Weeknotes 5

AI, of course.

Mar 24, 2023

AI you worried?

Turns out I’ll be writing about AI *a lot* over the course of the next… forever? AI is going to be a similar paradigm shift, like the democratisation of the web in the late 90s. Or perhaps like the advent of mobile that changed the way how we interacted with web content in the late 00s? I’m also probably underestimating the impact of it here. There are so many interesting topics to unpack, and the most fascinating to me are human-machine interactions.

We will see uncanny valley-esque worries. We will see a raise of mistrust caused by bullshit content looking 150% legit. It’s hard to say what else we will see, we’re entering what Cynefin calls Chaotic problem space. The only way forward is to act and observe. What could possibly go wrong?

A friend recently recommended reading Superintelligence - if you’re curious about what could possibly go wrong then read this book. I can guarantee a few sleepless nights. They say curiosity killed the cat - I guess humanity might as well, possibly, become the cat in the AI story.

AI crooks

But there’s more immediate worry I have. A worry that’s much closer and way more immediate than AGI.

It won’t take long until tools like Chat GPT get integrations with mass messaging platforms. Chat GPT produces content that can easily pass for a human. Based on these two premises, and on what we know about the world of phishing and spam, I dare to predict

Wave of AI-generated spam in our inboxes
Smarter phishing messages
Phishing calls / interactive phishing messaging bots

This is definitely coming for us. We should brace for impact.

Noindex

Remember robots.txt?
Remember <meta name="robots" content="noindex" />?
If you’re unsure check this out.

We desperately need something like this for web content. Perhaps you noticed the University of Chicago’s Glaze Project enabling artists to protect their visual art against being used as training data for AI text-to-image generators. I love it for the principle that every author should be able to define how their content gets used.

Yesterday, I saw Chat GPT now has a Bing plugin that allows it to access web content through Bing search. I can imagine a lot of authors will want their content protected against AI training for many reasons (whatever they are). Again, to me, this is entirely legitimate, and not dissimilar from the robots.txt case. Based on my understanding, Chat GPT will use ChatGPT-User super agent based on human user instructions (so it’s not a web crawler). Looks like Disallow will still work.

Once there’s Chat GPT web crawled then… we’ll see I guess.

It also reminds me of the publisher x platform wars. Content created by e.g. news publishers getting consumed via aggregators or social networks, which then benefit from user engagement and advertising.

The power of the web is in diversity, and I don’t think we want to end up having all web content getting fed to AIs.