Tomorrow, I’m meeting with a Google VP who’s working on Humane Tech issues - user exploitation, privacy, etc.
I’d like to present him with an estimate of the total of all users’ extra time wasted by YouTube each day due to recommendation algorithms. In other words, if we know that users spend, say, 100 million hours a day on YouTube, can we put a rough guess on how many of those hours are due to recommendation algorithms?
If I can do this, I can multiply those hours by some $/hour to give an estimate of the social cost of YouTube recommendations just for time wasted (this would not include the social cost of opinions changed by extreme or “fake news” videos).
Lots of analyses for governments do this - for example, when there is a traffic light that wastes commuters’ time, a traffic engineer will estimate the amount of time wasted, multiply it by some $/hour (cost of commuters’ time), to calculate the social cost of that faulty traffic light. The traffic engineer then uses this to calculate whether it’s worth fixing the intersection + traffic light.
Of course, it’s highly subjective which recommended YouTube viewings are “wasted time” and which are not. I would even be OK saying, “Youtube users spend XX million hours a day on recommended videos, and if 50% of these are wasted time, then the social cost of YouTube’s recommendation system, just accounting for wasted time, is $XX million/day.”
Does anyone have any ideas on this?
I’m getting help on this from Guillaume Chaslot, ex-YouTube recommendation system developer and current YouTube critic. He’s helping with the number of hours/day wasted. Before more fine-tuning, our very rough, preliminary estimate is that about 175 million hours a day are wasted.
So, I still need to estimate the social cost per hour of time wasted on YouTube. Traffic impact analyses for governments do this - for example, when there is a traffic light that wastes commuters’ time, a traffic engineer will estimate the amount of time wasted, multiply it by some $/hour (cost of commuters’ time), to calculate the social cost of that faulty traffic light.
It’d be easiest to estimate one $/hour cost, but the audience is quite diverse - kids + adults, poor + rich, US + other countries. Even a low rate of $2/hour would imply that the social cost of YouTube recommendations is $350 million/day.
Does anyone have any ideas on the social cost/hour of time wasted by YouTube recommendations?
Thanks for the post. But I don’t think YouTube would think much about the social cost wasted watching videos. The more engagement on their site the better for their business. A new paper from Google Youtube parent company propose an update of Youtube algorithm to make it more addictive to users. https://www.technologyreview.com/s/614432/youtube-algorithm-gets-more-addictive/
Yep. I think the big challenge here is one person’s soul-sucking rabbit hole on Twitch is another person’s diversionary indulgence to escape from the horrors of daily life for a respite. Our standards are different and product builders can’t be all things to all people. So the challenge is a level deeper than that.
This makes YouTube more like a bag of potato chips. For some once in a while could be an Ok thing. For others it’s a life sentence to obesity, heart disease, and early death.