AI Content Archives - DanThornton

It’s always been a good idea to think critically about the information you’re given, or why people might be acting in a certain way. That was true before the digital age, but it’s become more important when you’re interacting with an anonymous person anywhere in the world.

We’ve gone from interacting with people in our village, to receiving news via print, radio and TV, to finding out information from random posts on social networks. With any online publication or community, it takes a little time and effort to figure out who seems to be more interesting or trustworthy, which means most of us will only expend that energy in groups that we’re heavily invested in, if at all.

And humans are psychologically built on making snap judgements at face value. Useful if you’re trying to work out whether that’s a sabretooth tiger about to jump from the bushes, and still handy if you’re being approached in a dark alley. Less helpful when you’re trying to judge the accuracy and honesty of a random internet message.

What happens when we can't believe anything online? — *What happens when we can’t believe anything online?*

Bad and manipulative people have always existed. But technology has helped them scale. Someone untrustworthy in a small village will struggle to find new targets indefinitely, whereas direct mail scams could reach people across the country or around the world.

And online we’ve seen the growth of automated bots and sock puppet accounts. We all receive an influx of phishing emails and messages to our phones on a regular basis. And even those most informed and aware about security can become victims. The old rule about the police and criminals has been flipped, and now they can fail hundreds of times, but we only need to mess up once to have our bank accounts emptied or identities stolen.

But what happens when we can’t trust anything?

There are a lot of ethical questions about AI, from data sources and algorithm bias, to the accuracy of the output.

But what about the impact on all of us, including many who aren’t interested in artificial intelligence, and have no idea of the current and future capabilities of it? What’s the cost to our minds and energy levels when we have to not only assume that we can’t trust anything, but also that it may never have even been near a human? And when that misinformation or manipulation can be done more effectively, at a greater scale than ever before?

The concept of decision fatigue has been debated, but there is some evidence it can impact your cognitive ability, and even your physical wellbeing. Now imagine having to spend time and energy evaluating every single incoming message, every piece of text, every sound or song you hear, and any video or photograph you see for clues about the origin. Not just asking whether someone is trying to spread misinformation for a political or financial gain, but whether your friend has actually gone on holiday to Barbados, or just used AI to fake it. Or even if your friend is a real person at all.

I’ve worked in and around technology for a long time. And there’s always been a trend that regulation comes long after progress, and tends to ignore most of the impacts on normal people. You can argue pretty successfully that the Industrial Revolution raised living standards in general, but it took decades and centuries for that to make a meaningful difference for the general population, compared to changes to society, the environment, and to the individual wealth of mill owners and factory bosses.

And there’s no reason to believe the AI Revolution is going to be any different. We can’t predict all of the impacts, both positive and negative. But we make an educated guess about where the benefits are likely to land, and the disadvantages are probably going to fall.

Image credit: Zoltan Tasi on Unsplash.

Every day brings more generative AI startups, or examples of the apparently amazing things that are now possible. But the maths of AI content don’t add up, and I’m not talking about their ability to get simple addition wrong.

There’s growing analysis of the technology costs, including the infrastructure and servers, training large language models (LLMS), and more. Along with the carbon emissions, and the use of invisible human labour to provide reinforcement learning with human feedback (RLHF). For example, this Ars Technica op ed, or this Andreessen Horowitz analysis. Whether you advocate for an AI future or have strong views against it, the huge cost factors are undeniable.

So there are issues if you plan on creating your own large scale AI service without the funding to match Google and Microsoft. And there are environmental and moral problems, depending on your views about the planet and exploiting cheap labour.

Then there’s the issue of massive private companies using publicly-available content to train the models which replace the original writers, artists, designers et al. At some point there will be a legal resolution to the debate, but in the meantime, those services are using work to profit without any form of permission or compensation.

And a further question about the value of AI content. The current model for creators and publishers is that they are either paid directly by their audience, or monetise the attention they receive through advertising, affiliate programs and more. But when I pay for music and entertainment, it’s partly for access, and partly to compensate the people who made it.

One argument for paying rather than pirating is that it’s quicker and easier for those with money to get access legally. But even when an author like Cory Docotorow has made pretty much everything available for free through legal means, I’ve still bought copies for myself, friends and family. Because I want to offer some financial compensation for the time, effort, knowledge and skill he’s invested in the creative process.

Even if AI could perfectly create new works indistinguishable from the novels of William Gibson, I’m not going to pay for them. Not only do I want some amount of financial recognition and reward to reward the author, but at that point, I could just pay the AI company for access and produce endless works myself.

The Maths of AI Content Don't Add Up — *The Maths of AI Content Don’t Add Up*

What about the workers and craftspeople that AI is inevitably replacing? The always insightful Danah Boyd recently published a great article about the problem of Deskilling on the Job, which highlights how we have dismissed the training and socialisation which comes from automation replacing training at work, socialisation and existing knowledge.

That’s a risk if you’re part of what Boyd terms Camp Augmentation, who promote AI as a tool which will boost human abilities (As opposed to Camp Automation, which sees it as a complete replacement).

If you talk publicly about the risk of jobs completely vanishing, then you’re normally answered by either the belief AI will bring new opportunities, or that we’ll end up introducing universal basic income (UBI) for everyone to support their needs while pursuing whatever interests they fancy.

But the ‘new opportunities’ argument doesn’t stack up. Even in a very quick analysis, creating a short video for Youtube might typically require one or more people to take on roles including scriptwriter, camera operator, presenter, video editor and graphic designer. Depending on the level of production, you could add makeup, directing, audio production, music composition and more.

Using AI, that can all be replaced with one person creating the right prompts. Not only do you potentially lose employment for several people, but the AI version can be created much more quickly, and alongside other projects. So that one individual still employed will be earning for 30 minutes, compared to potentially days of income.

The new opportunities, which don’t yet exist and no-one can yet define, aren’t going to offer the same quantity of roles. That’s the entire point of automating in a capitalist system. And there’s a finite amount of people who are young enough or able to retrain in a role which isn’t as quickly replaced. Barriers to becoming a nurse, plumber or builder include supporting yourself while you retrain, any kind of aptitude, and having the physical ability to do those roles.

Then comes UBI, which even co-founders of private AI companies are suggesting governments will need to fund. With less people working, and therefore less tax revenue given that the leading tech companies behind AI are generally experts in avoiding their responsibilities in that regard.

My favourite take on this comes from OpenAI CEO who suggests UBI could free up people to pursue more creative work. Having removed their ability to earn from it, with Open AI. So his company profits from replacing creative workers, but governments can then subsidise people to be creative.

Again, it doesn’t really add up. Especially when a proportion of creative people are driven by the urge to reach people with their work, as much as the process of creation. But that’s already hard enough if you’re not picked up by large publishers.

There’s a massive amount of discussion and research into UBI, and a number of potential issues even if funding isn’t an issue. While there are benefits including to health and reducing poverty, it has massive implications for social mobility, morality, creativity etc. In Arthur C Clarke’s 1953 novel Childhood’s End, the golden age of prosperity is at the expense of creativity, before humanity becomes extinct. And while AI isn’t directly equivalent to alien Overlords, our species always seems more threatened by the banal and mundane.

And ultimately, I come back to the AI question that always perplexes me. Given that AI can achieve amazing results in analysing large data sets, particularly in narrow fields of expertise, why have so many people chosen to focus on generating content rather than investigating potential symptoms and cures for chronic diseases, or other tasks more valuable to us as a species?

I’m not aware of any chronic content shortage in human history? Access might be limited through censorship or paywalls, but I’ve already got more books in my various wishlists than days left in which to read them. Which is good, considering once AI has devalued the worth of publications, I’m not particularly interested in reading them…

So how do you resolve the fact that the maths of AI content don’t add up? Short term profits for large companies and big investors.

Share this:

Share this: