Document Interrogation: Reading 2.0
Most information is trapped in texts.
Essential facts are hidden deep in the pages of Geological surveys, technical manuals, and thick textbooks.
Reading is how we extract information from these textual knowledge containers. Word by word, we read the text to get the information we want. Some of us scan. It’s a little faster. And many look for a shorter text that summarizes the long text we didn’t want to read.
To get knowledge, one must read.
“The answers you get from literature depend on the questions you pose”
- Maragaret Atwood
This extraction is a one-way conversation.
The text exists and you read it. You can ask a question but the text doesn’t answer. It won’t adjust its tone or shift its diction. The text is static. It’s being read.
An alternative way to get knowledge is talking to someone who read all those long texts. Conversation is interactive. You can stop when the story drifts. Ask for clarification. Even challenge points and receive a revised argument.
Books can’t talk though.
But what if they could?
Turning Documents into Chatbots
Artificial intelligence (specifically Large Language Models) offers a new way of retrieving knowledge. Talking directly to it.
By training AI models on specific documents, reading can become an actual conversation. This is a process I call “Document Interrogation”. By turning any text into a chatbot, you can talk to it, question it, even ask it to explain “like I’m five years old”. You’re no longer a captive audience to a text’s static monologue. You are in conversation with a living text.
The idea is not restricted to single texts either. Ideas, bodies of thought, intellectual movements can be transformed into query-able chatbots. Any body of knowledge contained across a discrete number of books can be compiled and chatbot-ized.
The idea of Document Interrogation has many, many applications. Companies can turn their FAQ pages into chatbots; annoying kitchen electronics can turn instruction manuals into Air Fryer User Manual chatbots; Politicians’ websites can have their platforms condensed into a chatbot; the notable wits of history can have their interviews, articles, and tweets modeled into an emulated version of them. Any substantial body of knowledge can be collected and turned into a chatbot.
Information can talk back.
What’s Ahead?
Where this technology leads is unclear.
How we get knowledge is changing radically. As information is increasingly chatbot-ized, more knowledge will be more readily available every month. Right now these chatbots require prompting, but soon they will talk to you of their own volition. Information will follow you around your digital life, suggesting itself where possible.
Reading won’t disappear. Not in my opinion. There will still be value in gathering knowledge closer to the source. Text formats may evolve though, and elements originally added for human readers could be discarded to produce writing more tailored for AI consumption. Yes, the way SEO-optimization influences writing, AI-optimization may influence writing even more. A strange idea.
Many companies are already releasing products around uploading documents into chatbots. Docuchat, Filechat, and Corpora all offer document-chatbot products. At Pickaxe too, our Document Interrogation feature is now live, allowing anyone to upload a document and transform it into a chatbot.
For instance, I created a chatbot using a film grants database that recommends relevant grants based on unique film aspects.
This innovation presents countless opportunities, and we’ve just begun to explore its potential.
