Twitter has become a one stop shop for any information — from tweets from the President of the United States to memes and Tik-Toks, multiple generations have used Twitter to taken in large masses of data. Twitter has taken advantage of this by showing recommended Twitter Ads.
However, with the influence of social media on the world today, how consistent and precise are these types of ads?
Researchers at Microsoft have created 18 guidelines to analyze Human-AI interaction in interfaces. I have taken four of these guidelines in four different interaction stages and rated them out of 5 to analyze this interaction.
Initially: G1: Make Clear What The System Can Do (Rating: 2/5)
Although Twitter gives you the option to see why you received this ad, it does not give specific, consistent reasons why you get those recommendations. If it does give you anything, it uses the same two reasons: location and age. This lack of metrics and inconsistent information makes it unclear to users what Twitter is analyzing when they use this application.
During Interaction: G4: Show Contextually Relevant Information (Rating: 3/5)
Sometimes, the Twitter Ads will show me relevant information depending on what I am interested in, and what I follow. Additionally, it will also include information relevant to what is going on in the US, including presidential candidate campaigns and activist petitions. However, sometimes, it will show information from locations I am not at — such as California — months after I have visited that place. It has even done that today.
When Wrong: G11: Support Efficient Dismissal (Rating: 5/5)
Twitter allows you to choose “I don’t like this ad” after every ad and informs you that it will be used to create better ad recommendations later on. Additionally, Twitter Ads are an unobtrusive feature — it looks like a Tweet, making it easy to ignore if it doesn’t interest the user.
Over Time: G13: Encourage Granular Feedback (Rating: 3/5)
Although a Twitter survey will show as an ad every once in a while, but not often. Additionally, it is not clear what exactly the surveys are used for. However, it will give you the option to use more of your information to get more personalized ads later on, so it is the user’s choice whether or not they would like more precise ads or not.
If this is how Twitter ranks up to Microsoft’s AI interaction, what other interfaces prove to be the best (or worse) in terms of other guidelines?
Initially: Make Clear How Well The System Can Do What It Can
Do. (Siri — Rating: 1/5)
It is unclear what Siri can do when it comes to suggestions and interactions with Siri, and the suggestions usually just pop up at the bottom of the screen. It can get frustrating when this happens because it is never clear what we can ask Siri to do and what she can recommend.
During Interaction: Time Services Based On Context. (Google Home — Rating: 5/5)
Google Home can set up a ready-made routine. For instance, if you say “Good Morning,” Google can state information based on the time of the day.
When Wrong: Scope Services When in Doubt. (Google Gmail— Rating: 5/5)
When Google finishes a statement for you in an email, it will not completely autocorrect it for you. Instead, it will show the words in light grey in order to give you the option to keep typing or tab and finish your sentence with their help.
Over Time: Learn from User Behavior. (Spotify — Rating: 5/5)
Spotify has been the best AI interaction I have had in terms of recommendations. In fact, Spotify even states, “Get better recommendations the more you listen.” In my experience, Spotify has been able to learn from my ever changing music taste and adapt as necessary to give me more and more music to discover in my Discover Weekly and Daily Mixes each week.