I wrote a couple of posts early in June talking about how understanding the way certain search tools functioned could explain away their offensive or merely bizarre results. A few days after those posts, I was put to the test by my own piece of technology.
It happened when I was (for reasons I cannot recall in hindsight) typing ‘businesswoman’ into my phone. Readers should know that I don't own one of those mini-computers the kids these days call ‘phones’. Rather, I have a flip phone, and a fairly advanced one at that, but it’s still limited in terms of texting by my having just ten numbered keys for typing. This means words like ‘room’ and ‘soon’ are spelled using the same sequence of keys on my phone. As I type, my phone will try to guess what I’m typing. Anyone who has used a search engine will know that this autofill feature can lead to some unexpected results.
Anyway, back to business, or to ‘businesswoman’ – as I was typing the word, I noticed that the phone filled in ‘businessman’. I laughed a little bit when I first saw this, almost triumphantly, because I knew that I had some easy material for TOA. When I got around to starting this post, I remembered my own words from those aforementioned pieces about the importance of understanding how search engines work. I decided to confirm how my phone was actually functioning, step-by-step, before I jumped to any conclusions.
Here are the results of my experiment – below is what I typed into my phone followed in parentheses by what my phone filled in:
b (a)
bu (at)
bus (bus)
busi (bush)
No surprises so far from my perspective. These first four entries are a good example of the phone’s autofill function in action – it looks at the letters and provides a rank order of the most commonly used words.
busin (curio)
busine (crime/buried)
busines (arrives)
This is an illuminating block. It starts innocently enough – I guess ‘curious’ is more commonly used than ‘business’, though I didn’t care enough to find out for myself. The next letter (d, e, or f) told the phone ‘curious’ likely wasn’t my target so it went off in search of a new word. It gave me two options (I ran multiple experiments): ‘crime’ and ‘buried’.
This was an interesting outcome for two reasons. First, neither word fits based on my spelling. In other words, in order for the phone to suggest these words, it assumed a misspelling and tried to correct me. This is a problematic point that we’ll talk about more in a moment. The second reason is that by giving me different options seemingly at random, it dispels the notion that the phone is using a technique where it knows the most likely final word for a given combination of keys. Instead, for certain cases it appears that the phone is choosing at random from two or more plausible choices. If it tracks my choices and updates its internal set of probabilities to give me my most common choice over time, it would qualify as an elementary example of machine learning! My flip phone is more advanced than it appears to the naked eye.
business (business)
businessw (choppy)
businesswo (businessman)
businesswom (businessman)
This is the key sequence. It appears initially that we are back in business (!) but my phone loses the plot immediately when it spits out ‘choppy’ after I append a ‘w’ to ‘business’. Yikes.
From here, it’s like watching a dumpster fire roll slowly down a hill. First, I add ‘o’ and feel my eyes roll right out of my head as the phone makes its best guess. I follow up with the ‘m’ in a vain attempt to save my text but it seems nothing can divert the phone from the erroneous precedent set by its ‘choppy’ autofill choice.
At this point, we've clearly established the reason for my post. Let’s wrap it up first, and then consider some final thoughts.
businesswoma (businesswoman)
businesswoman (businesswoman)
Eventually, the phone gets the message and fills in the right word. However, it required some extra effort – I needed to type in all but the final ‘n’ to get the autofill on the right track. I’m sure there is a plausible technical explanation here but it wouldn’t be much of an excuse. Is it really possible that someone who types in ‘businesswom’ meant to type out ‘businessman’? Was this a common spelling mistake in those glory days of longhand? The saddest thing about all this is that as I continue to use my phone, it’s very likely I will discover additional applications of yesterday’s logic.
I’ve often joked that one day the science museum will confiscate my ancient flip phone and place it in a display case. I’ll certainly make that joke again, but for different reasons. When people laugh, I’ll have to consider an important question – whose job is it to correct a harmless assumption? My phone has clearly taken up a position just a couple steps to one side of this fine line. Maybe the more relevant question is, if I remain quiet, do I symbolically join my phone on its side of the line? I suppose for what it’s worth the logic of autofill would certainly say so – the words it places in my mouth are, if left uncorrected, the words that become my own.
