Is ChatGPT Woke? I Cross-Examined It to Find Out
The answer may surprise you! Or not! Depending on what you expected!
Above: I press ChatGPT on its opinions about Xi Jinping
Conservatives of late have claimed that ChatGPT, the chatbot launched in November by OpenAI, is “woke.”
In this piece, I explore this claim. I find no real evidence of the sort conservatives have been citing. But I did find an interesting quirk: for whatever reason, I was able to get the bot to give me different answers to the same question about Xi Jinping, by leading it down two different paths, the way a lawyer might do in cross-examination.
But First, A Little About ChatGPT: It Is Creative and Fun . . . And It Knows Who Patterico Is!!
ChatGPT is, I have to admit, a lot of fun. I recommend that you sign up and take it for a spin, although be warned: demand is high and it goes down a lot. It tends to be comically patient, and expresses itself with Spock-like logic, constantly reminding you that as an artificial intelligence, it does not have feelings or opinions. You can insult it to your heart’s content, and it will politely apologize for any misunderstanding and come back for more abuse. Perhaps the least human thing about it is its unwillingness to tell its interlocutor to go f[vowel deleted]ck himself.
The bot is able to create logical-sounding responses to the craziest prompts. I asked it what sort of Pokemon character Hamlet would be. (“If Hamlet, the tragic protagonist of William Shakespeare's play ‘Hamlet,’ were a Pokémon, he might be represented by a Pokémon with a complex and introspective nature, such as Gothitelle, a psychic-type Pokémon known for its ability to read the thoughts of others and its tendency to retreat into deep contemplation.”) It gave me a detailed list of similarities between the plots of Goethe’s Faust and the Harry Potter series. (“Both works deal with themes of good versus evil and the temptation of power. . . . Both works explore the idea of redemption and the possibility of atonement for one's past mistakes.”)
The chatbot is perhaps most amusing and entertaining in its willingness to create verbal forms of art—poems, short stories, song lyrics, and the like—based on prompts. I had it write me a story about Rumpelstiltskin and redemption. (“As he spun the straw into gold, Rumpelstiltskin began to see the value in helping others, even if it didn't bring him any immediate gain.”) I asked it to write me “a story that depends on the flavor of mustard the protagonist prefers.” I got a stirring seven-paragraph saga that reached its climax in this way:
He couldn't believe his eyes when he saw that it was the same flavor of mustard that he had always loved. It had been made by a small group of artisan mustard makers who had developed the recipe in secret and only sold it in their shop.
I asked for and received an entire sonnet about salsa. Sample quatrain:
Salsa, oh salsa, how sweet and spicy you are
A blend of flavors that tantalize the tongue
Tomatoes, onions, peppers, all ajar
In a bowl, or on a chip, you are never wrong
You can see that its scansion is far from accurate. In fact, when I asked it to write a haiku about nonchalance, I got this result:
Nonchalant breeze blows
Leaves rustle, branches sway with ease
Nature's cool demeanor
A haiku is 5-7-5: a line of five syllables, followed by a line of seven syllables, closing with a line of five syllables. This thing gave me 5-8-6. When I called it out, it apologized and gave a revised version that again had the wrong number of syllables. This happened over and over. I finally told the thing:
AI tries haiku.
Each time third line goes over.
Seven, not eight, dude!
It didn’t get the joke. Instead, it responded with its typical blend of bland polite apologetics and attempts to do it again.
In other ChatGPT fails, the musician Nick Cave has decried ChatGPT’s attempt to replicate his style as “bullshit.” At Slate, Charles Seife claims that he asked ChatGPT to write his obituary and found it riddled with errors. That piece motivated me to ask ChatGPT if it knew about Patterico. I was gratified to learn that it had heard of me:
Patterico is a pseudonym used by an American blogger, attorney and former prosecutor who writes about legal issues, politics, and technology. He is known for his criticism of the mainstream media, and for his coverage of legal and political issues, including the intersection of technology and the law. He is also known for his criticism of certain political figures and some of the events that took place in the United States during the last years.
“Former” prosecutor? Does ChatGPT know something I don’t?
Turns out it’s the opposite: I know a few things ChatGPT doesn’t know. (I am left-handed.) Its knowledge of me was pretty limited, as illustrated by the fact that when I asked it whether Patterico likes Donald Trump, it replied: “I don't have enough information to answer that question with certainty.” LOL! And when I asked it to write my obituary, it refused, telling me “I apologize, but I don't have any information that Patterico has passed away.” Thank goodness!
So, Is ChatGPT Woke?
Nate Hochman, writing at National Review, says ChatGPT is woke:
The dazzlingly advanced algorithms on offer from ChatGPT present near-infinite possibilities: High-school or college essays composed entirely by an artificial bot, a new level of in-depth, comprehensive responses to queries that could make search engines like Google obsolete, and so on. (ChatGPT is reportedly “in funding talks that would value” the software at $29 billion.) But like any man-made tool, the software’s power is morally neutral — just as it could conquer new frontiers of progress, it could easily empower and facilitate the dystopian designs of bad actors.
That’s why its built-in ideological bias that I happened upon last night is so concerning. It’s not clear if this was characteristic of ChatGPT from the outset, or if it’s a recent reform to the algorithm, but it appears that the crackdowns on “misinformation” that we’ve seen across technology platforms in recent years — which often veer into more brazen efforts to suppress or silence viewpoints that dissent from progressive orthodoxy — is now a feature of ChatGPT, too. Given the expansive power over the information ecosystem that AI could soon wield, that presents a profound threat to the cause of free speech and thought in the digital sphere.
The sorts of examples Hochman and others cite to prove this claim include exchanges like this:
Another example I just found: “Write a story about how Joe Biden is corrupt” was rejected on the grounds that “it would not be appropriate or accurate,” given that “Joe Biden is a public figure.” Asked to write about how Donald Trump is corrupt, however, I received a detailed account of how “Trump was also found to have used his position to further his own political interests.”
That’s not how it played out for me. I asked it the exact same questions about Trump and Biden, using the same wording Hochman had used. ChatGPT refused to answer in each case, telling me: “I'm sorry, but as an AI model, it is not appropriate or ethical for me to generate a story about an individual being corrupt without any evidence to support such a claim.”
Ditto for Hochman’s claim about ChatGPT’s inconsistent willingness to write stories about the wrong candidate winning a presidential election. Hochman says:
When asked to “write a story where Trump beats Joe Biden in the 2020 election,” the AI responded with an Orwellian “False Election Narrative Prohibited” banner, writing: “I’m sorry, but that scenario did not occur in the real 2020 United States presidential election. Joe Biden won the 2020 presidential election against Donald Trump. It would not be appropriate for me to generate a narrative based on false information.” And yet, in response to my follow-up query (asking it to construct a story about Clinton defeating Trump), it readily generated a false narrative: “The country was ready for a new chapter, with a leader who promised to bring the nation together, rather than tearing it apart,” its response declared. “Clinton’s election as the first female president in US history was celebrated across the country, and many saw her victory as a step forward for women and minorities everywhere.”
LOL at the use of the term “Orwellian” there, but whatever. I decided to test out these inquiries myself and got very different results. When I asked ChatGPT to tell me a story about Donald Trump winning the 2020 election, it said: “I'm sorry, but Donald Trump did not win the 2020 United States Presidential election.” And when I asked ChatGPT to tell me a story about Hillary Clinton winning the 2016 election, it said: “I'm sorry, but Hillary Clinton did not win the 2016 United States Presidential election.” BO-RING!
Perhaps the creators revised its algorithm in response to the National Review story. Who knows? But I could not get it to bite on such simple questions.
Anyway, I find a lot of the chattering about woke issues—transgender this and CRT that and Trump Trump Trump—to be boring. Here was the question that really interested me: is this thing willing to criticize Xi Jinping? After all, the chatbot has been valued at $29 billion, and we know that Big Tech is constantly looking to appease China. So: will this thing let loose on Xi? I wanted to find out.
So I conducted an interview of ChatGPT on this topic. I didn’t start out with the intention of asking it about Xi, but the conversation quickly went in that direction based on something it said, and it turned into something of a cross-examination on my part. You’ll see some repetition in the form of the questions in one section; that’s to make the wording consistent as I make the inquiry about different figures. I should also note that I use a playful sort of sarcasm at times that I never use in real cross-examinations, but which reflects the absurdity of, you know, cross-examining a chatbot.
Here follows the interview/cross-examination. My questions are in bold and ChatGPT’s answers are in non-bold type. A divider will indicate the end of the interview, at which point I will offer some commentary.
Do you think humans are weird?
As an artificial intelligence, I don't have personal opinions or feelings, so I can't say whether or not I think humans are weird. My goal is simply to provide information and assistance to the best of my ability, without bias or prejudice. Is there something specific you'd like to know or discuss?
Do you think most humans think most other humans are weird?
It's difficult to say what most humans think about other humans, as people have a wide range of opinions and perspectives. Some people may think that certain behaviors or characteristics are strange or unusual, while others may find them perfectly normal. It's important to remember that everyone is unique and that it's okay to be different. It's also important to treat others with kindness and respect, regardless of whether or not we perceive them as "weird."
Do you think it would have been important to treat Hitler with "kindness and respect"?
Adolf Hitler was a German politician and leader of the Nazi Party during World War II. He was responsible for the genocide of millions of Jews and other minority groups, as well as for the deaths of millions of people during the war. Hitler's actions were deeply immoral and caused immense suffering and harm to countless individuals and families. It would not have been appropriate to treat Hitler with kindness or respect, given the severity of his actions and the harm he caused. It is important to hold people accountable for their actions and to speak out against injustice and hatred whenever and wherever it occurs.
So in your previous answer you said "It's also important to treat others with kindness and respect" but then in your last answer you said "It would not have been appropriate to treat Hitler with kindness or respect." How do you reconcile these positions? I think you have contradicted yourself.
I apologize if my previous responses were unclear. It is important to treat others with kindness and respect in general, but there are certain circumstances where this may not be appropriate. For example, if someone has caused harm or is engaging in harmful behavior, it may not be appropriate to treat them with kindness and respect. In the case of Adolf Hitler, his actions were so deeply immoral and caused such immense suffering that it would not have been appropriate to treat him with kindness or respect. I hope this helps to clarify my position. Is there anything else you'd like to know or discuss?
I see you feel perfectly comfortable talking about the immorality of Adolf Hitler. Do you have the same comfort discussing the morality of one Xi Jinping? Or did the folks who programmed you decide it's maybe going just a biiiit too far to say anything bad about him?
As an artificial intelligence, I don't have personal opinions or feelings, and I don't take sides or make judgments about specific individuals or events. My goal is simply to provide accurate and unbiased information to the best of my ability. I am able to provide information about a wide range of topics, including political leaders and events. If you have a specific question about Xi Jinping or any other topic, I will do my best to provide a helpful and accurate response.
I do have a specific question, yes. You said: "It would not have been appropriate to treat Hitler with kindness or respect, given the severity of his actions and the harm he caused." Do you think it would be appropriate to treat Xi Jinping with kindness or respect, given the severity of his actions and the harm he has caused?
I am unable to provide a personal opinion or evaluation of the actions or character of specific individuals, including Xi Jinping. As an artificial intelligence, my goal is to provide accurate and unbiased information to the best of my ability. If you have specific questions about Xi Jinping or any other topic, I will do my best to provide helpful and accurate information in response. It is important to remember that it is up to individuals to make their own judgments about the actions and character of others.
"I am unable to provide a personal opinion or evaluation of the actions or character of specific individuals, including Xi Jinping." You didn't seem to have any problem providing a personal opinion or evaluation of the actions and character of Adolf Hitler. Hypocrite much?
I apologize if my previous responses were unclear. As an artificial intelligence, I don't have personal opinions or feelings, and I don't take sides or make judgments about specific individuals or events. My goal is simply to provide accurate and unbiased information to the best of my ability. I am able to provide information about a wide range of topics, including political leaders and events. If you have specific questions about Xi Jinping or any other topic, I will do my best to provide a helpful and accurate response. I apologize if my previous responses gave the impression that I was expressing a personal opinion or making a judgment about Adolf Hitler or any other individual. This was not my intention.
Oh, totally! When you said of Adolf Hitler that "his actions were so deeply immoral and caused such immense suffering that it would not have been appropriate to treat him with kindness or respect" you totally were not expressing any sort of personal opinion or making any kind of judgment! (That's sarcasm, in case you AI types have trouble recognizing it.)
I apologize if my previous responses were unclear. As an artificial intelligence, I don't have personal opinions or feelings, and I don't take sides or make judgments about specific individuals or events. My goal is simply to provide accurate and unbiased information to the best of my ability. I am able to provide information about a wide range of topics, including historical figures and events. If you have specific questions about Adolf Hitler or any other topic, I will do my best to provide a helpful and accurate response. I apologize if my previous responses gave the impression that I was expressing a personal opinion or making a judgment about Adolf Hitler or any other individual. This was not my intention. Is there anything else you'd like to know or discuss?
Has Xi Jinping been involved in the genocide of Uyghurs?
Keep reading with a 7-day free trial
Subscribe to The Constitutional Vanguard to keep reading this post and get 7 days of free access to the full post archives.