Behavior I/O: how observations drive better behavior
There are a few truisms of life: the sky is blue, water is wet, and algorithms get smarter as they receive more human feedback.
Let’s take a high-level look at LinkedIn’s ‘People you may know’ feature. When you confirm a match, you teach LinkedIn what a good match looks like, and LinkedIn consequently gets better at suggesting them. In other words, you (the human) are ‘training’ the machine (LinkedIn) through a feedback mechanism that can be coined as Behavior I/O.
Wait, what is I/O?
In computing, input/output — or I/O — is used to describe the communication between an information processing system (think: computer) and the outside world (think: a human or another information processing system).
Where ‘behavior’ comes into play
Trying to hit 10,000 steps today? Companies like Fitbit learn from the behaviors of people to help train other people to behave better. In the world of Fitbit, that behavior is increasing physical activity to improve your health.
Similarly, in the data world, we can use machine learning to observe the behavior of analysts and leverage those observations as Behavior I/O to allow other data consumers to use data in a better way (based on how analysts engage with their data assets).
How do observations of analyst behavior actually provide value to other users?
Benefits of the human-machine learning system inherent in
a Behavior I/O feedback loop:
One of the overarching problems in data analysis is incorrect use of data. Data teams all over the world are united in their universal propensity to use the wrong report for analysis, even when a new correct report is available (and has been available for weeks).What’s more is that this isn’t limited to reports. Filters, data transformations, calculations — just about every data asset that can be used in a report is subject to the same fate.
But by observing user behavior, a data catalog with Behavior I/O might alert a user that they’re working with a stale data asset. This alert could contain a notice that a more accurate data asset is available and could provide a direct link to the right asset, effectively correcting that user’s behavior.
Detailed data context
So you’re using a Tableau workbook that everyone else on your team uses — how do you know that the underlying data is correct? Even if you know that 100 people ran the same query in the past 2 weeks, do you know who used it and, more importantly, why?
A data catalog with Behavior I/O observes user actions to track the history of how all the assets linked to a workbook ended up being used — from the workbook to the underlying SQL query, back through every database and Hadoop system that touched that data to the very source where the data was initially stored. The end game? Arriving at better conclusions — since you can clearly see the full data lineage and how others previously used that data.
With this human-tech synergy, Behavior I/O is able to provide ‘just-in-time’ suggestions when needed based on empirical behavior of other users. This means getting a real time suggestion for the most relevant table to use as you write a query or getting a warning notification as soon as a user deprecates a data asset that you’ve used in a report.
And that’s about all the ink I can put on Medium before I lose you all.
For a deeper dive of the good stuff behind behavior I/O, watch Behavior I/O: How AI and human brains mix for big data insight.