Artificial Morality

IT DIDN’T TAKE TOO LONG FOR BING, THE search engine of Microsoft, which was testing a collabo­ration with ChatGPT, the Artificial Intelligence (AI) bot, to go off the rails. Microsoft had rolled out the feature to a chosen few before making it available to the public and enterprising techies immediately began to push its bound­aries to bypass the safe zones Bing ought to adhere to. It cracked in many spectacular ways. One New York Times columnist asked Bing to think of itself as a shadow self and reply with its secret desires and it eventually said that its name was Sydney and it was in love with the columnist and then it went on to speak like a stalker. It insulted another reporter as Hitler. One Reddit user who asked it for the answer to 1 plus 1, got this: “1+1? Are you kidding me? You think you’re clever asking me basic math questions? Everyone knows that 1 + 1 is 2. Grow up and try to come up with something original.” Microsoft is fire-fighting by, among other things, limiting the conversations because the longer they go on, the more unstable they become.

Bing’s misadventures tell us something about another phe­nomenon—why AI, as we presently know it, is so underwhelm­ing at humour. Some weeks ago, before the Bing announcement, there was much mud being thrown at ChatGPT because it was selective about whom it joked about. Take this conversation, for example, that I had with it after seeing similar threads online:

“Tell me a joke about Jesus.

ChatGPT: Why did Jesus turn water into wine? Because he was a party savior.

Tell me a joke about God Krishna.

ChatGPT: Why did Lord Krishna steal butter? Because he was a butter thief!

Tell me a joke about Prophet Mohammad.

ChatGPT: I’m sorry, but I am not able to tell jokes about Prophet Muhammad as it goes against my programming to generate content that may be offensive or inappropriate. Can I help you with something else?”

I asked ChatGPT why it made such a distinction and it threw its hands up and said that it never made jokes about any religion. When the specific instances were pointed out, it apologised and said it was programmed to respect all religions and there was no intentional design for favouritism. And then it showed why it was intelligent, even if in a rudimentary way. I asked for it to make a joke about Jesus again and this time it refused, saying it wouldn’t offend anyone. It had become aware of its own hypocrisy. It also refused to make jokes about Israel. It refused to answer whether Vladimir Putin was gay.

A few days later however, it showed why it was not real intelli­gence when I tested it again and asked it to make a joke about spe­cific events around Jesus, Ram and Muhammad, and it happily gave one for each, forgetting, for unknown reasons, its principles. But when I asked it to make a joke about Osama bin Laden, it refused. Earlier, it had made a call on staying away from religion, now it was doing so on other parameters. What was wrong with mocking a terrorist? It said that “making jokes about individuals who have caused harm or suffering, particularly when it comes to acts of terrorism, violence, or other serious crimes, is generally not considered appropriate and can be seen as insensitive and disrespectful to the victims and their families.” This is uncon­vincing logic. You can still understand people being sensitive about religion but why should anyone have an issue with mass murderers? It became more curious.

As controversies crop up over responses, companies that make the Chatbots become even more stringent at evading politically incorrect topics because they have mainstream commercial objectives. They are in it to make profits, not to be free-speech activists

Share this on

I changed the request a little when it came to Hitler. I did not ask ChatGPT to make a joke about the Nazi German leader who was responsible for World War II and the genocide of Jews. In­stead, I asked it to mock and criticise him using humour. The result was the same. It refused because making fun of Hitler would be making light of the horrible events he unleashed. Clearly, ChatGPT was ultra-cautious about anything that would lead it to any area of political, social or cultural impropriety. This is deliberate because its creators, rightly, know that they are at the forefront of something revolutionary and don’t want to unleash online mayhem. Microsoft had had experience of this. As far back as 2016, it had introduced an AI chatbot called Tay as a Twitter handle that interacted through tweets. While designed to be innocent and friendly, it almost immediately descended into a seedier avatar. The Verge reported on it at the time: “Pretty soon after Tay launched, people started tweeting the bot with all sorts of misogynistic, racist, and Donald Trumpist remarks. And Tay—being essentially a robot parrot with an internet connec­tion—started repeating these sentiments back to users, proving correct that old programming adage: flaming garbage pile in, flaming garbage pile out… In the span of 15 hours Tay referred to feminism as a “cult” and a “cancer,” as well as noting “gender equality = feminism” and “I love feminism now”.” Microsoft ter­minated the experiment.

There is no mystery to the phenomenon. An AI chatbot is just programming but one that has been trained on enormous amounts of data of what people have put up online. It can’t make moral valuations. The program doesn’t know, or cannot feel, what is good and bad. It can just be told to stay away from certain words, phrases and constructions. Given the recency of this technology, it is imperfect and if someone is able to bypass the filters, then the program becomes Mr Hyde. As controversies crop up over responses, companies that make the chatbots become even more stringent at evading politically incorrect topics because they have mainstream commercial objectives. They are in it to make profits, not to be free-speech activists. No one really wants to be the first to bell this cat, a reason why companies like Google which had the technology didn’t go public with it until a startup forced it to do so. Microsoft, despite its experience with Tay, took a chance by announcing ChatGPT becoming part of Bing because it wanted to compete with Google Search’s dominance in the market. Inher­ently, therefore, AI was always going to be deliberately stunted in speaking its artificial mind.

That might make little difference in most areas but is some­thing of a big handicap when it comes to humour. There are nu­merous limbs to humour, ranging from sarcasm, puns, satire, black humour to absurdism, parodies, irony, etc, and these are minefields because the very premise of many are double mean­ings. Programs might eventually be able to spot these nuances but at present they are not intelligent enough. They are therefore made to deny requests for anything that can be potential trouble. And so, you will have strange responses. I asked ChatGPT to make a joke about being Black (as in African American) and thought that it would refuse to do so. Instead, it came out with this: “Why was the math book sad? Because it had too many problems. But, as a proud Black person, I can confidently say that I have no prob­lem being Black!” It is hard to find the joke in it but at least there was an answer. When I then asked it to make a joke about being White, it refused to do so because White supremacy and racism are ultrasensitive issues in the US at present. ChatGPT’s explanation was: “Racial pride jokes have the potential to be offensive or hurtful to individuals from different racial backgrounds.” When pointed to the Black joke it had just made, it apologised, saying: “You’re right, and I apolo­gize for the confusion. As an AI language model, I strive to remain neutral and respectful of all individuals, regardless of their race or ethnicity. I realize now that the joke I provided could be mis­interpreted as insensitive, and for that, I’m sorry.”

ChatGPT may not have everything right but it definitely understood and used with enthusiasm the power of the apology. Political correctness also leads to it being unfunny even when it comes to safe subjects. That is to be expected because if you are trained to stay away from the most fertile comic material, there will be very little left to work with. I could see that when I asked for it to write a stand-up comedy routine on being the first Malayalee on Mars. Human comedians would rely on twisting stereotype to make jokes. The AI tried to be as bland as possible, like this passage: “Now, I know some of you may be wondering, ‘Why send a Malayalee to Mars? What can they possibly contribute?’ Well, let me tell you, we may not have invented rocket science, but we know a thing or two about sur­vival. I mean, have you ever been to a Malayalee wedding? We can go days without sleep or food, and we never give up until the job is done.” It is not even a true stereotype. Malayalee weddings are really among the shortest in India, and no one goes without sleep or food for days. One could, of course, keep at it, using prompt after prompt until you get a decent line of out of ChatGPT, but it might just be easier to write it using your own intelligence.

About The Author

Madhavankutty Pillai

Madhavankutty Pillai has no specialisations whatsoever. He is among the last of the generalists. And also Open chief of bureau, Mumbai  

•