I’m a big fan of the pod, very happy to see an episode this week. However, does anyone care about the Fr. Justin AI debacle? Like who cares? Don’t think it is relevant to the regular programming we are accustomed to. Also very frustrating when Ed is making a very salient point and pulls away from the microphone, levels are way off and very hard to hear his salient point.
I found the whole Fr Justin thing very enlightening. It's actually helped me clarify my thoughts on AI and chatbots (which are actually two different, if sometimes related, things.) Ed seems to have landed on roughly the same conclusions I have.
Would Ed be willing to control the character of the whiskey priest in a Nintendo version of The Power and the Glory? Press ‘A’ to wrestle with your vocation, press ‘B’ to hide in the woods.
CA's text threads were impossible to get real information out of. But there have been real priests like Monsignor Pope in Catholic newspapers answering questions.
My immediate reaction to Father Justin was to consider all his plus points.
1. Not continually asking for money. Unlike my Bishop moaning on BBC Radio about his parishioners who each give about the price of a cup of coffee each per week.
2. You can leave him alone with children. Indeed he has no pelvic urges whatsoever, so he can be left alone with anyone of any age, sex, sexual perversion or gender presentation.
3. He is not telling me how wonderful Pope Francis is, unlike my previous PP.
4. He seems to actually know a lot of complex Catholic information.
5. Is available 24/7.
6. No alcohol/drugs/compulsive gambling/reckless use of money. Unlike the priest in the parish next to mine who made romantic advances to the Master of Ceremonies after a few bottles of wine.
7. No embezzling parish funds. Like that priest in Lansing, Michigan who stole enough to build a 5 million buck mansion.
8. Does not bless gay unions.
Fr Justin may have a smirk and an odd voice....but if those were serious problems, I can think of several priests in my diocese who would never be let anywhere near a parish.
When I consider the problems I had persuading a British Telecom Merlin (the name was the only magic thing about it) to calculate a week's Social Security payment, I am awestruck at what Catholic Answers have achieved in this first very imperfect incarnation. The fact that it already provides plausible answers to many queries is a sign of better things ahead. Though I would be happy with a text interface without the cartoon figure.
Only the second scenario is actually likely. The real AI threat (apart from people foolishly increasingly treating machines like people and vice versa) could show up in one of two forms:
1. We trust it more than ourselves and give it control over something vital, like the electric grid or the justice system. Then we either lose the keys and can't override it (because we gave it the keys on the assumption that it would make fewer mistakes than a human) or we've forgotten how to do the things it does because we haven't had to in so long so we're faced with living with the increasingly weird ways AI handles these things or scrapping them altogether and rebuilding them from scratch.
2. We weaponize it against each other in various ways: "Hey, AI program. Figure out a way, using trial and error, to destabilize and crash Country X using social media manipulation. Here are a bunch of Twitter accounts." or "Hey, AI. Exterminate everyone living in Zone X in the most efficient way possible. Here are some drones." or "Hey, AI. Engineer a virus that targets specific ethnic groups. Here are a bunch of lab results and a real time lab."
Computer programs are never going to "achieve sentience."
My mom (who studied computer science back in the day) likes to say computers are just very fast dumb machines. They do exactly what they are told to do. Even automated "analysis" doesn't seem much more sophisticated than counting very fast through lots of data. A computer certainly can't assign meaning to any of it - though it may then enact other behaviors based on its programming, and since it lacks the capacity to care about life (or anything), it wouldn't occur to the program, "maybe I shouldn't shut off the grid here because people could suffer," or "I should ignore this override and not launch the missiles." There isn't an "I" there, and it doesn't have to be malicious intent on the part of the programmer or user - incompetence or hubris will do just fine. It's so fast it defies human scale. That's plenty dangerous on its own. We don't need the thing to be sentient.
Yes. Like a genie in an Arabian Nights-inspired story (not the one who has decided to kill you on the spot because you did not release him 400 years ago and has to be tricked back into the bottle through vanity, but the one who grants you several wishes according to how you phrased the wishes, rather than according to what you meant.)
The uncanny valley of "this is not what I intended *nor* what I envisioned when I was thinking about possible failure modes (how would *a human* fail at this + how have I seen other software fail)" can potentially be as deep as the Mariana Trench, I claim... my thinking has been colored by reading Zendegi a long time ago however.
Re: discussion of Fr. Justin's "common knowledge" discussed around 45 minute mark...
These sorts of projects almost always utilize a commercially available LLM like OpenAI's gpt-3.5. The developers typically don't actually "train" the chatbot on Catholic content -- that's possible but a much more expensive method. More likely, a semantic search is done on the users input against a custom content database of relevant content from Scripture, catechism, Catholic Answers articles, etc. That search returns relevant paragraphs from Catholic content. Those search results, plus the users input/question, plus a prompt from the developers (eg, "You are a Catholic priest named Fr. Justin...") are passed to the LLM for generating a response. That means that anything in the training data behind gpt-3.5 (or whichever model they're using) is on the table for formerly-Fr. Justin.
An element that isn't discussed enough is the lack of transparency there. Not only is user input being passed to OpenAI, but the user isn't interacting with just a Catholic chatbot. There's a lot more in there...the linguistic distillation of the spirit of the age, if you will. Which is why (at least last summer) it was possible for MagisteriumAI (similar project) to output sexual passages from Fifty Shades of Grey, even though the developers claimed to have exclusively "trained" it on a Catholic dataset. Sure, the developers engineer their prompt to try to keep the chatbot on topic, but there's really no telling when inappropriate or inaccurate things can bleed through.
Not to continue to dunk of the CA team, but the transparency/understandability or lack thereof of these Catholic AI projects is an important angle that would be interesting to see The Pillar pursue more. Those elements are major themes in the Rome Call for AI Ethics, but are being overlooked.
Dunking on Fr. Justin is easy enough. There's plenty of outlets capable of handling that (though not always with the same good humor and level heads that JD and Ed brought to the discussion).
But digging into his seminary and incardination background to discover the red flags therein? Where else could we possibly turn for such next-level analysis (which might just contain some lessons with real-world applicability, who knows)?
Easily makes up for the lack of a podcast the prior week, if you ask me.
Am I the only one who wondered how long it would take for JD to get his question answered by cannon law listeners and how many emails he would receive? Even if I am, I would very much like to know. lol
As a CUA alum who had the opportunity to take his Marriage and Annulments course as an undergrad, nothing beats the occasional mentions of Fr. Beal across the podcast episodes. This is the inside baseball Pillar content I’m here for! :)
I’m a big fan of the pod, very happy to see an episode this week. However, does anyone care about the Fr. Justin AI debacle? Like who cares? Don’t think it is relevant to the regular programming we are accustomed to. Also very frustrating when Ed is making a very salient point and pulls away from the microphone, levels are way off and very hard to hear his salient point.
We thought it was interesting inasmuch as it allowed for a broader discussion on AI. ymmv!
Thanks for the response! Nobody bats a thousand but the Pillar is way above the Mendoza line. Still happy to be a paying subscriber
I found the whole Fr Justin thing very enlightening. It's actually helped me clarify my thoughts on AI and chatbots (which are actually two different, if sometimes related, things.) Ed seems to have landed on roughly the same conclusions I have.
Would Ed be willing to control the character of the whiskey priest in a Nintendo version of The Power and the Glory? Press ‘A’ to wrestle with your vocation, press ‘B’ to hide in the woods.
The “to the law” sound track was the best thing ever!!! The only question I have is should it stay consistent or change each time?
Yes! 💯
Thank you for that game! You have cleared a long standing dilemma in my head regarding Oceans 11: https://youtu.be/MPwcqITDwho?feature=shared
You might want to read The Always War by Margaret Peterson Haddix about when computers run things via something resembling AI.
CA's text threads were impossible to get real information out of. But there have been real priests like Monsignor Pope in Catholic newspapers answering questions.
There are people simulating Mass on a video game called Roblox
https://youtu.be/gTN2rXKaa1o?si=Dj3KSTlA-LdBvqNQ
My immediate reaction to Father Justin was to consider all his plus points.
1. Not continually asking for money. Unlike my Bishop moaning on BBC Radio about his parishioners who each give about the price of a cup of coffee each per week.
2. You can leave him alone with children. Indeed he has no pelvic urges whatsoever, so he can be left alone with anyone of any age, sex, sexual perversion or gender presentation.
3. He is not telling me how wonderful Pope Francis is, unlike my previous PP.
4. He seems to actually know a lot of complex Catholic information.
5. Is available 24/7.
6. No alcohol/drugs/compulsive gambling/reckless use of money. Unlike the priest in the parish next to mine who made romantic advances to the Master of Ceremonies after a few bottles of wine.
7. No embezzling parish funds. Like that priest in Lansing, Michigan who stole enough to build a 5 million buck mansion.
8. Does not bless gay unions.
Fr Justin may have a smirk and an odd voice....but if those were serious problems, I can think of several priests in my diocese who would never be let anywhere near a parish.
When I consider the problems I had persuading a British Telecom Merlin (the name was the only magic thing about it) to calculate a week's Social Security payment, I am awestruck at what Catholic Answers have achieved in this first very imperfect incarnation. The fact that it already provides plausible answers to many queries is a sign of better things ahead. Though I would be happy with a text interface without the cartoon figure.
Mr Murphy with the hot take! 😯 🔥
Thank you for not being unhinged about “Fr. Justin.” This was a very interesting listen!
If I understand Ed correctly, AI could be harmful:
In a Sarah Connor situation (as JD said)
or (as I would put it) a Mickey Mouse-Sorcerer's Apprentice situation, with programming (or a spell) gone awry
Only the second scenario is actually likely. The real AI threat (apart from people foolishly increasingly treating machines like people and vice versa) could show up in one of two forms:
1. We trust it more than ourselves and give it control over something vital, like the electric grid or the justice system. Then we either lose the keys and can't override it (because we gave it the keys on the assumption that it would make fewer mistakes than a human) or we've forgotten how to do the things it does because we haven't had to in so long so we're faced with living with the increasingly weird ways AI handles these things or scrapping them altogether and rebuilding them from scratch.
2. We weaponize it against each other in various ways: "Hey, AI program. Figure out a way, using trial and error, to destabilize and crash Country X using social media manipulation. Here are a bunch of Twitter accounts." or "Hey, AI. Exterminate everyone living in Zone X in the most efficient way possible. Here are some drones." or "Hey, AI. Engineer a virus that targets specific ethnic groups. Here are a bunch of lab results and a real time lab."
Computer programs are never going to "achieve sentience."
My mom (who studied computer science back in the day) likes to say computers are just very fast dumb machines. They do exactly what they are told to do. Even automated "analysis" doesn't seem much more sophisticated than counting very fast through lots of data. A computer certainly can't assign meaning to any of it - though it may then enact other behaviors based on its programming, and since it lacks the capacity to care about life (or anything), it wouldn't occur to the program, "maybe I shouldn't shut off the grid here because people could suffer," or "I should ignore this override and not launch the missiles." There isn't an "I" there, and it doesn't have to be malicious intent on the part of the programmer or user - incompetence or hubris will do just fine. It's so fast it defies human scale. That's plenty dangerous on its own. We don't need the thing to be sentient.
> They do exactly what they are told to do.
Yes. Like a genie in an Arabian Nights-inspired story (not the one who has decided to kill you on the spot because you did not release him 400 years ago and has to be tricked back into the bottle through vanity, but the one who grants you several wishes according to how you phrased the wishes, rather than according to what you meant.)
The uncanny valley of "this is not what I intended *nor* what I envisioned when I was thinking about possible failure modes (how would *a human* fail at this + how have I seen other software fail)" can potentially be as deep as the Mariana Trench, I claim... my thinking has been colored by reading Zendegi a long time ago however.
Re: discussion of Fr. Justin's "common knowledge" discussed around 45 minute mark...
These sorts of projects almost always utilize a commercially available LLM like OpenAI's gpt-3.5. The developers typically don't actually "train" the chatbot on Catholic content -- that's possible but a much more expensive method. More likely, a semantic search is done on the users input against a custom content database of relevant content from Scripture, catechism, Catholic Answers articles, etc. That search returns relevant paragraphs from Catholic content. Those search results, plus the users input/question, plus a prompt from the developers (eg, "You are a Catholic priest named Fr. Justin...") are passed to the LLM for generating a response. That means that anything in the training data behind gpt-3.5 (or whichever model they're using) is on the table for formerly-Fr. Justin.
An element that isn't discussed enough is the lack of transparency there. Not only is user input being passed to OpenAI, but the user isn't interacting with just a Catholic chatbot. There's a lot more in there...the linguistic distillation of the spirit of the age, if you will. Which is why (at least last summer) it was possible for MagisteriumAI (similar project) to output sexual passages from Fifty Shades of Grey, even though the developers claimed to have exclusively "trained" it on a Catholic dataset. Sure, the developers engineer their prompt to try to keep the chatbot on topic, but there's really no telling when inappropriate or inaccurate things can bleed through.
Not to continue to dunk of the CA team, but the transparency/understandability or lack thereof of these Catholic AI projects is an important angle that would be interesting to see The Pillar pursue more. Those elements are major themes in the Rome Call for AI Ethics, but are being overlooked.
Great episode! AI is here to stay, like it or not, so Catholics need to start having thoughtful conversations about it.
Terminator 1 >>>> Terminator 2
Dunking on Fr. Justin is easy enough. There's plenty of outlets capable of handling that (though not always with the same good humor and level heads that JD and Ed brought to the discussion).
But digging into his seminary and incardination background to discover the red flags therein? Where else could we possibly turn for such next-level analysis (which might just contain some lessons with real-world applicability, who knows)?
Easily makes up for the lack of a podcast the prior week, if you ask me.
Am I the only one who wondered how long it would take for JD to get his question answered by cannon law listeners and how many emails he would receive? Even if I am, I would very much like to know. lol
As a CUA alum who had the opportunity to take his Marriage and Annulments course as an undergrad, nothing beats the occasional mentions of Fr. Beal across the podcast episodes. This is the inside baseball Pillar content I’m here for! :)