Everybody is already discussing ChatGPT, and since I’m a software developer and writer, I hope to have some useful insights to add.
in case you don’t know, ChatGPT is a chatbot. In other words, a software application that you, as the user, can chat with. You can ask it questions, and discuss things with it. That kind of software have been around for a while — I still remember Dr. Sbaitso from the early nineties. Of course, chat bots have changed since then. Dr. Sbaitso was horrible at understanding you. Chatbots have been getting steadily better, but they were still pretty limited.
ChatGPT is a scary good chatbot. It is exceptional at recognizing what users are asking for. It also has access to a large dataset of internet-based information. Wikipedia is at its fingertips, as is information from webcrawlers. It was also trained on GitHub, which is a hub of open source developer projects. This means it can actually help people with developing software. And, yeah, it’s pretty impressive.
So, I created an account and tried it out. You can ask ChatGPT anything, from “how do I make a sandwich” to “write an essay on Beowulf”. Being interested in how large its dataset was, I asked it about my old webcomic, Frank, Hank, and Charles. And, after some clarification, it came up with the following:
Wow. That’s pretty amazing! Er… wait… I started that webcomic in 2004 not 2005. And… Hank is not “a wannabe ladies’ man”. Over the entire 500+ page run, it has never even made clear if Hank is straight, gay, pansexual, or asexual. And ‘roommates’, no. Charles wants anything but Frank and Hank in his house. Finally, ‘bold colors’ might not be quite accurate for a grayscale comic.
So, okay. ChatGPT is not always correct. I proceeded to ask about the author. Me, in other words.
That was disappointing. Way to boost my ego, ChatGPT. It’s also eye-opening, because it just told me about my webcomic.
The limits of AI
When I look at the previous paragraph, I see a couple of things. First, ChatGPT is very good at reading what I’m asking. It can parse questions, and it retains information between questions, allowing you to refine your interaction. That’s amazing in itself.
On the flip side, ChatGPT, like all AI, is a form of advanced searching and pattern matching. This means it has no actual understanding of what it’s doing. It’s not capable of ‘common sense’, or truly understanding context. This is why an AI that generates pictures might generate people with six fingers, or animals with too few eyes or too many legs. It’s why ChatGPT can tell me about my webcomic, and then claim it has no idea who I am.
That doesn’t mean this technology is useless, or inferior. It’s really cool, but it’s not intelligent, despite it being named artificial intelligence.
So, I watched a video about how ChatGPT might change anything. Or not. The point of that video is, is that we don’t know yet what might happen. Will this AI technology disrupt the world, like the internet did, or iPhones, or Penicillin? Or is it just a new tool in our existing arsenal?
The video uses Napster as an example of how the internet eventually wiped out most of the music-on-physical-medium industry, and replaced it with streaming services like Spotify. So the question is: will tools like ChatGPT do the same? Will it replace entire industries?
The first industry that comes to mind, is the artistic industry. The writing community is already buzzing about that. The scifi magazine Clarkesworld closed its doors for submissions because spammers were flooding it with AI stories.
However, don’t read too much into that. AI will not replace authors, I’d say. Not yet, anyway. I asked ChatGPT to write me a story, and it wasn’t very good. It was grammatically correct, but it was not a good story. The problem, as Chuck Wendig eloquently explained on his blog is that in writing, the idea is just the seed, and the execution is where the work is. So you give an idea to an AI, and it can transform that to a story, but it can only pattern-match it into place. It isn’t creative, and it cannot give it a ‘soul’ for lack of a better word.
Also, an AI cannot apply common sense, meaning it might create a detective story where it adds a reveal scene for a killer that’s a character added only in the last chapter. An AI that has that context is beyond what we can do, and I’d argue it would have to be a self-aware life form.
When ‘okay’ is good enough
AI can only disrupt industries where the work being done doesn’t require a lot of common sense. Where it doesn’t matter if it makes some mistakes. Writing is not a place for that. The same goes for medical treatment, or driving a car. And I’ve already stated that I think AI is not capable of content moderation for this very reason. But it might hurt other industries.
ChatGPT is not going to replace authors. However, fiction writing is the world championship of the writing art. The Mount Everest of the writing profession. ChatGPT can replace professional writers in places where ‘okay’ is good enough.
When I open my fridge and take out almost any food item, there’s a label on there with a description of the product. That description gushes about the natural water in my soda, or how the tomatoes were grown in a field near a ley line and harvested during the full moon. That kind of writing… well, I think if that’s your job, you should be worried.
The same is happening for artists. The first AI generated book covers are already making their way to the market. That genie is not going back into the bottle. Artists are going to have a harder time, because AI generated art is only ‘okay’ but that is more often than not ‘good enough’. I’d like my works published with actual commissioned art. However, if I have to choose between publishing my work at a loss with commissioned art, or with a profit with AI art… Well, I’m glad I’m not in that situation. And believe me, publishers are making that calculation. And not just for book covers. AI advertisements are no doubt coming.
There are places where okay is fine. For a growing number of companies, Chatbots are the preferred way of interacting with customer support. ChatGPT will move into that space right quick. If you can increase the people you help through chat by only a few percent, the savings are enormous. And ChatGPT offers more than a few percent.
AI will start appearing in tools that can help us, too. With human guidance, it will be a very useful tool to make humans more productive at their jobs. But it will take away mundane work. My biggest worry is that our society says we have to work to live, and removing work will be like playing musical chairs with people’s livelihood. That will hurt, but that’s not because of AI, that’s because of how we set up our society.
Down the rabbit hole
Where ‘okay’ is also good enough, is in the education space. Kids are getting wise to ChatGPT, and AI essays are on the rise. This causes a stir already, but I think you can’t put that genie back in the bottle either. AI essays are here to stay.
And that doesn’t have to be a big problem. But I also expect ChatGPT-like AI will make its way into Whatsapp and the like. A few years from now, you might just tell your smartphone ‘tell my mom I’ll be late for dinner, nicely’ and a perfectly nice app in your style will pop out.
Again, not worrisome, until that happens all over the place. The rise of messaging and smartphones has already damaged writing and math skills in students, so far that 15% of Dutch youths is functionally illiterate. That means they cannot even understand letters from the government, or a doctor recipe.
This trend is worrying, and ChatGPT is not helping. If we continue like this, that percentage might go up higher and higher. And ChatGPT and other AI might actually be able to help those people as well. However, errors will start to creep in, because AI has no common sense. The gap between the haves and the have-nots will increase, and worst case, we come to Eloi and Brave New World territory.
Disruption of War
And that brings us to the really scary part. You know where ‘okay is good enough’ is very dangerous? War. Russia is okay with hurting a lot of civilians to conquer Ukraine. They don’t mind AI drones that sometimes miss their targets. That’s a war crime, of course, but Putin doesn’t seem to care.
We’re entering an era of mass AI weaponry. And that is scary. Because AI has no common sense. Go Play Horizon Zero Dawn if you want to know where that can lead. It’s a scary time when having the same height, weight, and facial structure as Zelensky might get you gunned down by a drone. Where automated drones might blow up half your neighborhood because they can’t be arsed to see the difference between your house and a weapon factory.
Beyond that, we see that ChatGPT is capable of developing software. That’s scary too. Not because it will bootstrap itself to intelligence, but because that too is a weapon. I’ve warned about the dangers of our automation fetish, but AI is pushing the dangers to new heights. Hackers are getting faster and faster at finding exploits in software, and better at exploiting the human part of IT security at companies. ChatGPT can help even more. AI could make it even easier to find and exploit bugs in software. ChatGPT could help create phishing e-mails that could gain access to systems.
We’ve seen a steady increase in hackers hurting the real world for fun and profit (and war). That’s going to go up another notch or two. The internet might well become untenable, even as it is unmissable. I have no idea where that will end, but ChatGPT is the canary in the coalmine for some funky shit.
In the end, ChatGPT and its like will make our tools even smarter. I doubt it will disrupt a lot of industries. but I could well be wrong.
I am far more worried about what it might do in the hacking and military spaces. Not so much disruption, but escalation. The potential for serious damage is there. That scares me.
That said, the genie is out of the bottle, and we’d all better prepare for the roller coaster ride ahead.