Gaming, Voice Chat, and Transcending Toxicity

Even before COVID-19 forced us all to stay inside our homes, social interaction was becoming virtual. Old-school hangout spots like malls, bowling alleys, and movie theaters have been slowly but surely finding themselves replaced with Fortnite, Roblox, Minecraft, and RecRoom. These new platforms, once “mere” games, have rapidly taken on a greater significance, as a new generation of players discovers that online environments offer greater freedom for self-expression, easier connectivity with friends across the globe, and a wider variety of diversions to keep themselves entertained.

Unfortunately, online interaction isn’t perfect. In particular, it’s painfully clear that people socialize differently online compared to in the real world, resulting in online communities which often contain “toxic” or “disruptive” behavior which comes up far less frequently in the physical world. There are several reasons for this difference. For starters, players are more likely to act as passive bystanders, and less likely to intervene or punish negative behavior, when online. Anonymity clearly plays a role as well, as players who enjoy being anonymous also enjoy a wider range of disruptive behaviors during play. Players can also more easily self-sort online, resulting in increasingly homogenous groups, which decreases our ability to appreciate and respect diversity and may also cause outsized negative reactions when someone joining a group is perceived as not belonging.

(By the way, if you’re skeptical that this is a real problem, 75% of U.S. online gamers report having their experience shaped by harassment, and 68% of players further report severe harassment like physical threats or stalking. What’s more, about 22% of players will quit games due to toxicity, according to a 2020 report from the Anti-Defamation League.)

In reaction to these kinds of toxic experiences, many players have soured on the idea of socializing online, especially those from minorities or marginalized groups who often face the brunt of this kind of harassment — which in turn only amplifies the homogeneity of those left over, reinforcing the bad lessons gamers and others are often learning.

One ‘solution’ to this problem could be to give up — start cutting out voice, or even text chat, in order to make games more isolated experiences. This certainly would reduce harassment, but also sacrifices the enormous potential of digital worlds to be a positive thing for society. (Not to mention that this would also hurt the experience of many more players. 88% of online gamers report getting social value from that time — including 51% of gamers who have made new friends through online gaming. And that can’t happen without chat tools — Chris Priebe at Two Hat Security has noted that users participating in chat are more than 300% more likely to continue playing. Even beyond the building of friendships, games present a social landscape which is fundamentally different from that encountered in the real world — another study found that nearly 80% of women gamers felt at least partially empowered by the opportunity to interact with others in games differently from how they would in the real world — even while 75% of those gamers reported experiencing active harassment while playing.)

So what’s the alternative? Well, the alternative is that we, as an industry, need to fix online interaction, and make it accessible, inclusive, and empowering to players of all types.

How do we do this? It’s not enough to make the high points high — that 2020 ADL report shows that 95% of players of online multiplayer games still report having positive experiences when playing, even as 22% of them quit those same games due to toxicity as mentioned above. In other words, positive experiences aren’t enough if people are also subjected to toxicity — our only option is to curb the kind of harassment that, even sporadically, might cause a truly negative experience. And given that the same report identified the in-game voice chat as the primary way toxicity is experience, we’ll specifically need to focus on the voice experience — text moderation just won’t be enough.

In order to do this, we need two things — first, we need to understand why toxic behavior arises; and then we’ll need to design interventions to stop it.

Where toxicity comes from

Danluu, a technical internet blogger, constructed a small experiment to test gender differences in Overwatch, which involved playing some games while signaling a male gender identity, and others while implying himself to be female. As one might expect, Danluu found some comments which were explicitly sexual harassment — though only in about 1% of the games he played as a woman. Much more frequently, he found he was simply treated with less respect as a woman. He found others tried to tell him how to play and assumed ignorance on his part roughly three times as often (19% vs 6%) when he used a feminine username compared to a male one. (For clarity, Danluu chose to ignore voice chat for his test, as he lacked a sufficiently believable voice changer, so he “changed his gender” purely by changing his username.)

A similar paper from 2015 saw Australian researcher Michael Kasumovic analyze an earlier study which had conducted just over 150 sessions of Halo 3 to examine how men and women were treated differently. This study did use voice chat, and found some fascinating results which provide deeper insight into Danluu’s discoveries.

Men were given the same amount of praise regardless of their skill level (2.5 compliments on average), but women received less praise (1.5 complements on average) at low skill levels and more (4–5 compliments on average) at high skill levels — and especially when they were very skilled but their teammates were even more so. In contrast, women received vastly more criticism when they failed to get a large number of kills, while the criticism men faced was independent of kill count. (Similar to Danluu’s results, they found that only a small percentage of the negative comments explicitly used gendered or sexual language, though they still saw this 13% of the time compared to Danluu’s 1%, hinting that the voice channel may be where much of this harassment happened. The rest of the negative comments were simply disrespectful or derogatory of skill level — though of course, the statistics show that there was a hidden, systemic sexism underlying these comments as well.)

What this suggests is that the negative experiences many women face online aren’t as simple as “sexist jerk actively trying to be hateful” (though there are some of those, unfortunately, too.) Instead, many disruptive and toxic behaviors may be traceable to an over-aggressive sense of competition, where male players (especially those who are relatively unskilled) are desperate to protect their place on the social hierarchy, and so become more hostile to anyone who has the potential to disrupt that hierarchy. In other words, at least part of the toxicity we’re seeing comes from people who are trying to protect their social position — so if we can find a way to encourage them to do that without toxic behavior, they should be willing to follow our lead.

Whether or not this is always the case, it seems safe to say that most toxicity is contingent, only emerging in certain situations — whether that situation is “outsider threatening my social status” or simply “had a bad day.” Indeed, Riot Games found that only 1% of its players are actually consistently toxic (though these few players cause an outsized 5% of all negative experiences on the platform.)

That said, of course, there are also those more explicit comments we mentioned before. And they are hardly contained to sexism — toxic behavior is well documented between players of different races as well, and in some cases doesn’t require a demographic difference at all. Much of this harassment comes from deeply-rooted cultural mores: things like gaming’s historic framing of itself as a “boy’s club”; to the pervasive and flawed idea that only offering one — white — racial option was a neutral position; to attitudes learned in the physical world which simply seep into online behavior. The way we market and design games is deeply important for solving this problem — if we don’t start with an inclusive platform, we’ll never get anywhere. But these elements of culture are a bit deeper than this article is meant to cover — and besides, there are many better folks than me talking about designing inclusive games.

So let’s focus on our sliver of the problem here — given that most toxic behavior is contingent on outside factors, can we keep these situations from coming to pass, so folks have no reason to be toxic? And for the remaining few, those who are truly trying to be disruptive, can we decrease the chances they get to actually harass others, until they are no longer able to ruin the experience?

How to make things better

As discussed above, when players are causing toxic behavior, they are usually prompted by one of a few things:

  1. Fear of losing their place on a social hierarchy, causing protective aggression towards outliers
  2. Unintentional, or stress-provoked outbursts which weren’t actually intended, but which often escalate
  3. The rarer but uglier desire to ruin the experience of someone different from you

Starting on the first, we need to find ways to make the proverbial rising tide raise all ships, so that players are inspired to welcome new folks to their community, not to fear disruption of the status quo. To do this, we’ll need to actively motivate that kind of community-building behavior — finding approaches like rewarding support as much as “who got the most kills”, or even directly offering advantages to those who forge stronger and more positive social bonds. Imagine a game mode in which skilled players could act as coaches, helping other players grow and earning rewards for themselves in the meantime! This is already made possible by many third-party groups, but if we want games to foster true communities, it will be important for them to be able to enable this kind of support more directly.

Of course, that does have challenges — how can a game, at scale, monitor whether another player is treating you respectfully and helping you grow, or if they’re constantly harassing you? The trick will be in unpacking all of the information hidden in voice chat. Consider the difference between typing “gg” at the end of a match, versus saying it aloud. How much might we be able to learn from the tone of even those simple utterances, to understand whether a player is getting along with their teammates? Modulate has begun exploring this problem ourselves, and we’re not the only ones. Ultimately, it will become crucial for games to be able to tell whether or not you’re enjoying your experience, so that they can foster better and better ones through tools like matchmaking, social rewards, and perhaps coaching and other future features as well!

(I should note that this can and should be done carefully while respecting each player’s privacy — for instance, sentiment detection can run without needing to understand the words being said, such that the game only ever learns if someone is having a good or bad time.)

Moving onto the second point, we need to quickly mitigate bad outbursts, so that the provoked player calms down, rather than letting others get offended and causing an escalation. The answer here is real-time voice moderation. While it may be impossible to always bleep offensive terms before they can be heard, it is fairly straightforward (if technically involved) to notice the offending phrase immediately thereafter, triggering prompt punishment — which might be a temporary muting to force the speaking player to cool their head, or even potentially an in-game debuff. Once again, Modulate has taken advantage of our voice skin expertise to build a unique solution here, though voice moderation will always still require some humans in the loop — among other things, gaming vocabulary is too vast and evolving for any AI system to ever be “finished” learning about any given community. We’ll also need to all come together as an industry to set standards about the kind of dialogue that’s permitted, especially at major events like tournaments — as well as determining the flexibility of this moderation, for those playing with friends who are comfortable cutting loose.

Finally, we come to the third point — the fundamentally bad actors. I wish I could offer a clever solution which would make these folks see the error of their ways, but I’m afraid that problem is substantially harder than detecting good relationships or reacting to bad behavior. What I would argue, though, is that these individuals generally pick their targets carefully, looking for cues that give away information like the gender and race of a potential target. So what if the bad actors were denied this information altogether? Some games attempt to achieve this by forgoing voice chat entirely, but as we’ve discussed before, that has some major drawbacks. Instead, this is where Modulate’s third and longest-standing service, VoiceWear voice skins, shines. If every player chooses a voice based on their character, then how they sound is no longer giving clues about their gender, race, age, or anything else. This ensures all players can participate safely, regardless of their demographics, and leaves the malevolent few games without any cues upon which to build their harassment. It won’t completely stop harassment, of course…but we (and over 60% of our testers we’ve spoken to) see this as an obvious step which would hugely limit the damage these bad actors could do — while making the game more immersive and fun, besides!

Toxicity in gaming, sadly, may never completely go away — any more than we’ll ever be blessed to live in a world with no evil people. But cutting ourselves off from voice, and therefore from the vast majority of meaningful social interactions online, isn’t the answer. Instead, we should strive to create an environment which fosters and encourages the support of our fellow players, and provides tools to those who wish to go the extra mile to preserve their safety. Some game studios have already proven that these kinds of interventions really work in other contexts — so why shouldn’t we apply the same logic to voice chat, as well?

Gaming has the potential to become every bit as rich a social experience as any other. It’s our job to take it there.

CEO and cofounder at, building crazy cool machine learning tools to make online voice chat more immersive and inclusive.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store