Google launches “Perspective” to ward off trolls
Google has debuted Perspective, an artificial intelligence (AI) tool designed to moderate online comments. The tool is crafted with the intention of helping communities do more than just filter out curse words.
As the tech world fawns over the advancement, let us pause to talk about a substantial flaw in using AI to censor human behavior…
Language is a wonderful and beautiful thing
It is something so ambiguous, so contradicting, so clearly defined by its many grey areas; it is a perfectly human concept. Its innate human-ness has been further enforced with the introduction of Google’s “Perspective” API. Unfortunate, then, that the creation of Perspective and the reasons for its flaws are due to such negative circumstances.
Sticks and stones
Fact: Technology is amazing. Artificial Intelligence is making progress at a substantial rate, robots are ever-increasing in efficiency, and the Internet is leading us to a more connected world.
Alternative Fact: Whatever people can do, technology can do better.
Reasoning: Nothing can beat human creativity when it comes to finding ways to hurt and demean one another.
There is no questioning that the online environment has a significant “troll” population.
For those that may be unaware, I’m not talking about the kind of trolls that live under bridges attempting to eat unsuspecting billy goats.
Like Sauron’s army in the Lord of the Rings, they march forward, leaving a swath of destruction in the form of poorly worded hate comments in their digital wake. In the Lord of the Rings those were Orcs, of course, but their use of the English language was pretty similar, hence the analogy.
Often, a troll’s verbal vomit is not to express their true opinions, nor is to stir up an interesting ideological debate. Instead, it is merely to stir up chaos, just because they can. Sure, trolls can serve a purpose by pushing more people toward the center as they rant on the fringe, but most people just see their behavior as abusive.
In response to the often-volatile nature of the comment section, Alphabet, Google’s parent company, created its Perspective API. Using machine-learning, the service had been created to weed out the toxic language that so often turns a section created to express opinions and create open-ended debates into a War of Words. The concept is simple: using an algorithm, Perspective searches for pre-programmed words and phrases.
These comments and phrases are then given a “toxicity” ranking determined by the program.
Perspective is a great idea, created with what are presumably good intentions. However, there are a couple of major issues with the service.
Robots still don’t have feelings
The biggest, and most notable problem is that while it can detect toxic words and phrases, there is no way for it determine context, and even more so, sentiment.
Upon reading David Auerbach’s article for MIT Technology Review, I checked out the Perspective website for myself which offers a free test of the service. Drawing inspiration from Auerbach, I attempted several different phrases to determine their toxicity ranking.
Overall, the service could detect clearly mean or hateful phrases fairly well.
However, just by spending a few moments with it I was able to detect key words it would pick up, and by omitting or substituting said words, comments that would still be considered toxic by human eyes received a comparatively minuscule toxicity ranking in the system. A prime example of this would be the use of swear words, which very much differ in meaning depending on context.
Six degrees of censorship
Another, more potentially troubling issue, is the fact that Perspective was created and is maintained by human programmers. As such, what it deems toxic will generally tend to fall in line with the programmers’ own ideological views.
Though it may be unintentional, one would naturally input words or phrases that they, personally, would deem offensive, while potentially omitting phrases they may be unaware that others would find offensive. Even more so, however, is the fact that Perspective would be marketed to businesses. [clickToTweet tweet=”Google Perspective: companies could weed out comments that do not fall within their beliefs” quote=”In effect, companies would have the potential to be able to weed out comments that did not fall within the range their own ideological beliefs.”]
This would be especially harmful in relation to entities such as news sites and forums, which should serve as platforms for users to express and debate their opinions.
Let me restate my belief that Perspective is a great idea created with genuinely good intentions. However, it will be some time before it is, if ever, one-hundred percent effective. And, like all powerful things, it has the potential to be dangerous if used in the wrong hands by creating echo chambers and ultimately, by eroding free speech.