Google had made interesting and visible changes to the most loved page on the planet, Google Search page. One of the interesting changes is how the Search Box shows a list of search items that we can use instantly like “pizza near me” etc.
Well, from the list that was presented to me I was not interested in using any of those terms. I’m sure this will improve search experience and engages AI very early in the search process. In my opinion at first this may be annoying because it is a distraction from what I actually want to search.
This approach may be useful in the long run and defiantly any previous search history will be used to come up with accurate or matching predictions of search terms that we most likely perform.
To set the context, I have not logged in to my Google or Gmail account but am using Chrome as my web browser. I wonder weather the behavior would be the same or I would be presented with different predictions when logged in. I believe being logged in, Google knows more about me than myself anyway! 😉
In my previous article titled “When AI has a Person-al Problem” I had discussed about the AI design architecture with limitations and one of the limitations is, it’s not designed to allow capture of feedback which is referred to as feedback-loop (FBL). I strongly believe that FBL is a crucial step through which AI becomes smarter and keeps Machine Learning Models relevant for accurate predictions.
Integrating FBL in AI design is critical ML Model survival and one of the main reasons it’s important to have a well-defined architecture for any Data Science/Machine Learning/Artificial Intelligence projects. Clearly FBL should be one of the core pieces of that architecture.
By looking at the changes to the Google Search Page, it is very clear that the team behind the AI predictions had decided to make the FBL visible and easily accessible. Search users are now empowered to provide feedback and, in my opinion, tremendously help the Machine Learning Model behind the Google Search page.
With the feedback loop option, Machine Learning Model is now actively listening!
Not all Machine Learning Models are designed to accept direct feedback from the user or service, but as an AI Architect it is important to consider FBL (“feedback loop”) in the designing stages of the overall AI architecture.
Out of curiosity I had made a few attempts to test the FBL features and used the “Report inappropriate predictions” option, which is clearly visible by the way, to send the feedback. Google Search FBL requires two critical pieces of information, which predications are inaccurate and why they are inaccurate.
I ran a few more tests to see if my feedback is really accepted or considered for future predications. In my observation I did see some changes in the search items and in some cases with a few variations of the same search items, which is a good sign.
Well, Machine Learning Models need some time to digest the feedback, improve and customize future predictions. If you explore further, on the Google feedback feature there is also a link to a legal page in which Google highlights the importance of feedback from a legal perspective.
This clearly demonstrates the importance of the FBL in designing Machine Learning Models and it is an important consideration in overall AI Architecture.
FBL is the key to make users/services to trust the AI and make the AI Smarter.