Can a Twitter Bot Become Sentient? Will it follow me?
Now that twitter is a mainstream marketing and advertising platform, users with large numbers of active and engaged follower become valuable commodities, as they can be important mediums to advertise and influence. But what kind of audience do these users really have? Are they really engaged? Do they actually care about whom they follow? Do they even really exist? The last question might be the toughest to answer. Twitter has very loose restrictions on who can start an account, and a single person can and do run several. Inflating the number of users and followers a company can claim.
An account whether for parody or another voice is fine and good, but for marketers the problem is that these accounts do not buy products. Because Twitter depends on advertising, it is critical to get an accurate estimate of real, engaged followers, and equally important for those who want to be influencers to amass a large group of followers. There is now a commercial incentive to both create humanoid-like profiles, and an equally strong incentive to discern the difference. This cat and mouse game, where algorithms try to identify fake accounts, and twitter bots grow more sophisticated to try an fool them, could lead to a new level of artificial intelligence. People, this is Blade runner!
For now, the techniques of “Blade Runners” seem pretty crude: analyzing the number of followers, the number of tweets, quality of tweets, and so on. From Twitter Audit:
Each audit takes a random sample of 5000 Twitter followers for a user and calculates a score for each follower. This score is based on number of tweets, date of the last tweet, and ratio of followers to friends.
It seems to me that many causal users would be considered fake under these criteria. Hyperficial ran it’s own audit, and Twitter Audit estimated we had 143 fakes, though it wouldn’t identify them.
88% real. Yeah, I feel that way too sometimes.
The newest techniques, like the one developed at Berkeley involve analyzing the quality of tweets, the pattern of handle and e-mail address, and other technical considerations such as the speed of which the account was created. The researchers boast of a 99% accuracy rate, but more sophisticated accounts are sure to be created that can overcome algorithm-based filters.
Any check based on a pattern or norm can be fooled fairly easily. The best way to find out if someone is real might be simply to ask them. If an account response with a relevant answer when you tweet at them, there’s a good chance there is a person behind it. Right now, even the most advanced AI can only carry on the most basic of conversations, and it doesn’t take much to get a nonsensical exchange. However, if the development of these Turing tests follows the pattern of other technological trends, lucid, engaging, fascinating dialogue will soon emerge.
The inevitable development of twitter artificial intelligence raises interesting philosophical questions concerning sentience and consciousness. But, this is not about trying to become a person in the real world, but a “real” account in the twitter world; that is all that matters to anyone involved. Based on the inherent limitations of twitter (140 characters), and its generally cryptic diction, sentience, as defined by marketers, could be very close.
What if the twitter world exists as a battlefield of robots acting as humans and robot hunters, where the stream of conversations and messages come mostly from artificial accounts? Both sides would attempt to look “real” to avoid detection by the other. Conversations, re-tweets, friendships would form between hunter and bot. Would real interactions be lost in the din? What if one witty twitter bot rises above the rest and becomes a…verified account?