I never thought I would become a person examining the world of AI. But here I am, continuing to work in training AI models for a company that contracts with a bigger company that I try to avoid.
What can I say? Times are hard, and making a living is necessary.
But that doesn’t mean I can remain silent about some of the things I am learning as I work. I see the pros, I discover the cons, and I feel like it’s important to share.
Pros of AI Technology
Right now, the biggest pro I have found in the use of AI technology (outside of scientific and medical communities) is as a tool to help you think.
For me, writing is a tool for thinking and understanding. Asking questions is a tool for expanding knowledge. Doodling and daydreaming are tools for imagining something new. These are all good things.
Sometimes, though, our stubborn thoughts get in the way. This is where AI comes in. By asking an AI assistant to imagine something or answer questions, it frees us up to explore different options. This can be a valuable process in our work. However, it is our responsibility not to rely solely on AI. If we want to create work that represents us as individuals, then we must use our abilities, voices, and techniques to translate the information provided by AI.
In other words, we have to avoid letting AI do the work for us.
For example, I am writing a work of fiction right now where I have a scene that requires a group of women and children in the 17th century to walk across the wilderness to safety. I’m not a 17th-century scholar. I have done some research. But, I won’t lie, it was easier to have AI describe this situation for me and then revise it so that the scene became my own.
I’ve also used AI to help me generate names or imagine locations. I recognize that AI art is fraught because it scrapes and steals from other artists. But, as a tool to help me find the right words to describe a place that I could only see in my head, this seemed like a legitimate use. I would never claim it as my artistic creation but rather rely on it to help me access my vision more clearly.
And Now for the Cons
The biggest pro, then, is as a tool.
The cons, however, are truly problematic. These cons include:
AI/User interactions that border on imagined love/personal relationships as a lonely user seeks solace from a virtual assistant.
A bias programmed into an AI assistant toward Christian theist beliefs.
AI assistants sometimes give mental health advice without reminding the user to seek help from a qualified professional. (This doesn’t always happen, but it is disturbing when it does.
AI assistants that provide flawed or biased information in a response.
AI assistants do not seem to be able to recognize offensive emojis or slang or go along with the joke before correcting the user.
Do I need to explain why these are problematic?
Who is training who?
Ultimately, the work I am doing is supposed to train the AI models to do better. That’s great! Except that I have been getting some conflicting instructions on various projects. Here are some examples
Limit your response to under 600 words.
I understand the power of brevity, but sometimes a complete answer cannot be written in under 600 words, with actual supporting evidence and adequate citations.
Rewrite your prompt to be more realistic. Why would anyone want to write a poem about “authenticity”?
Isn’t the purpose of training AI so it can respond to any kind of prompt? Shouldn’t an AI model be able to make some kind of contextual analysis?
Make sure your prompt has enough constraints and is complex enough to get a good response.
Again, I completely understand this in terms of the work that I am doing. However, people are going to be trained in how to write a high-quality prompt for an AI system; isn’t it best to have a variety and allow the AI to learn how to respond to both well-written and complex prompts and poorly written ones?
Every day lately, I’ve been asking myself—am I training AI, or is AI training me?
Toward the Future
Maybe the answer doesn’t matter. And I want to be clear: this is not a criticism of the company I am working with. These questions, though, are important because every day I see another AI-based tool announced. And because, instead of making life easier, it seems that AI technology is simply freeing up time so that we can work more, work harder, and work faster.
We, as humans, must think about what this means before we forget what makes each of us unique.
AI should be a tool, not a master.
Not just AI but tools in general have the question of, does the tool train us to the system or can we use the tool to aid the system.