Project: ChatAI

The motivation

The reason why I wanted a chat AI to create a powerpoint to the conversation is that instead of having to actively step out of the conversation to organize my thoughts or google stuff and pull in information, it would be easier if an AI automatically did it for me. It could provide that exoskeleton superpower or fighter pilots heads up display to enable the most accurate robust informed streamlined communication possible.

The development

While I can’t discuss the how yet, I can discuss the what. There have been two main developments:

  1. Outline functionality. So from any conversation the software can split up the conversation into chapters with summarized headings. This can be done with however much granularity as you’d like meaning you can have chapters with sub-chapters with sub-chapters etc. This is actually a pretty straightforward but useful feature that helps you not lose track of the conversation.

  2. Conceptual auto-correct. Very interesting stuff, basically if you say something like “Cairo is the capital of France”, it will tell you that is incorrect and Paris is the capital of France. It can also pull up supporting evidence, but since so much would likely be correct and fill up too much space, it’s better to stick with just the incorrect things. Although providing supporting evidence is part of the core motivation of creating that powerpoint for your conversation, the design challenge in terms of presenting you useful information from a torrent of information is a substantial one.

  3. Questions are automatically google searched. Basically questions are automatically hyperlinked to open a new tab and conduct a google search with that question, completely eliminating the necessity to invoke a virtual assistant. It’s always running.

  4. (Future feature) One idea that I have that can probably only be implemented by Google is this idea called reverse text search. Basically it takes the text you have now, and figures out which search term is most likely to pull up what you’ve said. So you might be describing “I want some food with tomatoes, cream, and spices”, it will reverse text search that to indian tiki masala. The reason why I’d want to do this becomes more apparent when I describe some random car part based on the way it looks or what it interacts with, and it reverse text searches to provide me the name of that part. More generally the issue this is solving is that you don’t know what you don’t know, and by doing this, it may educate you about a complete blind spot in your knowledge.

Ultimately this could become the killer feature for AR and MR technologies, I personally see this being most useful with the screen real estate provided by AR/MR. And then either laptop or smartphones, and there’s the trade off of real estate with laptops and always having it with you smartphones. Now that I think about it though, if you want to have a fully AI augmented conversation, it’d be probably really intentional, so having a laptop with you when you have that conversation is more likely. So it’d be AR/MR form factor, then laptop, and then smartphones.

Alright the details to how some of this is done is juicy and I can’t share it yet, but when I do I bet there will be people who will face palm and say of course that’s how it should be done. 😃 Until then!