This is the regular weekly post where I share updates about things happening in gaming, machine learning, audio, or frankly anything that catches my eye and seems relevant to share or discuss here. Let’s hop right in!
Generally speaking, when people talk about “voice tech”, they are talking about voice-as-a-user-interface-to-control-a-computer (think Alexa, Cortana, Siri, etc) rather than voice-as-a-channel-for-activity-or-self-expression-on-the-platform (think of the growing use of voice chat in gaming and social apps). That said, I still like to keep an eye on roundups of the “voice tech” space, as the more comfort people gain using voice to interact with machines, the more it will feel odd when they’re unable to do so with other users as well.
Someone made a tool called “Genderify” which purported to use machine learning to detect someone’s gender from their username alone, and which offered such terrible assessments as the username “scientist” being almost certainly male. It’s already been taken down due to backlash, but even beyond the obvious badness of this specific application, I think it’s important to address a key underlying misconception. Machine learning is not magic, and it cannot find information that isn’t there. There was no reason to assume that usernames should encode gendered information – and in fact it would be trivial to choose a username which suggested a different, or no, gender – so it was always guaranteed that machine learning would only be able to pick up on “patterns” which happen to be artifacts of whatever specific training data was used, rather than any real patterns of different behavior in the real world.
One of the best-known text-to-speech startups for a while was Lyrebird, which was acquired by Descript around a year ago now. Descript just launched Overdub, their new audio-replacement product which leverages Lyrebird’s underlying technology. They’ve done a lot of great work to productize what was previously more of a research tool, but it remains to be seen whether text-to-speech is at a level where this tool will become widely used. Worth keeping an eye on!
Also in text-to-speech news, a developer named Brandon Thomas built an app called Vocodes, which allows users to create text-to-speech content in the voices of a wide variety of recognizable celebrities. The underlying technology is impressive – still significantly imperfect, of course, but frequently offering plausible-sounding snippets – but there are a lot of (in my opinion justified) concerns around Thomas’s handling of the ethical challenges here. While he recognizes the risks of fraud or imitation and concedes that new legal work might need to be done, the Vocodes platform does nothing to restrict use or prevent imitation today, which, despite the imperfect quality, already poses a risk we should all be thinking deeply about. Any pioneer of new technology will find that the law has failed to perfectly predict the new tool – that’s why it’s up to us as the builders of that new innovation to go beyond just what’s legal, and really focus on setting an ethical standard that the law can eventually codify.
Riot began to work with, then promptly dropped, a Saudi Arabian sponsor this week, after backlash related to apparent injustices ties to Saudi Arabia and the sponsor (NEOM, a planned new smart-city which will likely be forcibly displacing those from already underprivileged populations) emerged. While it certainly seems everyone would have been best off had Riot simply skipped the sponsorship offer entirely, it’s at least interesting to see how quickly they’ve taken action in response to these concerns. It will be important for Riot to quickly bake a deeper understanding of what their community values into these kind of high-level decisions – but while they are iterating on that, it’s good to see that they’re at least willing to admit mistakes and react quickly.
Nintendo has always been a bit of the odd one out in the gaming space, treating its console design process, library of games, and distribution philosophy very differently from other platform builders like Sony, Microsoft, or Valve. This recent piece by Matthew Ball does a great job explaining these differences as coming out of Nintendo’s mindset of prioritizing excellence over virality or revenue, as well as highlighting some interesting challenges they’ll face should they continue to hold onto that mindset in the coming years.
Finally, Tim Sweeney talks about the economics of the ‘metaverse’. The key point, harped on repeatedly throughout the conversation, is that content will need to come from all kinds of places, so he sees it as absolutely crucial that platform fees are kept to a minimum to encourage as much novel content to come together as possible. I certainly agree with this at first blush, though I wonder about all the new economic opportunities the metaverse will unlock – perhaps those who build the platform will discover new ways to monetize that position which are less exclusionary than rent-seeking distribution fees, such that cutting those fees isn’t such a sacrifice after all. (For instance, as an economy grows, managing the economy will be a valuable and important service – and would also be something the core platform developers would be well positioned to do, and justified in taking fees on. Some companies even do this today, but at metaverse-economy scales, the revenues they see from this might easily eclipse the money they could earn through gatekeeping access to new content.)
That’s all for this week! As always, any thoughts or feedback are welcome. Stay safe, healthy, and sane, all!