Twitter recently took drastic action as part of an effort to slow the spread of misinformation through its platform, shutting down more than two million automated accounts, or bots.
But Twitter shuttered only the most egregious, and obvious, offenders. You can expect the tricksters to up their game when it comes to disguising fake users as real ones.
It’s important not to be swayed by fake accounts or waste your time arguing with them, and identifying bots in a Twitter thread has become a strange version of the Turing test. Accusing posters of being bots has even become an oddly satisfying way to insult their intelligence.
Advances in machine learning hint at how bots could become more humanlike. IBM researchers recently demonstrated a system capable of conjuring up a reasonably coherent argument by mining text. And Google’s Duplex software also shows how AI systems can learn to mimic the nuances of human conversation.
But technology might also provide a solution. In 2015 the Defense Advanced Research Projects Agency ran a contest on Twitter bot detection. Participants trained their systems to identify fake accounts using five key data points. The resulting systems are far from perfect (the best worked about 40 percent of the time), but the efforts reveal how best to spot a bot on Twitter. We may come to rely on these signals much more.
- User profile
The most common way to tell if an account is fake is to check out the profile. The most rudimentary bots lack a photo, a link, or any bio. More sophisticated ones might use a photo stolen from the web, or an automatically generated account name.
- Tweet syntax
Using human language is still incredibly hard for machines. A bot’s tweets may reveal its algorithmic logic: they may be formulaic or repetitive, or use responses common in chatbot programs. Missing an obvious joke and rapidly changing the subject are other telltale traits (unfortunately, they are also quite common among human Twitter users).
- Tweet semantics
Bots are usually created with a particular end in mind, so they may be overly obsessed with a particular topic, perhaps reposting the same link again and again or tweeting about little else.
- Temporal behavior
Looking at tweets over time can also be revealing. If an account tweets at an impossible rate, at unlikely times, or even too regularly, that can be a good sign that it’s fake. Researchers also found that fake accounts often betray an inconsistent attitude toward topics over time.
- Network features
Network dynamics aren’t visible to most users, but they can reveal a lot about an account. Bots may follow only a few accounts or be followed by many other bots. The tone of a bot’s tweets may also be incongruous with those of its connections, suggesting a lack of any real social interaction.
A horrifying new AI app swaps women into porn videos with a click
Deepfake researchers have long feared the day this would arrive.
The therapists using AI to make therapy better
Researchers are learning more about how therapy works by examining the language therapists use with clients. It could lead to more people getting better, and staying better.
DeepMind says its new language model can beat others 25 times its size
RETRO uses an external memory to look up passages of text on the fly, avoiding some of the costs of training a vast neural network
2021 was the year of monster AI models
GPT-3, OpenAI’s program to mimic human language, kicked off a new trend in artificial intelligence for bigger and bigger models. How large will they get, and at what cost?
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.