This is in the realm of AI but there is a way to do it in Airtable - it’s not pretty and it has a lot of moving parts. My approach …
- We have the ability to run scripts based on events (like adding a new record to a table or view). This can fire off a process that reads related (historical) data and sends it to a webhook service or at least pointers to the historical data.
- The webhook service must be constructed by you to listen for these calls and the associated pointers to the historical data or the data itself. Free services to build such a webhook include autocode.com and Google Apps Script.
- The webhook service is then able to submit the historical data to another platform called OpenAI. You will have to join the beta program to use this, but they are pretty much allowing access to anyone and relatively quickly once you request access.
- While a lot of buzz centers around OpenAI Codex, what you are really interested in using the GPT-3 platform which is designed for auto-completion. This is a platform designed to be easily trained and then respond when given a context. You can “program” GPT-3 by showing it just a few examples or “prompts”, aka - field-level data values. They’ve designed the API to be both simple for anyone to use but also flexible enough to make machine learning teams more productive.
- Each call into the webhook service may be used to re-train GPT-3 with the latest historical data and from this derive the most appropriate field entry recommendations. Over time, this approach will become smarter because it is constantly being given new historical data. This is how machine-learning works.
- The webhook service would then respond to the Airtable script action and pre-populate the desired fields in the new record.
Like I said - it’s not a cakewalk, but it is quite impressive.
I have used this approach to create some amazing solutions. In one case, a system literally writes a text report narrative based on a bunch of key performance indicators. You can literally shape the output of the text in any way desired.
This is the underlying AI system being used to develop GitHub’s new CoPilot code-writing platform which will likely revolutionize how we write code in the future.
Side-bar: Airtable engineers should add integration capabilities into all of its code editors to allow us to make coding far simpler.
