Read Time:7 Minute, 21 Second
This is likely to ramble a bit, so grab some snacks.
I have been mulling over this excellent episode of the Publishing Rodeo for a couple of days now, trying to put my finger on why it rattled around my head so. https://open.spotify.com/embed/episode/1lMUiKqG7PSlJjx7LoOHkU
If you do not listen to the Publishing Rodeo and you have an interest in the traditional publishing world, I suggest you do. Sunyi Dean and Scott Drakeford debuted with the same publisher in the same year in the same genre and so far have had very different careers —she lead author, he forgotten midlist (sorry, Scott). The podcast explores the decisions out of either of their control that led to said state of affairs and is an excellent education on the industry.
The episode in question was an interview with Dr. Kerry Spencer Pray an author and scientist who published research that essentially proved that the only things that mattered in whether or not your book sold was how famous you were and how much money you got as an advance.
Now, obviously, things are more complicated than that summary. Fame actually had only a small effect, the advance size is a proxy for how much marketing budget you got, and there are a couple of other marketing items that had small effects, such as cover likeability and hitting the market at the right time for a specific sub-genre. But measures of quality had no discernable effect on the success of a book. I suspect that is a bit overblown — the acquisition process likely ensures a floor of quality, so I suspect the true statement is more like once you reach a minimum quality, the quality of your book does nothing to help your sales.
Dr. Spencer Pray was able to combine this information into what she called a minimum viable marketing score and use that score to very accurately predict whether a book was going to sell. Hit her numbers, and your book would be successful. Fail to, and your book would sink like a stone.
I want to reiterate that I in am in no way questioning Dr. Spencer Pray’s results, qualifications, intentions, or methodologies. I am absolutely certain that she is correct in her findings, and I am frankly fascinated by her work. But it all felt a little bit off-center to me, in the same way that much of the A.I. as art discussion has. It might just be that as someone who wishes to be a published author, I find the idea that quality having no meaningful impact on sales distressing, but I don’t think that is the reason. Rather, I think that Dr. Spencer Pray’s work, as fascinating as it is, is an exemplar of how we sometimes use data as the easy way out.
Perhaps even easy way out is too strong a term. Data is a good marker of what works in terms of what we can measure, but it is only that. Dr. Spencer Pray discussed that she could find no correlation between the numerical rating of an author’s work on Goodreads or Amazon and the sales of their books, for example. Scott Drakeford echoed her findings with work of his own. The proposition then is true — as far as it goes. But as Dr. Spencer Pray would tell you, I am sure, data is only as good as our ability measure it. And it is easy to fall into the trap of discounting items that we cannot measure.
Recently, an anime account named Bigolous Dickulous, for our sins, tweeted that they loved the book This is How You Lose the Time War. Dickies tweet was picked up and retweeted by other fans of the book, all extolling its excellence. As well they should have — it is a brilliant little story, well told. Go buy a copy if you haven’t already. As a result of this word of mouth, the book, which had been out for about two years at the time of tweet, rocketed up the best seller lists. All because people rallied around a book they loved.
Obviously, that is a unique situation. But books get pressed into the hands of friends and loved ones all the time. People gush about their latest favorite reads to their literary friends constantly. Just try and get me to shut up about the best books I have read. Canut had better luck with the tides. But tracking such behavior is well-nigh impossible and generating it as part of a marketing plan is not something marketers have ever been very successful at.
I think that is what made me uneasy with the lessons of the episode. I think we as a society, as an economy, as cultural stewards may rely too heavily upon data driven decisions. Data is about what works now and in the past. And again, I stress: I am not anti-data. I am a dork. I love numbers. I love their exploration, and data driven decision making certainly has a place in the world. I am just afraid that we sometimes give it too much credence, too much authority.
It can be the easy route — if you do this, then the data says that you should have some measure of success. It looks to the past because the past is where the data comes from. And it looks to what it can clearly measure, because that is what data is — the measurement of what can be measured, of what can be seen. But what can be measured and what can be seen is not the entire universe of what can be done, of what options are available. By relying on data so heavily, we can blind ourselves to trying non-data approved actions. We can fall into ruts, refuse to experiment outside the bounds of the easily measurable. It is hard to take risks, hard to do new things. If you rely too heavily on data, then you risk atrophy as you always do what you have always done, because the data says it has always worked. Until one day, it doesn’t.
What has this to do with A.I. and art? Maybe nothing. But generative or imitative A.I. functions by remixing data from the past, by calculating what word should come next based on what word had come next in similar situations in the past. The wits among you may say that this is all any writer does, but I don’t think that is the case.
During school and corporate icebreakers I always say that I have never done anything interesting, but that is not actually true, as it would not be true for any human being. I have worked myriad of terrible jobs, from comic book warehouses to blackjack dealer, to pay for college. I have been in fights, been held up at gun point, sometimes unsuccessfully. I have been well off and so poor that I didn’t know where my next meal was going to come from. I have outlived people I shouldn’t have, regretted saying no, and regretted saying yes. None of those are unique experiences, of course, but the combination of them at the specific times and places I experienced them are unique enough that they bring a different perspective to my writing than purely recombing different styles alone does.
Humans, unlike generative/imitative A.I.s are not word calculators. They are not just remixing the things they have read. I could be wrong of course — I have yet to be published, yet to achieve even representation. As far as the publishing world is concerned, I suck. Be that as it may, I am convinced that humans create art not by remixing alone, but by adding their perspective to the styles they have learned. That is what makes a story worth reading, as corny as that may sound. The little bit of the writer that shines through in a piece of art is what makes art, whether it is low, high, pop, or the snobbiest snob that ever did snob, art. It all starts and ends with someone putting a bit of themselves into its creation.
Data, no matter how well remixed, cannot provide that spark. That is why the danger in A.I. is not replacing humans but rather expecting humans to force their spark into the superstructure created by A.I.s for less pay and credit.
A lot of words, I know, for probably not a deeply profound revelation. Sometimes, though, it’s worth reminding ourselves of the simple truths. Take it from your friendly neighborhood computer dork: data is not where knowledge ends. It is where it begins.
I hope you enjoyed your snacks.
Happy
0
0 %
Sad
0
0 %
Excited
0
0 %
Sleepy
0
0 %
Angry
0
0 %
Surprise
0
0 %
0
0
