AI In The World of Publishing
Working with some authors on their websites, I realized quite how viscerally many people in the publishing industry are reacting against AI.
In particular, there is an extreme level of concern around exposing their work to AI of any kind, largely driven by a legitimate fear of machine plagiarism.
As a technologist, I believe that the horse has already bolted. Actually safeguarding the privacy of any work-in-progress would require abandoning many of the tools we take for granted (email, online collaboration, cloud storage). Although many of these services purport to provide privacy by encrypting data, most of us rely on vertically integrated stacks (google, apple) which could easily be compromised to enable the provider to snoop on us. For privacy, we are forced to trust Big Tech. Personally, I don’t. My gut feeling is that if Google, Apple, or Microsoft wanted my data for AI training they would just change some detail of the pages and pages of T&Cs I click through every time they force an ‘upgrade' on me, and they’d have it. Microsoft’s USA ‘privacy statement’ (https://www.microsoft.com/en-us/privacy/privacystatement), for example, includes the following clause:
As part of our efforts to improve and develop our products, we may use your data to develop and train our AI models.
However, I do believe there is cause to hope. We have already established that AI generated work cannot be copyrighted. This is a necessary first step and must be defended.
AI providers are starting to watermark generated text. This is not for the benefit of authors. For their business model to survive in a world where AI is ubiquitous, AI providers have to be able to distinguish between AI and human writers. Otherwise, the majority of data used to train AI systems will be AI generated, which would be a huge problem. Because it is in their interest, I trust that this will happen far more than I trust Big Tech in general to safeguard information which, ultimately, wants to be free. Legislation to mandate this watermarking would be an important next step and I think activism on this will ultimately be more useful to the publishing industry than trying to prevent an author’s work from ever being used to train AI.
There is even cause for optimism. If AI is embraced as a critique tool… life as a writer could become much easier. Imagine a world where publishers and agents are able to train or prompt AI models to give instant feedback based upon their own editorial policies and preferences. Imagine writers were able to submit work to hundreds of publishers and agents simultaneously, receive instant feedback, and then focus their attention on those which found their work most attractive? Languishing on someone’s slush pile for months waiting on tenterhooks for some crumb of feedback could become a thing of the past.
On reflection, this might not be a bad model for the software industry. Developers, unlike writers, generally have a positive feeling about working with AI, perhaps because software development has always been a collaborative process, often massively so. Or maybe because developing software can be tedious in the extreme, and AI can help us avoid the worst jobs completely.
However, concerns remain about the impact of the technology in the long term. AI is certainly very good at developing new applications, but developing a new application was never that expensive relative to the total cost of ownership in the longer term. Whether AI is as good at maintaining software that is in use remains to be proven. AI can certainly help, but the case for requiring a human in the loop when managing production issues is much stronger. Vibe-coding may be acceptable, but vibe-maintaining is an altogether riskier strategy.
If AI leads us to create more software whilst also de-skilling the people who must maintain it, we are heading for a problem in the longer term.
So… should AI-generated code be watermarked and free? Should we try to steer the industry towards using AI as a critique tool and trusted advisor for building commercial software, rather than as the primary engineer?
Comments welcome.
Photo by Steve A Johnson on Unsplash
