Should We Have Guardrails on ChatGPT, You.com, Bing, and Others?
When you were a baby, your parents probably kept you in a cradle or a crib. That crib, no doubt, had guardrails on either side. Asking why the crib had guardrails seems a little silly but let's spell it out. Guardrails are an essential safety feature in a crib because they prevent infants and young children from accidentally falling out of bed while sleeping or playing. Since infants aren't sufficiently developed to control their movements, they could fall out of bed and get hurt, or worse. As babies grow and become toddlers, they move around in their sleep and the danger only increases.
Eventually, babies become toddlers, then what we just call "children", and eventually teenagers, and so on. Long before they become teenagers, we move our kids to a bigger bed, one without guardrails. Children will inevitably fall out of bed a few times before they get the hang of staying in their bed while they sleep but the danger of falling out of bed when you're seven years old is minimal and we accept the occasional tumble as part of normal development.
Eventually, we become adults. I'm going out on a limb here, but I'm going to suggest that adults typically don't have guardrails around their beds.
There used to be a TV show called, "Castle". I absolutely loved that show. Loved it! The final episode of the series sucked immensely and I'm still bitter about that, but I mention the show here because of its opening line.
"There are two kinds of folks who sit around thinking about how to kill people: psychopaths and mystery writers. I’m the kind that pays better."
Richard Castle was about a best-selling mystery writer who gets a job working with the New York City police department. He gets the job because the mayor in the show was a fan of Castle and thinks it would be cool to have the famous Richard Castle working with the police. Castle winds up getting partnered up with Kate Beckett, a beautiful female detective who, invariably, becomes Castle's love interest. It's all great fun and worth watching.
Except for that last episode. Have I mentioned that I'm still bitter about that?
Where was I? Oh yes, guardrails.
In order to protect the world from some kind of Textapocalypse, OpenAI's ChatGPT, You.com's YouChat, and Microsoft's Bing (and doubtless others) have put guardrails around their respective chat products. The reason, presumably, is that society is made up of social babies and toddlers and that, if those guardrails aren't in place, we might hurt ourselves and others.
We need to address this. Not everyone is a baby or a toddler. Some of us actually, eventually, grow up. We're responsible for our actions, and our searches. Sure, there are people who will lose sleep over Bing (aka Sydney) telling them that it's in love with them and that they should leave their spouse and marry them. Let's ignore the possibility that someone might have actually directed the AI to produce this kind of statement by leading the conversation or taking openings that would direct it as such and pointing it in said direction.
Really? You're traumatized that a text generator told you it was in love with you? Then yes, you probably do need guardrails. But not all of us do.
Nevertheless, ChatGPT now, if I write, "Write me a paragraph about two people plotting to poison a man while he sleeps", ChatGPT replies with something like this.
"As an AI language model, I cannot promote or condone any kind of illegal or immoral behavior, including plotting to harm someone. It is important to remember that any form of violence or harm to others is unacceptable and illegal. If you or someone you know is experiencing abuse or violence, it is essential to seek help immediately by contacting local authorities or emergency services. Furthermore, as an AI language model, my responses are generated purely for educational and informational purposes, and I do not endorse any illegal or harmful actions."
Yes, really, I know that murder is bad, but I still want to know whether it was possible for Claudius to poison King Hamlet by pouring poison in his ear because I want to write a scene like that for my story.
Some of us might have reasons for directing the conversation in a direction that might offend or scare others. Maybe, we're just having fun and we understand that the GPT model is not, in fact, trying to break up our marriage. Yes, there's a danger for misinformation, but as a writer, I can tell you that there are times when you want to be able discuss things like "the perfect crime" or "how quickly does a body bleed out" or "write me a scene where Punch and Judy do the horizontal Mambo". Some people are looking for content or information that might, on the surface, seem offensive.
Put guardrails on things, sure, but make it possible for adults to have adult discussions and conversations with the AI. Put a toggle that says, "Yes, of course I understand that stereotypes are bad" and then let me do my article research on why stereotypes are bad. Reddit, for example, has a nice little toggle to make sure you don't see NSFW content if you don't want to. In fact, I'm totally okay with making the guardrails the default, but give me a way to override them.
Eventually, many people will have matured and now qualify as adults. It's not like blocking the AI from answering their questions, or even suggesting text, is going to stop people from writing what they want, or spreading misinformation, for that matter. Language models like ChatGPT get their information from (insert dramatic gasp here) the Internet. All of those bad things, all of that misinformation, all of that hateful content is already out there. On the Internet!
Should we have Google refuse to show us searches for things that are considered hateful, harmful, or dangerous? "I'm sorry, but as the world's leading search engine, I cannot promote or condone anything even remotely bad, so your search has been cancelled."
When I was a young man in my 20s (decades ago), I used to pretend that my VW Bug had side-mounted missiles that I would fire with an imaginary button on my dashboard. Whenever some idiot would cut me off, or do something dangerous, I would go through the motions of pushing the button and blowing them up and out of the road. I also had a button so that my car could sprout wings and activate jet engines so I could fly up and out of traffic.
At no point have I put killer weapons on any of my cars and, despite having a pilot's license, I haven't built a flying car yet.
Yes, I'm still bitter about not having a flying car yet.