[The author makes the common mistake of believing that Deep Learning Is Magic (flavor of the day: regret-based learning is magic), but it’s still a very interesting post. -egg]
Swarms of real life, human trolls have already been able to achieve some remarkable things.
For example, there’s the well-known incident where Time’s Man Of The Year Poll met 4chan. Twice.
But real-life trolls have to sleep. They have to eat. Whilst it might not look like it, they get tired, and angry, and dispirited.
Chatbots don’t.
And the only limit to the number of trollbots you can control is the amount of processing power they require. That might initially look like a pretty major limiter, given that machine-learning applications tend to require at least a single graphics card of some power each. But a), thanks to cloud computing that’s actually pretty affordable – an Amazon GPU instance on Spot Pricing will cost you $0.13 an hour or a little over 2 dollars a day – and b) there’s no reason that one instance of the trollbot software can’t control hundreds of social media accounts all posting frantically.
So what does this mean?