The Primitive Puppet Show
21% of the internet is AI pretending to be human (2). How can we use AI to improve communication rather than thwart it?
It is a social failure. A dozen emails slip past my spam filters each day that look personal, and are not. This is also true of many of my text messages, and it’s starting to show up in scripted phone conversations. The social media platforms I once used to build and maintain business relationships are filled with AI-generated garbage. It’s the same with the Discord server that we use as our community chatroom. The whole internet feels like Five Nights at Freddy’s - a nightmare of automatons with no human beings in sight.
It’s not just me that is encountering this, of course. I recently talked to city employees in Arlington, Virginia about the explosion of well-constructed legal FOIA (Freedom of Information Act) requests clearly written by AI. The initial prompt must have been something like “Can I cut down my neighbor’s tree?”.
This is all going to get much worse.
This social failure has far-reaching consequences. I run a tech startup. We offer support and tools around an open source project we maintain. Community is really important to us. It’s how we find our customers, and how we sustain our team. We–like much of the rest of the world–need a functioning internet to survive.
I believe in our future. Not because of AI, nor in spite of it, but because people in a free society tend to both embrace change and find balance through collective effort. In this series of articles I’ll take you on our journey to apply AI in ways that foster collective effort, bring us together and enable greater collaboration and open communication.
The First Objective: a trusted feed of fresh information. I can’t get that from Google searches anymore. LinkedIn is no longer in the business of helping me keep up with my connections and their interests. Facebook isn’t even worth opening. I want a curated list of news from blogs and RSS news feeds I respect. I want to get clued in on emerging trends and new voices.
The First Attempt: the simplest possible Python program. This is just research - a throwaway application. Without getting caught up in anyone else’s libraries, tools or frameworks, can I write a minimal amount of code that will create a good news feed? Is it even possible, and what happens when I attempt to do it?
You can see the results so far on GitHub in an open repository called research-agent. You don’t have to go read the code, but it is there for those that want to dig deeper.
Here is what I have learned so far:
Building a list of trusted sources is hard work. Just like having friends and maintaining good business relationships - this is an act of eternal vigilance. Most of the world isn’t going to put in this effort. That should jump out at you as an opportunity to get ahead.
The internet is learning to hate bots. More and more of the resources on the internet–particularly the good ones–are not accessible to automation tools. Everyone wants to be discoverable, few want to be “pre-digested.”
Kagi is super helpful. I am not affiliated with Kagi, I just give them my money each month. Kagi is a paid search engine with an API and some powerful search tools. It is proving invaluable to gather open data from the internet. Their Lens feature and ability to search by date range are useful in this context.
Diversify your AI dependencies. The ability to use different models for different problems is important for many reasons - money, flexibility, security. Everything I will evaluate moving forward will allow me to switch easily from one LLM to another.
I’m running this code manually, and I’ve been happy with the results. I’d welcome pull requests to improve it, or forks that show how this would run better and easier with leading frameworks.
Next week we will get this out of raw code, and into visualization tools like N8N and SpiffWorks to see if we can make these kinds of tools easier for non-programmers to read, understand and update. The real power of AI is ultimately in making software like this easier for everyone to create.
I am one of the founders of SpiffWorks, a business process orchestration tool. This series of blog posts is intended to be an honest journey. I will bring in some of my company’s software from time to time. I want to make sure our software is really good at building agents that are useful and not harmful. To do so, I’ll need to research what everyone else is doing. Follow me. Comment. Keep the human conversation going.