AI and the Cult of the Lazy Amateur

We are all familiar with the phrase. We learn better when we write things by hand, rather than when we type them on the keyboard. For decades people have been against typing, saying “It’s cold, it’s dead, it’s impersonal”. Those same people will then say “you have awful handwriting” and yet still feel nostalgia for hand written notes. In the age of AI more is being lost.

Ten Thousand Hours of Practice

When people my age were growing up, if you wanted to sing, you needed to devote ten thousand hours to develop tone, timber and a good voice. I am not a singer, which you can tell by what I just wrote. The point is, to be a singer you needed to dedicate yourself to the craft.

Photography

Some might say that digital photography is much easier than analog, because if you make a mistake you see it instantly. You can also take plenty of photos until you get it right. With film cameras you have 36 photos per roll, and it’s when you develop the film that you see if you made a mess or not.

Video

Years ago I had the VX-1E video camera, and it was replaced by the VX-2000, and then another after that. With broadcast cameras, and with video cameras from a different era the zoom was manual, not servo controlled. You could adjust aperture, focus, and the zoom manually. With newer cameras you lost more and more direct control. Control is now electronic, rather than tactile.

Now, with DJI drones, and other camera gear cameras pan and tilt automatically to follow the action and a single operator can control four cameras at once via a console. A human is no longer behind each camera.

We hear about AI taking jobs, but robotics replaced camera operators a decade or two ago.

As if that wasn’t enough, editing tools do more and more of the work, with the aim of replacing the professional video editor with an amateur influencer. This is where we get into Andrew Keen’s cult of the Amateur and AI topic.

Our mobile phones are now our cameras, and image stabilisation makes phone camera footage usable. This means we do away with the camera operator. With new video editing tools that automagically edit video, very badly for now, the need to learn Montage Theory is vanishing.

If you watch YouTuber content, it’s recognisable. They all use the same sound effects, the same transitions, the same lazy approach to presenting to camera. If any YouTube content was shown within a university or film school setting, it would be panned as kitsch, cliché and worse. On YouTube, where the cult of the amateur reigns, the new sub-culture has become the norm.

I use the word sub-culture, because although I hate the style, it is popular, and those from newer generations than mine, appreciate and enjoy the style.

When I was growing up I read the Film Sense by Sergei Eisenstein, and I read books about basic betacam work, basic editing, and many other books. Because editing systems would cost 30,000 CHF for a player, and 30,000 CHF for a recorder deck, and then yet more for the edit controller I learned to edit on paper, and in my mind, impatiently waiting to have access to edit suites.

It’s with the Miro DC 30+ and Adobe Premiere that I could really learn to edit, as well as with the DHR-1000 Sony edit deck. When I first bought Final Cut Studio it cost me 1600 CHF. A few years later I bought Final Cut Pro as an individual app for 300 CHF, and that license has been valid on every mac I have owned or used since then.

As I look on Threads especially I see amateur video makers, more flatteringly for them, referred to as influencers, or YouTubers speaking about how long editing takes, and how hard it can be. The more you edit as a professional, the faster, and more efficient you become as an editor.

Years ago I watched an editor called Jesus, editing on Avid Media Composer and he was really fast and efficient with the edit suite. He knew the edit suite inside and out. For him that edit suite was part of him. I have seen many professional editors who are like that, who know the tool inside and out.

In contrast, with CapCut, and GoPro’s edit suite, and Virb, and other editing packages, the edit suite does the work for people. They shoot material, and then the edit software prepares the rough cut, and they fine tune it.

In some cases they use AI to tidy it up rather than learning how to do this themselves. They use AI to colour grade, and to add AI generated music, and more. In the end the art, and auteur aspect of editing is being lost, and replaced by something more generic.

Books and Audiobooks

I noticed on Audible that a lot of books that are included within the subscription have AI narration. For some, this might be welcome, but for me, if we pay as much as we do for audiobooks, they should be narrated by human beings, rather than AI.

When I see that books are written by AI I don’t see a point in paying for them. To me, the point of paying a person, is to have their vision and creativity, rather than paying for something generated by an LLM. If I wanted to read something generated by an LLM, I would pay for the LLM and provide it with my own prompts, and fine tune them.

Why would I pay for content generated by an LLM, if I can provide similar prompts and generate it by myself, for myself.

AI Content Farms and Spam

A few weeks ago we saw posts about people generating podcasts and blog posts with AI. When people wanted to create blog farms, in the past, they were immediately flagged as spam. For some reason today, the desire to generate blog posts and podcasts with AI is acceptable and yet the question I ask is “If you can’t take the time to create content with humans, then why should humans read, or listen to it? We are already innundated with human created content, without adding AIGC to the mix. AI Generated Crap.

In the age of information overload it doesn’t make sense to use AI t add to the noise. It makes sense for the opposite. No one benefits from AI adding to the noise.

AI for Recommendations

When I look at YouTube, and Google News, and other sites AI is used to recommend stories, but I feel that AI looks at the lowest common denominator, for everyone, rather than within our content niche. It would make sense for AI to learn from my habits, and recommend according to my interests, without using the interests of others to recommend content. Too often the recommendations are awful. At other times it’s too narrow. It’s not as tailored as it could be.

AI for Upscaling videos

In the early days of YouTube we uploaded videos that were as light as possible to YouTube because we lacked the bandwidth to upload bigger files, but also because people lacked the bandwidth to download larger videos. We have forgotten the times when we waited for videos to load.

Sometimes we would press play, and then pause, and let the video load while we went to get something to drink or a snack, and then come back once the download was finished.

One way to save time was to upload low resolution videos. The material might have been shot in 720*576 (SD), but we downscaled it for the sake of sharing on YouTube. People like me have our first videos dating to 2005 or so. Things have changed in 20 years.

In broadcasting upscaling has been around for years, for SD footage to be broadcast today. Look at Fasier, Friends, and other series. Plenty of old series were shot in SD, and are now broadcast in HD, after being upconverted.

YouTube is using AI to upconvert their back catalogue and although some people hate the idea, I see the value. A 240, or 320 video, on a 4k screen is a stamp. Some of my videos, by being upscaled, will gain a new lease of life.

For me, upscaling is a good use of machine learning/LLM/AI technology because it fills a niche that, as humans, we would spend months or even years, trying to do manually.

There is a chance that LLM/Machine learning will hallucinate and distort what people see, but at the moment low res videos can’t be played full screen. In an ideal world we would revert to the edit files, and re-export them in higher resolutions using new technologies, but as things stand, YouTube requires less returning to archives.

Archive Restoration

When I used to listen to MacBreakWeeely I found it interesting to hear about how machine learning was being used to compare every frame of certain films, in order to do a digital restoration of old films, as well as their film archives.

I have over a hundred tapes that I should digitise, and then restore. Beta SP, MiniDV, DVCAM and other tapes are great because information is stored on tape so if part of the tape fails, more of it is left. The drawback is that if you play old Beta SP tapes the sound is played more slowly than originally so it has to be cleaned up. The same is true of dropout with all of these tapes.

Although the cult of the Amateur might say “We don’t want our footage to be upscaled, because of hallucinations, upscaling so much footage will help train the models use to do a better and better job, and as we digitise and salvage raw footage from old digital video tapes, and analogue tapes, so we recover historic artifacts that could have been lost, or hard to watch.

And Finally

For me Ai/machine learning and large language models should be used to streamline the tedious parts of video, photography and more, without replacing human creativity. I don’t think that writing a prompt is creative.

With Photoprism, and Immich, we see how machine learning can be used to help sort and catalogue gigabytes of photos with locations, names of people, names of mountains and recognising colours, and seasons and more.

I will value human generated content over AIGC anytime.