I think I was kind of long overdue a case of covid given the amount of mid-pandemic international travel I did for work. It wasn't too bad, but did knock me about for a few weeks. I spent most of a couple of weeks either in bed or wishing I hadn’t got out of bed. Not fun.
If you read any substack newsletters, or have been anywhere near the internet in the past few months, you’ll undoubtedly have come across people talking about ChatGTP, the terrifying/useful (terrifyingly useful?) chat bot from Open AI.
As you might have gathered by now, I’m really interested in technology and how it intersects with people and their ability to make new things. Tinkering about with software has been a hobby of mine since the 90s, when I’d buy computer magazines to get at the software demo disks attached to their fronts. What a time to be alive.
That’s how I ended up doing a degree in Multimedia Design in the early 2000s, and that’s where I fell head over heels in love with Macromedia Director and the things it could do with 3d on the web. I ended up interning for an architectural visualisation company doing web3d for them using Director. Good times. (actually I was dirt poor while working there and lived in a hostel in central London for a couple of weeks eating basically no more than an apple a day to make ends meet)
Anyway Smash Cut! back to me laying in bed, a sweaty mess, missing out on all the festive festivities. I’d read a little about ChatGTP and gave it a whirl. I did the usual stuff first - get it to write diss tracks about dear friends of mine in the style of Tom Waits - the usual stuff. It’s impressive, for sure, but it felt a lot like I was using a fighter jet to trundle slowly along the road to fetch groceries. Not exactly maxing out the potential.
I’d heard that it could do a lot more - write working code in a variety of languages for example. I’ve got a beginners-level of ability with the Unreal and Unity game engines. I can navigate about and make things happen, but I’m in no way a programmer. Back in my Macromedia days, using Director and Flash, I got pretty good at coding and putting together interactive ‘things’ but in the intervening years, my ability to code atrophied as I used it less and less, and updates pulled the rug out from under my knowledge. I still had some of the structural thinking, but barely any of the syntax.
I decided, sitting in bed feeling sorry for myself, that I’d try and use ChatGTP to write a simple game in Unity. I had an idea (copyright, trademark etc) for a game in which you drive a car along a course towards a goal (a hole in the ground, a stack of boxes, fruit vendor’s cart etc) and then bail out, leaving the car to careen along by itself. The aim being to get the car to smash into the goal from the furthest distance from which you bailed out. Kind of a cross between Super Monkey Ball and Burnout. I called it ‘Car Golf’.
Asking ChatGTP to help me make this was weird in the extreme. It’s super competent, but at the same time, utterly inconsistent.
Asking “write a c# script to control a car in unity” would give you a really buggy script that would need lots of work to get it functional. Giving it complex instructions “write a c# script to control a 3d car in unity that the player needs to be able to bail the driver character out of when they hit a button, but the car needs to keep going, but also track how far it travelled since the driver bailed out and also the game is over when they crash into a particular thing” returned real hit-or-miss results.
“Is this the best way to do this?” would give me a “yes of course” response, but if I asked '“how can I improve this?” it’d take a big dump on what it just produced and point out many, many flaws. I started thinking of ChatGTP not as a single entity that was helping me, but a room full of eager ChatGTPs, all listening to me, but not listening to each other. I was never sure if I was going to get an answer from someone who had been paying attention the entire time, or had just done their homework on the bus.
After a little while playing about with it, memories of working in Director drifted back up and Unity snapped into a different level of focus and I did start to feel like I was more in control of, and had a better understanding of the project than ChatGTP had. It was up to me to structure the project and make decisions. It seems that while ChatGTP was good at doing some heavy-lifting in specific ways; getting chunks of code written, getting various balls rolling for example, it was really bad at fitting it all together to form a bigger picture. It’s a really smart thing, but it’s not clever.
I got the game to a level where it was playable, although very, very basic, and it was kinda fun. Not riotous ‘invite all your friends around for a whole evening of entertainment’ levels of fun, but certainly showing some promise;
The eagle-eyed amongst you will notice from the images that this code threw up all kinds of errors, over and over. Trying to get the bot to fix the bugs was pretty much impossible. I had to roll my sleeves up and get in there myself to get them squashed, at which point I realised that I wasn’t really using ChatGTP much. It had started to become less useful, the more specific I needed it to become.
I played around with a bit of level design, trying out courses that would be fun and challenging to line up the car, avoid/use the various obstacles then satisfyingly watch the car tumble through, or miss and have to reset. Then something terrible happened.
My 12-day headache subsided and the thought of walking up or down stairs didn’t fill me with despair. I started feeling better.
With no good reason to stay in bed any longer, and a big old list of things to do stacking up for my actual computer games job with Superbrothers, I put my weird ‘Car Golf’ game on a shelf and got back to work.
I enjoyed messing about making a quick game. I think if I was doing this solo, it would have taken me multiple weeks to get to where I ended up. But with a weird ai copilot, it only took a couple of days or so, and those were sick days too.
My takeaway from this is that I don’t think it’s really possible (at the moment) to get this chat bot to make a simple game like this without fairly significant human-side effort as well. I couldn’t give my kids a unity project and access to ChatGTP for example and expect them to pop out a game a couple of days later. There’s still too many knowledge gaps that need filling with experience. I leaned heavily on 20 year old memories of working in Actionscript 3 and Lingo, making stuff in Lightwave and 3DS Max and a healthy portion Everything Else I Have Ever Learned Ever.
I think what it does well is provide a sounding board to bounce your own thoughts off of. I don’t trust anything it writes, but I found it gave me a way of seeing my own thoughts from a different angle, and I usually find that really useful.
Although ChatGTP creates the things you ask it to, I don’t think of it as a creative tool. I think of it as a way to rapidly get some of the tedious stuff out of the way so I can get into the exciting creative bit myself.
Who knows how this is going to develop. My hunch is ‘rapidly’. I can foresee versions that integrate with something like Siri/Alexa/Cortana, which feels like Star Trek territory to me.
The thing is, now something like this is possible, it kind of becomes inevitable. It’s probably likely that as natural speech interfaces improve and integrate with other platforms, systems and software, sitting down at a desk and describing a game like the one I described above and have a functional game pop out might well be possible eventually.
To me, if it does develop that way, I’ll feel about it the same way I do about Midjourney/Dall-E, the generative AI image robots; Pretty cold. Yes, they can make impressive images, but I find creative work far more interesting and compelling when I know a human has made decisions, mistakes and scratched a creative itch. At galleries, I love to read the little cards next to a picture and get some insight into the work’s creation, and I feel that in a way, AI work doesn’t have those cards. The work sits alone, without context and without the warmth of a direct human touch. About as interesting as the pictures that come with picture frames; disposable.
Maybe all this now curses us to live in interesting times. At least it’ll be interesting.
I might come back to my Car Golf game at some point, I don’t know. As with most things I do, it gives me more ideas for other things I’d like to do. Knowing that there’s a way to get these kinds of balls rolling a little quicker increases the chances I’ll actually do something about it, and that’s what I think is the positive outcome of this little experiment.
Found this really interesting. (I also got started with Director MX in uni!)
Well done on doing the deep dive with it all. I hate the AI stuff honestly because it's being 'touted wrong' as a one size fits/does it all when it feels like its purpose will/should be doing some labour intensive but essentially simple tasks reliably. The up-rezzing of old game backgrounds seemed like a great use of this sort of stuff until the videos of "What if we could replace entire art communities" started cropping up.
I remember trying to code a Gameboy Advance game for a Uni project using what was a simple editor, but that drove me near insane every time it kept crashing so I've forever dipped on touching more. I could see ChatGPT perhaps being more useful as like a Clippy overlay (that scamp), pointing out where you might be going wrong causing errors rather than it doing all of it and having to go in and fix a mountain of piled up disasters. Who knows, maybe it will just punt out AAA videogames in a decade, but yeah, don't like that.
I'm way more interested in VR stuff that opens up different mediums for the human people to tinker with than what feels like the last three years of non-stop "what if art was made by awful tin-openers" news.