December 8, 2021 • 3 min read
I haven't been this excited (about work stuff) in a while.
When founding Aito.ai, we defined a benchmark for how simple to use, yet highly effective machine learning should be. We looked at an Excel sheet and said that the ML should work right there - "inside the cells" - automatically.
The diagram below shows a perfect example of the simplicity we were aiming for. The spreadsheet contains training materials in the Category column, so wouldn't it be natural that you could "just know" what the next value would be? And the next, and the next … you get the point.
This Excel idea went on the back-burner for a while. We were busy implementing the Aito Core to be able to do real-time machine learning from tabular datasets. We built Aito Console, where users can manage Aito cloud instances, upload data and evaluate prediction performance. We also built our Python SDK for SW developers. But it wasn't just that we needed to build other stuff first. The Excel idea was great in itself, but also quite difficult to achieve.
Excel columns don't really have data types, so there are countless things that can go wrong. Just start from the fact that a number can be an identifier (categorical) or amount of something (continuous). This may not be a big thing for humans, as we can easily spot the difference, but it's critical for machine learning. Also, Excel isn't really known for its open API to build extensions.
Before too long, we noticed that Google Sheets added similar features. I admit, we were slightly worried that Google would solve the same things and render us obsolete! For instance, the
FORECAST function allowed you to predict a number using linear regression …
About a year ago, Airtable launched its Apps, Marketplace and Automations. For me, it was love at first sight! Finally, here was a platform that truly performed like a database in how, for example, each column has a predetermined type (or schema, as my inner DB developer from 20 years ago wants to say). And Airtable literally invited the community to build on it! So, what did we do next?
We set out to build an Airtable app that would make machine learning accessible to every no-coder.
Our first prototype was built in the second Covid summer of 2021, and the app was been refined for launch during spring and early winter. It's now ready and available, and we couldn't be happier now that we're able to show it to the world!
Let's dig a bit deeper into what we built, starting with the three main features of our app.
First, it's so easy to maintain the training data set in Aito by choosing which data from your Airtable base is used. Once you pick the right view, Aito shows if all your columns are supported, and then it's just a click to upload. Whenever you want to, just repeat the same and re-upload data. (We're thinking about continuous sync next, but want to see your use cases first and get feedback.)
Second, just select a cell in a table and Aito predicts the value. Of course, Aito is most suitable for categorical data (typically in columns like Single line text, Checkbox, Multiple select, Single select, Collaborator, Email, Rating and so on). However, we chose to show whatever results Aito brings for any column. It's then up to you to let Aito either autofill the cell value or do it manually.
Third, hover over the question mark of any result for Aito to reveal the main contributory factors for the prediction. Both are the ones that have the biggest positive or negative impact. Explainable AI without a hassle! How cool is this?
While we have so many ideas for what to do next, what we're really looking forward to is your feedback and use cases! So get the Aito Instant Predictions app for your Airtable base right now from the Marketplace, and show us what you can do with it!Back to blog list