If we need to talk about precision, let’s be precise
We measure the world, you already know that, thanks not only to our daily work in retail, events, or smart cities but also to other spectacular projects. We gradually unveil how our Artificial Intelligence platform works. You see, at Goli Neuromarketing, we’ve been delving into this AI that has now become so popular for years.
One of the common questions we face is how we know that our Golineuro AI platform measures reality and not other things. Well, it’s because it’s calibrated periodically.
Every so often, we subject our platform to a series of known situations in a controlled environment and compare the results to the actual situation.
For example, to verify that eye-tracking works correctly, a group of voluntary subjects looks at marked points on a ruler. Then, we extract information from the platform about where they looked, calculating if those points match or, if there’s any, what the maximum error is.
In the case of emotion recognition, the control group of users wears devices that measure their brain activity, heart rate, and skin conductance, using scientific literature (publications or scientific studies) to identify emotions based on the parameters these instruments provide. This way, we compare the emotion indicated by the platform with what is deduced from their biometric activity.
For the recognition of behaviors and actions, users are asked to randomly choose a behavior from a list and perform it, comparing the chosen behavior with what the platform indicates.
This way, we not only keep our platform always up to date and ensure that the data is as objective and reliable as possible but also progressively reduce the margins of error that may occur.
After the last calibration, the maximum error margins are as follows:
– Maximum eye-tracking error: 0.6 mm
– Maximum counting error: 0.8%
– Maximum age estimation error: 15.4%
– Maximum gender estimation error: 11.6%
– Maximum emotional identification error: 1.2%
– Maximum flow direction error: 0.01%
– Maximum behavior detection error: 1.4%
