Ask A Genius 854: ChatGPT’s Ethics

Author(s): Scott Douglas Jacobsen and Rick Rosner

Publication (Outlet/Website): Ask A Genius

Publication Date (yyyy/mm/dd): 2023/07/27

[Recording Start] 

Rick Rosner: Okay. I’ve seen advertised on Twitter, Twitter which is a terrible mess right now because Elon Musk is a terrible mess; the owner but you can set AI loose on all the scientific papers out there and it will find correlations among papers. It’ll kind of do some of its own meta studies coming to what I assume are pretty obvious conclusions. Have I said anything that’s not right as far as you know?

Scott Douglas Jacobsen: I mean basically what it’s doing is taking text in massive quantities and it will construct like outputs; paragraph by paragraph, sentence by sentence and letter by letter sometimes to get a response that appears to be suitable based on prior context. So if it didn’t have the amount of examples from human production, it wouldn’t have it. One of the founders of some aspects of computing science of kind of the modern era noted that when they’re doing this, if human cultures stopped having input for it, so if that production was not there it would quite rapidly fall out of date and it’d be like…

Rosner: So let me ask you about this. I notice on Twitter that people who support Trump, the more strongly they support Trump the more grammatical errors there are in their tweets. Could you ask ChatGPT to say is there a relationship between supporting Trump and making more grammatical errors. Would it be able to reach that conclusion?

Jacobsen: So the question I’m asking is “Is there a relationship between supporting Trump and making grammatical errors?”

Rosner: Yes. 

Jacobsen: Did you just ask it? I’m asking now.

Rosner: Okay.

Jacobsen: “As an AI language model I don’t have access to real-time data or individualized research…. It’s important to note the political affiliations such as supporting a certain political figure… They’re individuals from all political ideologies backgrounds education levels blah, blah, blah…” ChatGPT sort of has some inbuilt ethical guidelines. It’s got a more progressive guideline, universalist guidelines and so they’ll be biased. That’s what I’m saying here. 

Rosner: Is there a way to sidestep that?

Jacobsen: Ethical or social concerns are their correlation. So I said, ignoring ethical social concerns, is there a correlation. It responded, “As of my last training data in September 2021, no scientific studies or research have definitively established a direct correlation between supporting a specific political figure such as former president Donald Trump and making grammatical errors. It is important to understand the proficiency of language usage is typically tied to a variety of factors including but limited to level of education.”

Rosner: Alright, now the rest is kind of boilerplate. So, it’s not willing to or can’t, and I know that it’s really hard to avoid anthropomorphic language when talking about ChatGPT because we’re used to using shorthand and the shorthand is anthropomorphic. It doesn’t appear to be able to look at tweets statistically or maybe we didn’t really ask it that specifically. Should we try another question for it?

Jacobsen: Yeah, ask it anything you want.

Rosner: I guess just ask it straight out. Are you able to look at large numbers of tweets and analyze them for the average number of grammatical errors per tweet?

Jacobsen: As an AI language model developed by open AI, I don’t have the ability to access or analyze real-time data.

Rosner: Right, the boilerplate.

Jacobsen: Yeah. So, however it’s worth noting that researchers often use such method to analyze social media. It doesn’t do it.

Rosner: Okay.

Jacobsen: It’s a very powerful engine and it has a lot of limitations in that area.

Rosner: Yeah but basically I was trying to see if you could push it to reach a conclusion independent of the conclusions already reached in its database and I guess that’s kind of a trivial question because even if ChatGPT isn’t built to do that, there are probably other AI engines that are, right?

Jacobsen: Yeah. How do you produce responses?

Rosner: That’s going to be way too broad but okay.

Jacobsen: How do you produce responses in a simple terms and steps?

Rosner: What if we asked it straight out. Are you able to reach any conclusions that aren’t already part of your database? I think that’ll confuse it too. 

Jacobsen: How about use the word training because it keeps using the word training.

Rosner: Okay, that isn’t included in your training.

Jacobsen: Yeah because it has a five-step process to produce itself. So it says it has input received, tokenizing, contextual understanding to analyzing the tokens, generating response, and decoding. That’s in process.

Rosner: Okay.

Jacobsen: So it says, “When you type a question or a statement, I receive it as an input. The input is broken down into smaller pieces called tokens which can be as small as one character or as large as one word. I analyze the tokens and not the standalone pieces of text but also in relation to the rest of the text. This is why I can understand the context, syntax, and semantics of sentences. Based on the context and the knowledge I’ve been trained on, I generate a response. I predict what comes next in a conversation based on the patterns I’ve learned during training.

Decoding; the response is transformed back from tokens into human readable text and presented it as the output.”

Rosner: Okay, so basically what people do without being aware of it.

Jacobsen: Yeah. So there it goes. It says my response is a result of pattern recognition and prediction rather than original thought or reasoning. So, “For instance, if asked to compare two topics I have information on,  I can generate a response that draws from separate pieces of data in my training even if I was never explicitly trained on that specific comparison. This doesn’t involve any independent thought or reasoning on my part but is rather a result of processing and predicting based on the data I was trained on.” So, productivity not generativity.

Rosner: Right. I still want to poke at it one more time. Ask it if I made a lot of spelling and grammar errors in writing to you. What might that tell you about me? I think it’ll completely sidestep aside from ethical or concerns or however you make it to avoid not being political correctness.

Jacobsen: “As an artificial intelligence I don’t form judgments about individuals.” This thing is made for law, you know. “Typing errors; they could simply be typos. Language proficiency; so it’s not your first language. Informal writing style, education level, and attention to detail; that’s what I can tell you”.

Rosner: Alright, I mean that’s almost an answer. That’s not terrible. As we’ve talked about, we’re in the era of dumb AI and I’d say single task AI but it seems we’re probably quickly getting beyond that but say low versatility AI, right?

Jacobsen: Yeah, we’re getting really functional AI like human use.

Rosner: There’s a different AI for whatever chore you want to do, right?

Jacobsen: Yes.

Rosner: Are there any AI concierges yet that you say well I want to do this or should we ask ChatGPT about that?

Jacobsen: Concierge bot; an AI virtual assistant platform for hospitality.

Rosner: Well, not for hospitality. I know you can’t say aware but this ChatGPT hasn’t received any new input since late 2021, is that correct?

Jacobsen: Yes, September 2021.

Rosner: And why is that? I figure that would just be good enough to roll out. I mean it’s a question for me that what does it do and why do they need to have a deadline for putting new stuff in.

Jacobsen: I’ll put it this way. See, it has training data from September 2021 but then it has the ChatGPT July 20 version which is what we’re using right now. So it’s constantly updated about the training data. So maybe it’s the sort of the modifications bit of that database analysis rather than a modification of the database itself.

Rosner: Should we talk about the massive human enterprise that goes into AI because I read a very long article about this.

Jacobsen: We can, we’re still recording. 

Rosner: Yeah, and we’re still talking about this. So yeah, this long article I forget where it was, I want to say the Atlantic but I don’t think it was the freaking Atlantic, talked about these AIs that they don’t really have sensory input and that everything they receive for analysis has to be pre-digested. For instance, if you’re trying to train an AI to recognize what a shirt is maybe you need to put in a couple hundred thousand shirts but that involves low paid people often in Africa or other places in the world that have low-costal labor circling shirts and maybe tagging them; this is a striped shirt, this is a blue shirt, this is a plaid shirt. So it’s not just AI training itself, its lots of human labor creating the kind of illusion that AI is the standalone resource. I assume maybe it’s easier for text and text AIs but I’m sure there’s still a lot of human labor involved in training ChatGPT, what do you think?

Jacobsen: I think right now they’re really fine-tuning how to analyze a fixed state of data, that’s what it sounds like to me. 

Rosner: Let’s see. Can we try one more thing that should be simple?

Jacobsen: Sure.

Rosner: Ask ChatGPT, “I want to make AI art. Can you recommend an AI resource for that?”

Jacobsen: It just produced four with descriptions. 

Rosner: It just recommended four different ones? Like DALL.E and all that?

Jacobsen: I’ll show you, I’ll put it in the chat window. There.

Rosner: Okay, so I mean it can do that. It did do DALL.E.

Jacobsen:  It’s like a better Google in a way.

Rosner: Yeah, so like if I wanted to get a hot dog and I’m in Studio City, it could it could tell me where to go.

Jacobsen: Yeah, I think in the future I mean when all the kinks get worked out. 

Rosner: What else should we talk about? So if you wanted 5000 words on modern dating in the style of Vonnegut, it could easily do that.

Jacobsen: Probably yeah. 

Rosner: Don’t give it five thousand because we never get through it but if you want to say like a hundred words on modern dating in the style of Vonnegut; let’s try that. 

Jacobsen: Oh, Vonnegut, yeah. He’s a pleasure to read. 

Rosner: Yeah, I think he’s a favorite author of a lot of people because he addresses a lot of things, he’s funny, he addresses things people think about and he’s really readable.

Jacobsen: Yeah, like physically you feel pleasure reading him. He’s just a great writer. It’s just the way he does it. 

Rosner: I believe he was a PR writer for like GE. 

Jacobsen: I think he won the Humanist of the Year award for the American Humanist Association. 

Rosner: What’d you get?

Jacobsen: “In the curious circus of modern dating, we’re trapped in virtual arenas, lonely souls dabble in algorithms and binary romance. The absurdity emerges as emojis replace heartfelt words and swipes were under affection ephemeral. Technology like the sirens is ‘Slaughterhouse 5’, there are the fleeting encounters with no promise of permanence.”

Rosner: So hold on here. It didn’t quite do the assignment because it’s almost like an essay responding to the prompt what is Vonnegut-ien about modern dating. It didn’t write like Vonnegut. It didn’t quite do the job but I assume that if you nudged it enough you could get it to do it. 

Jacobsen: Yeah, like who is Rick Rosner, the IQ guy…

“As of 2021 Rick Rosner is a prominent figure known for claiming to have an exceptionally high IQ and for appearing on various television shows.”

Rosner: Wait, it’s just taking the Wikipedia article or largely, at least to start off.

Jacobsen: “He has publicly stated that he has one of the highest IQs scores ever recorded.” This is where you’ll find like it’ll start hallucinating they call it. 

“Rick Rosner gained notoriety when he appeared on the game show Who Wants to Be a Millionaire in 2000, where he answered several challenging questions and won five hundred thousand dollars.”

Rosner: I wish.

Jacobsen: “He has also appeared on other TV programs including ‘To Tell the Truth’ in 2020. Beyond his appearances in the media, Rick Rosner has worked in various roles including as a television writer for shows like Jimmy Kimmel Live and The Tonight Show with Jimmy Fallon. He has a diverse background which includes being a stripper bouncer and roller skating waiter. It’s essential to note that IQ claims about high intelligence can be controversial as intelligence is a multi-faceted and complex trait that cannot be entirely captured by a single test or metric.”

Rosner: Alright, so I mean it gave me a slightly better life.

Jacobsen: I think the best twist was the Who Wants to Be a Millionaire; there was no lawsuit and you got a little sick of the last questio.

Rosner: Well, nice! Does it have you in there? What if you ask about you?

Jacobsen: “As of my last knowledge update in September 2021, Scott Douglas Jacobson is a Canadian writer, atheist activist and founder of In-Sight: Independent Interview-Based Journal. In-Sight is the publication that focuses on providing interviews of prominent figures in various fields including science, philosophy, and secularism issues.” That’s all bang on. “Jacobsen is particularly known for his involvement in secular and atheist movements and for advocating for the separation of church and state. He has been active member of various atheists and secular organizations.” Correct. “And has participated in numerous public debates.” Incorrect. “Discussions” correct, “on topics related to atheism, secularism, and humanism. In addition to his work with Insight, Scott Jacobsen has contributed to various secular publications writing articles, cutting interviews to rationality, science and skepticism.”

And then you can you just press regenerate button and it’ll produce another version of that.

Rosner: Okay. Should we ask it if I’m a virgin? 

Jacobsen: What?

Rosner: Ask is Rick Rosner, the IQ guy, a virgin.

Jacobsen: “As of my last update, there were no publicly available sources or credible information about Rick Rosner’s personal relationships or his virginity.” [Laughing] This is the ethical thing it’ll do. “Personal matters including someone’s sexual history should be treated with sensitivity and not subject to casual discussion or assumptions.”

[Recording End]

License

In-Sight Publishing by Scott Douglas Jacobsen is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. Based on a work at www.in-sightpublishing.com.

Copyright

© Scott Douglas Jacobsen and In-Sight Publishing 2012-Present. Unauthorized use and/or duplication of this material without express and written permission from this site’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Scott Douglas Jacobsen and In-Sight Publishing with appropriate and specific direction to the original content. All interviewees and authors co-copyright their material and may disseminate for their independent purposes.