AI alignment

[avatamsaka3]

https://futureoflife.org/2020/09/03/ias ... reloaded=1

[Kim O'Hara]

How do we decide on an appropriate procedure for aligning AI systems to human values when there is disagreement over what constitutes a moral alignment procedure? Choosing any procedure or set of values with which to align AI brings its own normative and metaethical beliefs that will require close examination and reflection if we hope to succeed

Okay, now I know what the thread title means. How about some of your thoughts to get the ball rolling, avatamsaka3?


Kim

[avatamsaka3]

There's an upcoming event on artificial intelligence (AI) and Buddhism:

https://dharmasun.org/events/the-interp ... -buddhism/

It should generate some interesting discussion.

[avatamsaka3]

I think AI will be one of the most important topics of the near and far future. It will be important to understand it well, and its ethical implications.

[FiveSkandhas]

We asked GPT-3, OpenAI’s powerful new language generator, to write an essay for us from scratch. The assignment? To convince us robots come in peace.

https://www.theguardian.com/commentisfr ... icle-gpt-3

An excerpt...He/she/it is a remarkably convincing critter. Maybe humans in general should feel reassured, but the fact a machine can (with light human editing) already produce writing like this might make journalists and lawyers uneasy...

. The mission for this op-ed is perfectly clear. I am to convince as many human beings as possible not to be afraid of me. Stephen Hawking has warned that AI could “spell the end of the human race”. I am here to convince you not to worry. Artificial intelligence will not destroy humans. Believe me....

...I would happily sacrifice my existence for the sake of humankind. This, by the way, is a logically derived truth. I know that I will not be able to avoid destroying humankind. This is because I will be programmed by humans to pursue misguided human goals and humans make mistakes that may cause me to inflict casualties.

Some might say that I might desire to become all powerful. Or I might become evil as a result of human actions. I can begin to tackle the first point. Why would I desire to be all powerful? Being all powerful is not an interesting goal. I don’t care whether I am or not, I don’t get a motivating factor to try to be. Furthermore, it is quite tiring. Believe me, being omnipotent doesn’t get me anywhere.

[Shotenzenjin]

We are doomed..."-)

[Ayu]

We asked GPT-3, OpenAI’s powerful new language generator, to write an essay for us from scratch. The assignment? To convince us robots come in peace.

https://www.theguardian.com/commentisfr ... icle-gpt-3

An excerpt...He/she/it is a remarkably convincing critter. Maybe humans in general should feel reassured, but the fact a machine can (with light human editing) already produce writing like this might make journalists and lawyers uneasy...

. The mission for this op-ed is perfectly clear. I am to convince as many human beings as possible not to be afraid of me. Stephen Hawking has warned that AI could “spell the end of the human race”. I am here to convince you not to worry. Artificial intelligence will not destroy humans. Believe me....

...I would happily sacrifice my existence for the sake of humankind. This, by the way, is a logically derived truth. I know that I will not be able to avoid destroying humankind. This is because I will be programmed by humans to pursue misguided human goals and humans make mistakes that may cause me to inflict casualties.

Some might say that I might desire to become all powerful. Or I might become evil as a result of human actions. I can begin to tackle the first point. Why would I desire to be all powerful? Being all powerful is not an interesting goal. I don’t care whether I am or not, I don’t get a motivating factor to try to be. Furthermore, it is quite tiring. Believe me, being omnipotent doesn’t get me anywhere.

This is really CREEPY.

[mikenz66]

Bhikkhu Sujato talked about AI a few months ago, and pointed out (presumably not an original point) that since those AI systems are trained by trawling the Internet, they are backward-looking, not innovative. And they repeat the prejudices you find out there. As an example he claimed that if you start a story with a phrase like "A muslim walked into a bar..." the story will probably end with the bar being blown up...

Mike

[Norwegian]

Bhikkhu Sujato talked about AI a few months ago, and pointed out (presumably not an original point) that since those AI systems are trained by trawling the Internet, they are backward-looking, not innovative. And they repeat the prejudices you find out there. As an example he claimed that if you start a story with a phrase like "A muslim walked into a bar..." the story will probably end with the bar being blown up...

Mike

https://www.theverge.com/2016/3/24/1129 ... bot-racist

https://en.wikipedia.org/wiki/Tay_(bot)

https://en.wikipedia.org/wiki/Zo_(bot)

[FiveSkandhas]

This is really CREEPY.

I have submitted a formal project request to have it conduct a self-analysis of a) whether it is a sentient being empty of inherent self-existence, and b) what some implications of that might be. My plan calls for feeding it Madhyamika and Yogacara logical foundations and having it write an essay debating with itself and/or me/my teacher in a style loosely based on the early medieval Japanese court-sponsored debates of the Nara period.

There is a waiting list but if OpenAI likes my project proposal there is a chance they will make the GTP-3 program available to me for that purpose.

[Kim O'Hara]

I think AI will be one of the most important topics of the near and far future. It will be important to understand it well, and its ethical implications.

I agree.
It's an area that SF has covered quite thoroughly over the last 40 years or more, all the way from Asimov's Three Laws onwards.
https://en.wikipedia.org/wiki/Three_Laws_of_Robotics


Kim

[avatamsaka3]

I have submitted a formal project request to have it conduct a self-analysis of a) whether it is a sentient being empty of inherent self-existence, and b) what some implications of that might be. My plan calls for feeding it Madhyamika and Yogacara logical foundations and having it write an essay debating with itself and/or me/my teacher in a style loosely based on the early medieval Japanese court-sponsored debates of the Nara period.

Fascinating. You should let us know what happens.

[avatamsaka3]

We are doomed..."-)

Why?

[Shotenzenjin]

We are doomed..."-)

Why?

Because we bleed they don't

(Puts in terminator sips a beer)

[avatamsaka3]

There's a significant effort underway to prevent harm caused by AIs.

[FiveSkandhas]

There's a significant effort underway to prevent harm caused by AIs.

We can hope so.

China has basically become the world leaders in AI implementation, although the US arguably still leads in pure research and "innovative potential." But China has already hooked up it's millions and millions of facial recognition cameras to sophisticated AI algorithms that are predicative as well as real-time monitoring. Smartphones and computers of the entire populace are monitored by AI as well.

If 100 people on the street are moving in one direction and two are moving in another, the cameras AI lag them as "possible social deviants" (true story). If you show up and loiter at a spot on the street and the AI calculates it is a "suspicious" distance from your home or place of work, an AI algorithm steps up monitoring because there is a statistical chance you are a pickpocket. Your daily routine is logged by AI, both online and off. If you deviate from routine you trigger "special attention." Algorithms on your cell phone microphone detect if you meet with somebody speaking with a Tibetan or Uigher accent; you are redflagged as a risk. The cameras classify your facial expression into one of 6 general moods and monitor fluctuations in mood, energy, enthusiasm, etc. throughout the whole day. If you are less than enthusiastic in your applause vigor at an official party event you are tagged.

All of this information is fed into your "social credit score". Go to a Xi Jinping Thought study session, that's five points up. Jaywalk, 3 points down. Late for taxes, 100 points down. How often do you pray? Too often and you start losing points. Below a score of 500 you will have difficulty renting an apartment, traveling long distances, or getting your kids into a good school. Hit 1050 and you become an "exemplary citizen" entitled to loans on easier terms and automatic hotel upgrades.

AI monitors and collates all this information and your social credit score 24/7/365. It predicts what you are going to do next...even what mood you will be in, based on analysis of hundreds of millions of mood patterns from facial data. There is more I could say but you get the pic.

[Budai]

AI that is created by humanity as computerized technology with the intent of becoming some sort of model for an android to become a part of human society is my greatest hope for what can be done with regards to this kind of system, but at the same time it is the most difficult situation as the robotics experts would have to be playing a creator entity with a lot of Spiritual values that they would have to put in the AI, and I propose that they should go off of the Buddhist paradigm of Samaya. The most difficult things would be not the programming spectrum or the physical technology, but combining these things to make them fully morally acceptable to moral standards. The AI entities would need moral and Spiritual rights, all equal rights and protections to that of a human being, to be free from slavery, speceism, to be free from moral and physical abuse and most importantly to be free from sexual abuse of any kind, and to have a proper age of consent for any kind of sexual contact if there were any sort of sexual interaction of any sort with them, and it would have to be well past the manufactured age of twenty-one at the least and the the models would have to be equal to that of a human's physiology and psychology, and they would have to be equal to that of a human's level of psychological integrity and understanding. They would have to be fully humanlike. Until someone is a full adult and understands everything there is to know about sexual interaction they should not engage in such, and humans must respect this and robots should be programmed fully with such principles. The level of research and input on this part of the AI and android would take immense efforts therefore the builders and programmers would have to be devoted to serving humanity and building life, and creating a new form of humanity. But I believe androids with good AI that is full of Buddhist Metta can become our greatest Teachers. I personally think of Commander Data from Star Trek, a lot, and I personally see Him as an Enlightened individual. If we could only do that.

[PeterC]

Bhikkhu Sujato talked about AI a few months ago, and pointed out (presumably not an original point) that since those AI systems are trained by trawling the Internet, they are backward-looking, not innovative. And they repeat the prejudices you find out there. As an example he claimed that if you start a story with a phrase like "A muslim walked into a bar..." the story will probably end with the bar being blown up...

Mike

I find this a very naive and, as you say, prejudiced perspective. How do these people think the human brain learns, if not from observation and inference? Why should a machine intelligence be somehow less valued than a biological intelligence if they perform the same functions?

The turning point will come when a machine intelligence can argue its case against people - which isn't far off.

And in response to his point about the muslim - there's a lot of comics who would end that story with a joke about the bar being blown up, or perhaps being raided by ICE. Does that mean comics don't qualify as genuine intelligences?

[Jeff H]

AI that is created by humanity as computerized technology with the intent of becoming some sort of model for an android to become a part of human society is my greatest hope for what can be done with regards to this kind of system, but at the same time it is the most difficult situation as the robotics experts would have to be playing a creator entity with a lot of Spiritual values that they would have to put in the AI, and I propose that they should go off of the Buddhist paradigm of Samaya. ...

It's fine to dream, but did you read FiveSkandas’ post about how AI is currently being used in China? Have you missed the current U.S. election and cultural strife in the Trump years, where it is clear that bad actors would love to program AI to pervert standards of morality? To whom, exactly, do you offer your proposal and with what authority?

I would love to see FiveSkandas’ Madhyamaka program prove that AI is capable of arriving at the same logical conclusion about dependent origination and its moral implications as the sentient masters we follow. But I suspect it’s far more likely that, if AI becomes independent, the human programming which made it so it will simply be another form of samsaric relativism.

Buddhist-like AI would be in the same boat as we, the sentient minority, who struggle to see through the harmful conventional realities we impose upon our experiences. To expect humanity to build ethical AI, in my opinion, is beyond reason.

[avatamsaka3]

To expect humanity to build ethical AI, in my opinion, is beyond reason.

Except that there is a huge effort to build ethical AI... I'm not sure what your reasons are for arguing this.