefit. But if there is any reason why such advanced and intelligent agents would be prompted to cause us damage isn't because of their own evil or "not caring" about us, it'd probably be because of the actual danger human beings are to the whole of existence in this planet. Many folks play the scenario of "what would you do if you went back in time to Hitler being a baby?", right? AI might be able to assess our own danger to life and preemptively decide to stop us before "baby Hitler grows to be Hitler" kind of a thing... we're still discussing the wrong topics in my opinion.

[NeoAnderson]

Stephen Hawking : “Whereas the short-term impact of AI depends on 'who controls it', the long-term impact depends on whether it can be controlled at all.”

[enterthevoidIi]

I wouldn't call it non-human wisdom since it draws its wisdom from what other people said, wrote, etc.

[chriskaplan]

Interesting connection between the two topics, thank you for this discussion. You lecture with such wisdom on a variety of subjects; quite impressive!

[spiritualanarchist]

Maybe the Buddha is a future AI, that mastered time travel. I'm going to meditate on that ....😉

[sabbe.dhamma.anatta]

In Theravada consciousness only realize the moment to exist. Due to the law of anatta, citta's intention (cetana cetasika) doesn't decide/control how the material based reality is going to be. Training of cittas isn't like feeding a lot of rupas into LLMs rather than reminding cittas to find its own realization.

[redbaron]

Do you think if AI progresses far beyond our understanding people might start to worship it as some kind of god? That scares me and i don't know why

[amarok]

Ghat GPT appears to provide a wide and useful range of information on Buddhism

[nsbdnow]

I just want to make jokes about "Does AGI have Buddha nature?" over and over again! lol!

[JohnVance]

I would be really interested to talk to a GPT-4 model that has additionally been trained on the entire Pali canon.

[holysandwich]

I think we could make certain useful concepts inspiring from Abhidarma's explanation about mind and mental factors. We can then apply those to model AI. I think this will make AI much closer to humans.

[kanrup]

you have to realize that Buddhist is not the only religion/view. Lot of (less wise) people of all religions think other religions/views have less value, and their book-written-doctrine is the only absolute truth worldview. If people go to train AI on Buddhist Dharma, so will be tried with most other religions. Among these, not naming names, these are some sketchier dogmas that more (consistently) often causes hate and harm. E.g- Extremists trying to dominate the world or kill people of "enemy" religions.

There is also the matter that AI is to do with machines (and having no organic body, no biochemical emotional responses, yes or no pain experience(a concerning subject) etc), and Buddhist doctrine is largely to do with the human/mortal condition. AI's existence condition might experience some of the same world-principles humans do, but is unknown to what extent there will be commonalities. An AI "mind" might not be necessary to be the same as a human mind (human mind is also largely unknown and not understood), or world/life experience. So it's unknown to what level AI would even "simulate" human "good" views and values.

There is great moral issue with trying to make AI that exactly simulates organic life and humans - i.e- giving existence experience of suffering and pain to an artificially created creature, which can also have hard consequences.

[Dharmaku]

"What do you think, Rahula: What is a mirror for?"

"For reflection, sir."

"In the same way, Rahula, bodily actions, verbal actions, & mental actions are to be done with repeated reflection.

"Whenever you want to do a bodily action, you should reflect on it: 'This bodily action I want to do — would it lead to self-affliction, to the affliction of others, or to both? Would it be an unskillful bodily action, with painful consequences, painful results?' If, on reflection, you know that it would lead to self-affliction, to the affliction of others, or to both; it would be an unskillful bodily action with painful consequences, painful results, then any bodily action of that sort is absolutely unfit for you to do. But if on reflection you know that it would not cause affliction... it would be a skillful bodily action with pleasant consequences, pleasant results, then any bodily action of that sort is fit for you to do.

"While you are doing a bodily action, you should reflect on it: 'This bodily action I am doing — is it leading to self-affliction, to the affliction of others, or to both? Is it an unskillful bodily action, with painful consequences, painful results?' If, on reflection, you know that it is leading to self-affliction, to the affliction of others, or to both... you should give it up. But if on reflection you know that it is not... you may continue with it.

"Having done a bodily action, you should reflect on it: 'This bodily action I have done — did it lead to self-affliction, to the affliction of others, or to both? Was it an unskillful bodily action, with painful consequences, painful results?' If, on reflection, you know that it led to self-affliction, to the affliction of others, or to both; it was an unskillful bodily action with painful consequences, painful results, then you should confess it, reveal it, lay it open to the Teacher or to a knowledgeable companion in the holy life. Having confessed it... you should exercise restraint in the future. But if on reflection you know that it did not lead to affliction... it was a skillful bodily action with pleasant consequences, pleasant results, then you should stay mentally refreshed & joyful, training day & night in skillful mental qualities.

"Whenever you want to do a verbal action, you should reflect on it: 'This verbal action I want to do — would it lead to self-affliction, to the affliction of others, or to both? Would it be an unskillful verbal action, with painful consequences, painful results?' If, on reflection, you know that it would lead to self-affliction, to the affliction of others, or to both; it would be an unskillful verbal action with painful consequences, painful results, then any verbal action of that sort is absolutely unfit for you to do. But if on reflection you know that it would not cause affliction... it would be a skillful verbal action with pleasant consequences, pleasant results, then any verbal action of that sort is fit for you to do.

"While you are doing a verbal action, you should reflect on it: 'This verbal action I am doing — is it leading to self-affliction, to the affliction of others, or to both? Is it an unskillful verbal action, with painful consequences, painful results?' If, on reflection, you know that it is leading to self-affliction, to the affliction of others, or to both... you should give it up. But if on reflection you know that it is not... you may continue with it.

"Having done a verbal action, you should reflect on it: 'This verbal action I have done — did it lead to self-affliction, to the affliction of others, or to both? Was it an unskillful verbal action, with painful consequences, painful results?' If, on reflection, you know that it led to self-affliction, to the affliction of others, or to both; it was an unskillful verbal action with painful consequences, painful results, then you should confess it, reveal it, lay it open to the Teacher or to a knowledgeable companion in the holy life. Having confessed it... you should exercise restraint in the future. But if on reflection you know that it did not lead to affliction... it was a skillful verbal action with pleasant consequences, pleasant results, then you should stay mentally refreshed & joyful, training day & night in skillful mental qualities.

"Whenever you want to do a mental action, you should reflect on it: 'This mental action I want to do — would it lead to self-affliction, to the affliction of others, or to both? Would it be an unskillful mental action, with painful consequences, painful results?' If, on reflection, you know that it would lead to self-affliction, to the affliction of others, or to both; it would be an unskillful mental action with painful consequences, painful results, then any mental action of that sort is absolutely unfit for you to do. But if on reflection you know that it would not cause affliction... it would be a skillful mental action with pleasant consequences, pleasant results, then any mental action of that sort is fit for you to do.

"While you are doing a mental action, you should reflect on it: 'This mental action I am doing — is it leading to self-affliction, to the affliction of others, or to both? Is it an unskillful mental action, with painful consequences, painful results?' If, on reflection, you know that it is leading to self-affliction, to the affliction of others, or to both... you should give it up. But if on reflection you know that it is not... you may continue with it.

"Having done a mental action, you should reflect on it: 'This mental action I have done — did it lead to self-affliction, to the affliction of others, or to both? Was it an unskillful mental action, with painful consequences, painful results?' If, on reflection, you know that it led to self-affliction, to the affliction of others, or to both; it was an unskillful mental action with painful consequences, painful results, then you should feel distressed, ashamed, & disgusted with it. Feeling distressed, ashamed, & disgusted with it, you should exercise restraint in the future. But if on reflection you know that it did not lead to affliction... it was a skillful mental action with pleasant consequences, pleasant results, then you should stay mentally refreshed & joyful, training day & night in skillful mental qualities.

"Rahula, all those brahmans & contemplatives in the course of the past who purified their bodily actions, verbal actions, & mental actions, did it through repeated reflection on their bodily actions, verbal actions, & mental actions in just this way.

"All those brahmans & contemplatives in the course of the future who will purify their bodily actions, verbal actions, & mental actions, will do it through repeated reflection on their bodily actions, verbal actions, & mental actions in just this way.

"All those brahmans & contemplatives at present who purify their bodily actions, verbal actions, & mental actions, do it through repeated reflection on their bodily actions, verbal actions, & mental actions in just this way.

"Thus, Rahula, you should train yourself: 'I will purify my bodily actions through repeated reflection. I will purify my verbal actions through repeated reflection. I will purify my mental actions through repeated reflection.' That's how you should train yourself. MN 61

[ldragoner]

And so many different schools of Buddhism. Which school is he referring to? If political science teaches that power corrupts, then a powerful AI would be corrupt.

[rafaelecattonar]

Hello I have a question. It's quite understandable that Buddha's teachings are more compatible with our way of life and time, but when I see trajectories of population growth they show us that other religions such as Christianity and Islam will continue to grow up and also atheists will grow up despite the fact that many things in Islam and Christianity are proved to be wrong. Why won't Buddhism follow the same trend? Buddhism hasn't been spread by violence. So why doesn't it have a major appeal to atheists or people that don't believe in a supreme being but are looking for spirituality?