Agreed that the studios need to be held more accountable and their usage of AI is more problematic than open source last resort type work. I have noticed a degradation of quality in the last five years on mainstream sources.

However, the existence of this last resort tool will shift the dynamics of the “market” for the work that should be being done. Even in the open source community. There used to be an active community of people giving their voluntary labour to writing subtitles for those that lacked them (they may still be active I don’t know). Are they as likely to do that if they think oh well it can be automatically done now?

The real challenge with the argument that it helps editors is the same as the challenge for Automated Driving. If something performs at 95% you actually end up deskilling and stepping down the attention focus and make it more likely to miss that 5% that requires manual intervention. I think it also has a material impact on the wellbeing of those doing the labour.

To be clear I’m not anti this at all but think we need to think carefully about the structures and processes around it to ensure it does lead to improvement in quality not just an improvement in quantity at the cost of quality.

It is probably good that OS community are exploring this however I’m not sure the technology is ready (or will ever be maybe) and it potentially undermines the labour intensive activity of producing high quality subtitling for accessibility.

I use them quite a lot and I’ve noticed they really struggle on key things like regional/national dialects, subject specific words and situations where context would allow improvement (e.g. a word invented solely in the universe of the media). So it’s probably managing 95% accuracy which is that danger zone where its good enough that no one checks it but bad enough that it can be really confusing if you are reliant on then. If we care about accessibility we need to care about it being high quality.

I won’t rehash the arguments around “AI” that others are best placed to make.

My main issue is AI as a term is basically a marketing one to convince people that these tools do something they don’t and its causing real harm. Its redirecting resources and attention onto a very narrow subset of tools replacing other less intensive tools. There are significant impacts to these tools (during an existential crisis around our use and consumption of energy). There are some really good targeted uses of machine learning techniques but they are being drowned out by a hype train that is determined to make the general public think that we have or are near Data from Star Trek.

Addtionally, as others have said the current state of “AI” has a very anti FOSS ethos. With big firms using and misusing their monopolies to steal, borrow and coopt data that isn’t theirs to build something that contains that’s data but is their copyright. Some of this data is intensely personal and sensitive and the original intent behind the sharing is not for training a model which may in certain circumstances spit out that data verbatim.

Lastly, since you use the term Luddite. Its worth actually engaging with what that movement was about. Whilst its pitched now as generic anti-technology backlash in fact it was a movement of people who saw what the priorities and choices in the new technology meant for them: the people that didn’t own the technology and would get worse living and work conditions as a result. As it turned out they were almost exactly correct in thier predictions. They are indeed worth thinking about as allegory for the moment we find ourselves in. How do ordinary people want this technology to change our lives? Who do we want to control it? Given its implications for our climate needs can we afford to use it now, if so for what purposes?

Personally, I can’t wait for the hype train to pop (or maybe depart?) so we can get back to rational discussions about the best uses of machine learning (and computing in general) for the betterment of all rather than the enrichment of a few.