Forum: Author Hangout

Yes ai has pretty much taken over some image sites hasn't it, very suddenly! The images can be extremely perfect, as it were. Evocative and apparently meaningful.

I don't see why writing can't go the same way, rearranging existing elements into 'original' new creations.

I don't see why writing can't go the same way, rearranging existing elements into 'original' new creations.

I wonder whether that would split readers into two camps. Camp 1 would be those who are happy to read what are effectively the same stories time after time, with just a few name and plot location changes. For them AI would be the best thing since sliced bread.

Camp 2 would be those who seek out the innovative authors who come up with original plots, and who are possibly undervalued in the pre-AI era.

AJ

I find the whole thing hard to predict, I mean who'd have seen this coming, from LLM? But my current plot is probably easily predicted, as it's basically my usual teen girl power nonsense with a dash of The Power - it's one premise and everything follows from that. And is there a plot that ai couldn't arrive at? How original would that have to be? And yet credible...

Barring malicious AIs, AI cannot alter reality. What it can do is create plausible non-real works which are easily believed to be real. The problem is with the viewer, not with the tool.

While my cynical side would agree that some 'problems with viewers' are intractable given human nature, to the extent that there's a solution it will come from educating everyone to the potential misuses. The genie is long since out of the bottle.

The recent kerfuffle with a photo showing the Pope in a nice coat is exactly what we need more of. The more people see things that look real but are easily proven to be not real, the more most people will start to build filters to question what they see. It's much easier to build up those filters when the image itself isn't something most people have a strong emotional connection to. It's much easier to move from 'Pope in a coat can't be assumed true' to 'images in general can't be assumed true' than it is to move from 'politician X doing something bad can't be assumed true' to 'images in general can't be assumed true,' because many people are automatically inclined to either strongly accept or strongly reject the image based on their feelings about 'politician X'.

I'm comparatively less worried about written words. Written words have been used to lie to people for centuries, often successfully. The people who are going to fall for it are going to fall for it regardless of whether it's a misused AI or a human.

Social media is an enormous problem, but I think the reasons are very different for the most part.

Newspapers of various sorts have gotten agendas and only produced false narratives, and that goes back well over a century. That's why I continue to not be that worried overall.

Or, rather, I'm exactly as worried as I am about information silos and confirmation bias in the broader case, and I'm very worried about that. I'm just not sure that AIs make written (dis)information any worse. People have been doing an extremely good job at disseminating written disinformation for a very long time.

I was referring to newspapers. We've had newspapers with set ideological bents for quite a while (back into at least the 1800s). Anyone writing a story with a significantly different take would have to find a new employer (and that might be in some other city).

That's large, mainstream, middle-of-the-road newspapers. They mostly still do that, except for, say, the New York Post (or the Wall Street Journal, which is admittedly a national, not a local).

While I don't watch CNN, I do routinely see outrage posts because CNN had on a right-leaning guest, so I'm pretty sure that they have them regularly.

I was referring, say, to the Hearst era and like, where crossing the newspaper would end your career with that newspaper and likely have you widely blackballed.

I detest information silos, but I don't see AI making them meaningfully worse in terms of written content. In terms of visual and audio content, however, there's an absolutely enormous range of new misinformation/disinformation possibilities.

I thought the NY Post was more of a gossip rag then a middle-of-the -road newspaper.

It gets a lot of buzz from the gossip section. But the actual news stories they have are actually really good. Just to pick one - they're the ones that broke the Hunter Biden laptop scandal, in full detail, and were excoriated or ignored because of the 'disinformation' they published. Then, when the story was vindicated, they didn't quietly acknowledge how they were right, they shoved it into the faces of everyone who'd ignored them or otherwise dissed them. They're also leading the charge for the NYT to have the Pulitzer revoked after the Durham report came out and what they've said all along was vindicated. And they don't carry water for Mayor Adams - they're holding him accountable for crime and following up on his campaign promises.

Trying to avoid politics (and failing, of course): the Post's own reporters refused to report the 'Hunter Biden laptop scandal', and the Post published over their objections. As of today, the Post reporters' objections appear to be nearly as valid now as they were then.

If you believe any number of pundits and theorists (Joseph Campbell, Anna Quindlen, and Stanley Kubrick came on in a very quick search), every story has already been told. What remains is telling the same story better.

I'm not sure that I agree, but it's certainly the case that the vast majority of stories can, in one way or another, be reduced to well-understood archetypes and plotlines, sometimes with an amazing amount of embellishment.

My suspicion is that Camp 3, being the camp who wants 'original' plots (mostly this means well-written, with interesting characters, etc, rather than some abstract 'truly original' criteria being met) but is willing to accept relatively seamless AI 'fills' for parts of the story, will be the camp that comes out of this in the best position.

I haven't done this yet, but I've read comments from authors who started doing it pre-GPT3 and had good results. The core of their stories is theirs. The lead character driving across town to a meeting is an AI fill, as might be a description of the office they're visiting, etc.

As a very indirect but (at least for me) apropos example, Edsger Dijkstra, a legendary computer scientist (and professor) refused to use word processors (and very seldom used computers at all). At the time, there was a considerable controversy over whether he was a visionary standing for the importance of logic and mathematics over simply understanding progressing, or a luddite refusing to use one of the most basic tools his profession had created.

The truth is, of course, somewhere in the middle. Dijkstra had a point - the mathematical / 'theory of computation' side of Computer Science (by which I mean logic, proofs, etc, not manipulation of numbers) is critical to many elements and is often neglected. That said, one has more time to think about hard problems if one isn't spending time pushing a fountain pen around a piece of paper.

And, as another side anecdote, Steven King has said before he prefers using a fountain pen because it's slower and he needs to think about which word to write since the cost of correcting is higher.

An all-AI work will certainly satisfy some people, though I would argue that those 'few name and plot location changes' will be so obfuscated that most people will consider the work to be new.

And some people will only be happy if every word was written carefully, with thought, using a fountain pen.

But many people will be in the middle, wanting interesting characters and plots that (at least for now) are beyond the capacity of an AI to reliably generate but willing to allow much of the 'background' to flow from guided automation.

One important caveat: current state-of-the-art technology (GPT4, etc) is based on a convergence of some theoretical breakthroughs a while back, the availability of cheap high-end computing hardware, and inexpensive access to high volumes of source text. Breakthroughs are unpredictable, but the odds of computing power continuing to expand are high.

Simply from computing power growth and the relentless expansion of the internet, the second and third factors are likely to continue to expand, and when we get to GPT5, 6, 7, etc, it's likely that some levels of 'creativity' currently impossible will become commonplace.