In recent years, artificial intelligence (AI) has grown to infiltrate into our daily lives. What I find particularly interesting about AI is its extensive use in writing, whether it be to produce new content or to improve on texts people write. Grammarly is one of the most popular writing tools used to check grammatical mistakes. Using AI, it detects even the most trivial mistakes that people often overlook and provides helpful writing suggestions.
With the ever-growing influence of AI, interface designers have become more attentive to the legitimate ethical concerns regarding human-AI interaction; Microsoft developed a set of Guidelines for Human-AI Interaction. The guidelines are grouped in four different stages of human-AI interaction: “Initially,” “During interaction,” “When wrong,” and “Over time.” I will assess Grammarly’s features according to Microsoft’s guidelines and score them on a scale of 1 to 5, with 1 being “Clearly Violated” and 5 being “Clearly Applied.”
INITIALLY
G1 — Make clear what the system can do
Score: 5/5
Grammarly clearly presents itself as a digital writing tool and grammar checker. It goes over the user’s text to detect grammatical mistakes ranging from misplaced words to punctuation errors in compound and complex sentences.
DURING INTERACTION
G4 — Show contextually relevant information
Score: 5/5
When a user starts a new document in Grammarly, a pop up called “Set goals” appears asking for more information on the text that the user is about to write. The user can then input specific information about the text’s intended audience, level of formality, domain, tone, and intent. This helps Grammarly provide more tailored, contextually relevant writing suggestions.
WHEN WRONG
G8 — Support efficient dismissal
Score: 3.5/5
Grammarly can detect and flag a variety of mistakes in a user’s text; however, it also makes it extremely easy for a user to dismiss its suggestions. Removing Grammarly’s critical issue error alerts is as simple as clicking on the trash icon which always appears with the alerts.
However, removing the “premium alerts” which detect “advanced issues” does not seem to be possible. The user can only view the details of those alerts if they have the paid Grammarly Premium service and cannot easily remove and dismiss the alerts.
OVER TIME
G15 — Encourage granular feedback
Score: 5/5
Grammarly also contains a feature called “Personal Dictionary” under the tab “Customize.” When Grammarly detects words that it is unfamiliar with (on very rare occasions), it suggests the user adds the word to their “personal dictionary,” which can be easily done by clicking on the book icon with the plus sign that appears with the error alert. Adding the word to the user’s personal dictionary prevents Grammarly from flagging the words as misspellings in future incidences.
Assessing other interfaces
I will assess other interfaces according to Microsoft’s Guidelines.
Google Translate
INITIALLY
G2 — Make clear how well the system can do what it can do
Score: 1/5
When one writes a nonsense word like “pdfjdnfeyronakdfmakjlmdf,” Google Translate detects it as English and displays the same word in the language that the user wants to translate the word into; in this case, I have tried it with Spanish. While Google Translate asks the user to clarify what the user meant by asking “Did you mean: pdfjdnfeyronakdfmakjlmdf hdfh” when the phrase they try to translate is “pdfjdnfeyronakdfmakjlmdf iohsodfh,” it does not ask the same question when the user only wants to translate “pdfjdnfeyronakdfmakjlmdf.” As such, I believe it is not clear as to which guidelines the tool uses to determine when to ask the user to clarify their input or not.
I think that having an alert message such as “Google Translate could not translate the text properly” would be helpful for users, especially when they want to translate a text in a language that they are not familiar with. For example, in the following image, I typed “I’ve written something that makes no sense” in Korean with a couple of syllables (that also do not make any sense) before and after the sentence. Surprisingly, Google Translate translates the phrase into “I miss you.” While this might be because I did not space out the Korean words on purpose, to people who do not speak Korean, it seems as if Google Translate has translated the phrase seamlessly.
Google Autodraw
DURING INTERACTION
G3 — Time services based on content
Score: 5/5
In Google’s Autodraw, the user can choose when to draw freely or autodraw. When the user selects the “autodraw” feature, as soon as the user draws something, there are suggested drawings that appear on the top as “Do you mean.” When the user selects the “draw” feature, the suggestions are gone, and the user can focus on drawing freely.
Reflectly
WHEN WRONG
G10 — Scope services when in doubt
Score: 5/5
Reflectly is a journal/diary app that enables the user to write and upload photos of their day. The app uses AI technology and offers information such as quotes relevant to the user. For instance, when the user first starts a story, Reflectly asks about how they felt throughout the day provides a scale that does not contain specific numbers (as users may find it hard to “score” their day. It also allows the user to choose what made them feel that way and gives a variety of options.
NAVER Spelling Checker
OVER TIME
G12 — Remember recent interactions
Score: 1/5
NAVER spelling checker is extremely simple to use and is often accurate with detecting mistakes in writing, ranging from spelling errors to misuse of words. However, it does not maintain any short term memory of its interaction with the user and thus does not personalize the user’s experience.
