Technology should enhance humans, not replace them
This morning I was listening to an interview by Neil Gaiman on an Icelandic TV channel. The interview itself was pretty short, only around 7 minutes or so. He talks about a variety of topics but there a specific thing he mentioned that I wanted to comment on here. I don't remember his exact words but it was something along the lines of:
Humanities are disappearing. There's a lot of focus right now on the importance of kids (and adults) understanding science and engineering, and I think people sometimes forget that we created these in the first place for the humanities.
In other words: the goal should be to make our lives easier, not to replace the humanities.
There is a lot to be said here about how science is important but it's humanities that make life human, that make it worth living, but I wanted to constraint myself to talk about what the current LLM1 phenomena is doing to humanity. These are being marketed more and more with the goal of making it easier for anyone to create stuff. But who is actually doing the creation? Is it the person who prompted the model to create that beautiful image? Is it the one who asked for a wonderful kids' bedtime story?
I would say no. The creator is the model, of course. The users are just requesting what they want and the model generates it for them2.
If we take a moment to look more deeply into this then it becomes obvious that, if we advance time a bet, what AI is evolving towards is a situation where humanity is offloading its creative agency to "other beings", and relegating itself exclusively to the role of consumer.
This ties back to Niel Gaiman's comment. These tools have a huge potential to help humanity be better but the way they're being (in their majority) sold to us is as a replacement of what it means to be human3.
All the existing uses for LLMs and motivations people have for using them can be (mostly4) boiled down to the fact that we're each trying to leverage this new technology to get one-up on each other, to be more productive, have more social merit, get ahead in the race.
Coming back to LLMs' role in replacing the human creative agent, I don't really think this will happen. Or at least not entirely. Humans have too much of a creative spark in them for all the peoples of the world to go along with this story. I would actually argue that — in a world where Netflix doesn't record their entertainment offerings beforehand but generates them on demand based on the viewer — the real, honest creations will be a premium commodity and will themselves be more appreciated than they are today.
In some way we're already seeing this shift with lots of people abandoning (entirely or in part) the use of centralized social media sites such as Facebook and Twitter, and instead preferring to keep a blog and follow a handful of other people, preferring the intimate over the impersonal. We're seeing a shift from quick-easy entertainment to valuing more the individual thoughts of other humans5.
I imagine the internet user of the future will likely employ a mix of both.
On one side they will have Netflix-LLM or some other monstrosity that generates content for them on-the-fly based on their mood and what they had for breakfast. This content will be 99.99% accurate to the tastes of the viewer and will be manufactured in such a way that it will evoke a certain emotion. This is more or less how movies work now, but producers have to consider the sensibilities of a lot of people at the same time. It's likely that this same kind of fast-food entertainment (fast-entertainment?) will also be available in shorter form (for example, a nightmarish Reel-LLMâ„¢).
However, the people of the future will also have another form of content they consume which is directly made by other humans6. I imagine this kind of content will be long-form blogs much like the ones you can find on Bearblog.
Of course, other human created content (e.g. movies, books) is not really going anywhere. I imagine it will be sort of like owning handcrafted furniture vs IKEA stuff. The IKEA sofa is convenient and comfortable, but only the handcrafted will be considered art.
LLMstands for Large Language Model. They're a family of machine learning models that are especially good at dealing with language and manage to generate pretty good text. The most popular example, right now, is ChatGPT from OpenAI.↩To be fair, there is a nontrivial amount of skill that's needed to make the model produce what you want. But I would say this is only temporary and will become less and less important as these models get smarter.↩
This comments apply mainly the way humans have thought of using this models and not about research (which can't be stopped, there will always be someone curious enough to go poke the weird slimy blob).↩
I should recognize that not everything is bad. There are some legitimately nice uses out there. Like having LLMs help people understand topics, or helping those that want to make art but are not able to do so for some reason. The line is blurry at best, and it's located in different places for different people.↩
I'm most definitely biased on this, since I haven't done extensive research on the topic and the observation is limited to my personal experience and social sphere.↩
And it won't be informational (as in how to do X) since LLMs will be better at explaining things using the best pedagogical strategy for each individual. These human posts will, by nature, be about human connection and ideas.↩