I'm not entirely sure why we're doing a pair of technology-related articles, neither of which are particularly featuring a film or political angle, but after we talked about Twitter yesterday, I thought it would be fun to do another "get John's opinion" article about the other problematic tech app that everyone in the world seems to have a complicated relationship with: ChatGPT.
I have not been shy about my dislike of AI, and by proxy, ChatGPT (see here for a recent rant). But, like virtually everyone right now, there is a societal (and occupational) pressure to understand how ChatGPT works, and what its benefits are. So I, after occasionally tinkering with it, have started to (on occasion) use it. And I wanted to share a few observations I had about it that feel both helpful, and really terrifying.
Editor's Note: This article is about ChatGPT. For the record, no article on this blog has ever, or will ever, be written by ChatGPT or an AI-writing app. If I ever get to the point where I want to do that, I'll give up writing first. Wanted to state that straight away!
Here's the dirty secret-there are aspects of ChatGPT that I will own are genuinely useful. I think the biggest one that I've run into is in helping me create a new workout plan, which has been the biggest thing I've used it for. I have four separate goals I'm working on right now in terms of exercise: losing weight, running a sprint triathlon, climbing Pike's Peak, and (vainly) trying to protect gains in some of the "glamour" muscles because I'm a single gay man and that's something you keep as a calling card for dates. I have a finite amount of time each week, and can only devote so much of it to exercise, and so I had struggled for weeks to figure out exactly what to do in terms of my workout. So I asked ChatGPT to formulate a workout for these four combined, with at least one rest day, and one that worked largely around my work schedule.
And it did! The plan it came back with was genuinely impressive. It had room for my triathlon, was able to incorporate in extra leg days to start building on Pike's Peak, kept me focused on extra cardio for losing weight, and kept at least a couple of days for glamour muscles. It also amends really easily-you can tell it to, say, give you at least one day for yoga or have it track your weight-lifting progress (or share feedback like "my gym doesn't have that equipment-can you give me a comparable exercise that doesn't require that equipment?") and it does. There are a lot of ethical issues with this, and from a privacy/safety concern I'm not putting in extra information (i.e. it does not know my age or weight, which on my end I need to be able to keep track of since that greatly impacts a workout routine's ability to help), but what it is doing is, I must admit, a quality workout routine. In general, the thing I have most used it for is a brainstorming partner-not someone doing the work (though it is doing a lot of it), but someone that can bounce ideas off of and put it into something legible.
There are a lot of bad things about ChatGPT. There are the obvious (the ethical questions about its impact on the environment, a primary reason why I have limited it to only the exercise project and to learning how to use it given that its expectation in my industry to have a cursory knowledge of it) as well as the subtle (I think it's dangerous to rely upon AI for things like writing and reading comprehension skills, because those need to be continually practiced to stay sharp), but the most noticeable thing about it is the accuracy.
ChatGPT says in its description that it can be wrong (it's right there when you're typing: "ChatGPT can make mistakes"), but I think there's a really scary reality that most people accept whatever ChatGPT shares with them as fact, when it's decidedly not true. I had used the tool as a test to a question I don't know the answer to: who was the fourth woman to join the DGA? It's generally accepted that the first three women in the DGA were Dorothy Arzner, Ida Lupino, & Elaine May. As far as I'm aware, these are the first three women to direct feature films for major studios, and so they'd be the first three members, but I could not find any evidence of who was fourth on Google or through searching library databases, so I decided to ask ChatGPT.
ChatGPT initially said May was fourth, which I allowed could be true, but it couldn't give me anyone that might have been third without prodding. It eventually provided the name Shirley Clarke as fact for the third member, but given that Clarke was an independent filmmaker in the 1970's who never made a studio-driven picture (i.e. she didn't need to join the DGA in order to make the movies she made), I pushed back, asking for evidence that Clarke was the third...which it couldn't provide. It just had made an intelligent guess. It eventually came up with a few names that plausibly could've been the fourth woman: Lynne Littman, Nell Cox, & Dolores Ferraro are all names ChatGPT provided that make sense as the fourth woman, as they all made Hollywood-studio driven film & television, and if it is one of these women, it's probably Cox who directed an episode of The Waltons a couple of years before the others had such projects, and as a CBS-broadcast show she would've been required to join the union. But ultimately it could not find the answer-ChatGPT was not able to prove that there was a woman to join the DGA before Elaine May other than Arzner or Lupino, and could not definitively prove who the fourth member was.
And that would be okay...had it just said that to begin with-it's possible this is an answer the internet doesn't have the answer to, and the only way to find out would be to write the guild directly. I certainly couldn't find it, and I've tried to find this answer for a while. But ChatGPT initially, definitively, stated that May was fourth and Clarke was third, and it took me pushing back to make it think otherwise. That's a problem because you have to have a pretty extensive knowledge of a subject to be able to get at that level of detail-most people would've taken ChatGPT as fact in this situation, and thus provided the wrong information. And as more people publish research they have from ChatGPT as if it's fact in articles that will be data-scraped...it will be harder & harder to correct.
There's a lot of ugliness with ChatGPT. The biggest one is obviously occupational. Going back to my exercise example, there's a clear answer who could've helped me on this previously, and in fact had helped me in the past: a gym trainer. I have seen gym trainers and nutritionists through the years, and they are more than capable of doing this. Admittedly, I wouldn't have hired a trainer or nutritionist in this regard because I cannot presently afford one, and so in some ways this is not replacing a job more so than it is saving me time...but let's be clear, that's not how everyone is going to use this, and it is scary that we are lifting the human element out of a job like this, particularly given that it requires humans schooling and knowledge to do set jobs (and the humans, even with as well as ChatGPT did, are better at it because they see you as a person and not just something statistical to add together...and I suspect those same humans, through social media posts & videos, are probably the ones crafting what ChatGPT gave me to a large degree).
But for me, the biggest concern with ChatGPT, and the eeriest thing about it is that it was so freaking nice. People have mocked in TikTok's and other social media how ChatGPT is encouraging to the point of laughable-it will literally say any question you ask or any idea you have is a good idea (South Park had fun with this in a recent episode). But I will be honest-it felt kind of nice to have someone (or, more correctly, some thing) care about this project that most people in my life don't care about. Most people I know do not care who the fourth woman in the DGA is, and wouldn't have wanted to talk about it. Most people in my life would not have the time to help me pick a workout routine that specific. That ChatGPT does this, and is so gracious & encouraging, is a weird sort of Twilight Zone-thing. You can easily see people using this not just to replace people, but to replace human relationships. ChatGPT (for a price) always has time for your thoughts & expressions, and is nicer than an increasingly cruel world. I put a photo of Joaquin Phoenix in Her next to this section, because it's bizarre how closely it resembles his experience-AI is not mechanical, but it's warm and inviting like Scarlett Johansson's Samantha. That it's still a machine, and something that is designed to want to pull you in & use it as much as possible, makes it easy to see people forming parasocial relationships (that feel suspiciously like real relationships) with it, a terrifying thought particularly for a tool that (if I'm being honest) does have genuine usefulness.
No comments:
Post a Comment