When will AI pass the Robin Williams test?
Generative AI allows the creation of all manner of derivative works, including horrific family chocolate factory-themed days out. What about true inspiration?
Robin Williams, taken from us too soon, died in 2014. He is considered one of the top comedians of all time. Wikipedia has a dedicated article just listing his awards and nominations: It’s a very long list.
I asked ChatGPT 4 to tell me a joke about kittens in the style of Robin Williams (which it told me was an ‘energetic and whimsical’ style):
"Why did the kitten join the Red Cross? Because it wanted to be a first-aid purrr-amedic! You see, in the world of kittens, CPR doesn't stand for 'Cardiopulmonary Resuscitation.' Oh no, it's 'Cat Purrs Resoundingly.' And when they bandage you up, they don't use tape; they use fur. That's right, you get a fur-st aid kit. And let me tell you, those kittens are the only medics who will make you feel better just by walking into the room. Because nothing heals a soul quite like a purring kitten with a stethoscope, looking up at you saying, 'Don't worry, I've got nine lives of experience!'“
Good attempt? Bad attempt? One for the reader to decide just how strong a derivative work this one really is.
For every realistic deepfake out there, there is an example of incredibly poor use of Generative AI. A particularly terrible and also abject horrifying one was that of Willy’s Chocolate Experience. This event (NOTHING TO DO WITH WILLY WONKA AT ALL, LAWYERS) was held in Glasgow, with the website and Generative AI art promising a world of sweet-based wonders for families. The reality was mostly looking round a warehouse. There were tears, police and a lot of press on this.
To those with any experience of Generative AI, the images were obviously ‘a bit dodgy’. Right now, such images have an unmistakable derivative style to them - a sort of creative Uncanny Valley. Just to make it even more obvious, the images had spelling mistakes all over them (‘ENCHERINING’ instead of ‘ENCHANTING’, I guess?) — the models are not good at handling text, in the same way that hands are still problematic.
Robin Williams: The Test
We’ve established that Generative AI can, to different degrees of success, create derivative works. It’s even starting to flood the likes of Amazon with AI-written e-books. No best sellers though.
When I asked ChatGPT to come up with sitcom episode ideas for fun, it dished up the sort of tedious lukewarm affair that you would find clogging up early evening BBC programming at the weekends. Example: ‘Under One Roof" is a sitcom set in a vibrant co-housing community where families, singles, seniors, and young adults coexist in a large, renovated urban building […] The humor is derived from everyday situations, misunderstandings, and the quirky personalities of its members, while always maintaining a respectful and heartfelt tone.”
Ugh.
Right - The Test!
It’s really simple:
Can your Generative AI model create something inspirational and original such as Robin Williams?
This is the big gap right now. It can’t. Any of the output is always in the style of. Either because the user has asked for that, or because Generative AI can only give OUT what it took IN (from its training) in the first place. There is no mechanism for it to come out with something fresh or new.
Would Generative AI have computed the steps that took 80s music to 90s music? Or would we be listening to the 80s FOREVER? (Maybe that wouldn’t be so bad).
Would Generative AI have taken the steps to move beyond tired Peppa Pig to the blissful family delight of Bluey? Or would we be forever left with the slightly dubious stereotypical family unit concerns of Peppa and her family?
Would Generative AI have taken the creative steps to move from a TV show such as the X Files (ALIENS) to Breaking Bad (DRUGS AND MAYHEM)? Vince Gilligan, creator of Breaking Bad, also worked on the X Files. The idea for the later show was sparked by a glib comment about a meth lab during a telephone call.
This is why the current fears that Generative AI will destroy all the jobs in (insert industry here) are misplaced. Like with any sea change of technology, industries will change. We had all of this with Photoshop and the photography industry. I found articles from 2015 stating authoritatively that a combination of Photoshop and cameras in mobile phones will kill photography. Right now, there are in the region of 50 thousand wedding photographers in the UK - maybe someone should tell them?
It’s helping with developer productivity: Tools such as Amazon CodeWhisperer that can generate code suggestions in real-time are a great help. However, the tool can’t give you a great idea for what you should be creating in the first place. This is precisely why I needed to rely on the imagination of small children when creating my game, Avenger Sharks. I did try using various Generative AI tools to build assets for the game, but the results were always highly variable in quality and not consistent in style — I invested my time into learning rudimentary Photoshop instead.
There’s also ongoing nervousness in the industry about the legalities surrounding the technology. The fact that Generative AI can always spit out what went in is a factor in why 1 in 4 organizations have banned the use of GenAI (Cisco), with 48% admitting entering non-public company information into such tools. A similar regurgitation episode resulted in DPD’s AI chatbot swearing and criticising its own ‘employer’.
How you can help
The next time you are faced with someone telling you that Generative AI is going to create masterpieces or kill an industry, just do the Robin Williams test with them. You can replace Robin Williams with a better frame of reference to suit your conversation.
I suspect Taylor Swift would be more ‘down with the kids’ nowadays.
And would Generative AI have created Taylor Swift?
I thought the Robin Williams joke was excellent - actually pretty surprised GPT-4 generated that.
Great article, looking forward to more!
I use ChatGPT to create programming code and I'm amazed at its abilities to debug. I mean, you'd assume it would write perfect code, but it obviously does not because it only knows what it was trained on, which may or may not have been accurate.
But when I tell it that the code it wrote doesn't work, it will suggest solutions based on alternative ideas, or even make guesses on what might be happening. To me, that's the brilliance of ChatGPT.
Perhaps a future race of silicon-based life forms will produce its own Robin Williams counterpart, and we humans will never understand what it is about that being that makes it so creatively sentient, just like AI will never understand what made our Robin Williams so amazing.