The task this week was to create a microblog entry using the predictive text function. The phrase I choose to begin with was “this is not my idea of”. Here is my entry:
This is not my idea of what I have done for this week and I’m just thinking about it ha ha.
For interest, here are the predictive choices I was given through the process, with the selected words in bold:
This is not my idea of…..
- a what the
- you I the
- gotta want have
- to done a
- for at checkmark icon
- the this my
- year week app
- and I so
- I I’m then
- not also just
- thinking leaving not
- about about you thinking icon
- you it that
- and haha all
I had a number of observations from this exercise. One is that each of the three word choices I was presented with at each step were valid. While that seems a relatively small feat, I find it rather impressive from a relatively simply algorithm. Any of the choices would have made sense grammatically, and contextually at each stage. However, that was not the case for the overall sentence. Despite my best efforts to select words that would lead to something coherent and meaningful, I ended up with a statement that says nothing in particular and is borderline non-sensical. The overall conclusion is that the algorithm functions at a more detailed level, but not overall. The trees are there but the forest is weird.
I also noticed that the choices were more or less conjunctions, however, at a certain point, in this case at line 11., it knew to finally offer a verb. In this regard, the verb choices, however, where nothing that I was expecting or would have thought of for myself. Given the numerous verbs we have, there is a high probability that the three offered would not reflect what was on my mind.
Another observation occurred between lines 12 and 13. In line 12, I had a choice of simply the word “about” or the phrase “about you”. When I selected the single word, about, the following choice included “you”. Reflecting on the software’s seeming desire for me to put these words together, I recognized that I send a number of texts to people just to let them know that I am thinking about them. My guess is that the software has recognized my “thinking about you” message that is frequently sent in some form.
Before posting this, I looked at two of my classmates’ results and tried to reproduce them. In both cases, I was offered different choices early on. I became curious if I would be able to reproduce my statement. I found that the choices were identical the second time around, however, I placed a period after the 7th choice which gave me a whole different set of choices, specifically, I had a different choice of verbs. The second attempt ended up as:
This is not my idea of what I have done for this week. I just talked with her about it yesterday and she said that she would have it done for this afternoon but maybe she could come and pick it out if she wants.
Although a longer statement by design, I thought this second comment had a similar ring to the first. My immediate reaction to the first comment was that this was nothing that I would have put together on my own given the lack of logic or utility of the statement. However, given the second attempt produced something familiar to the first, I gave thought honestly about whether this was my voice or not. Disregarding the actual content of the message, I concede that this is reasonably close to how I would speak, or at least text. I rarely, if ever, text formally, to co-workers, or students, etc. My texts are always to close family or friends and would be casual in tone and language. This is reflected in the predictive choices I was given.
Finally, I thought I would search predictive text online, just to see what might be out there on the topic. My top search suggestions were about how to turn off predictive text. I had to turn on the feature for this task and my best guess is that at some point, I had searched how to turn it off. This search algorithm seems to have pegged me well.
One question for this week was what the implications of algorithms in public writing spaces might be. I cannot recall ever seeing comments like these and I have a hard time believing anybody would use predictive text to select words for them that they were not otherwise intending to use. If we co-opt communication to AI and algorithms, if even just to fill in a few blanks, then we have really lost the entire point of communication.