Generating language on Twitter

Faculty Member

Kay Kirkpatrick

In recent years, human-like behavior has been observed in automated programs known as internet bots, some of whose functions include benign ones, such as ranking results after a browser search, to malicious ones, such as those that target ad campaigns for political advantage.  These bots provide a starting point for studying artificial intelligence because of their ability to generate human-like responses.  The main goals of this project are to implement programmable bots on Twitter and to quantify their abilities to mimic human behavior.  These bots will be used to create a network that will interact both with themselves and other human users.

Team Meetings

Weekly

Project Difficulty

Intermediate

Undergrad Prerequisites

Some programming.

Project Results

Expanded Project Description:  In 1950, the mathematician Alan Turing put forth an idea in which an algorithmic machine would be able to replicate human behavior sufficiently well to convince another human of its own consciousness. This is known as the Turing test. Although current models of artificial intelligence fall short of real human behavior to a large extent, one could still examine ways in which such machines could be built to improve our understanding of our own cognition, perhaps using the Turing test as a benchmark.

In recent years, human-like behavior has been observed in programs known as internet bots. These are automated programs that perform a variety of functions online. Some of these functions include benign ones, such as ranking results after a browser search, to malicious ones, such as those that target ad campaigns for political advantage. These internet bots provide a starting point for studying artificial intelligence because of their ability to generate human-like responses.

One area where this ability is apparent is the online social media platform known as Twitter. Internet bots on Twitter ("Twitter bots") are automated user profiles that perform the same functions as a human Twitter user. They can interact with other users - both human and nonhuman alike - by tweeting, liking, and retweeting in an algorithmic fashion.

This suggests that a network of interacting Twitter bots can form a complex network, similar to a human social network. Moreover, just as humans possess the capacity for language, perhaps so too do Twitter bots. We propose to study the way in which these bots can create original combinations of words and phrases by programming our own bots and implementing them online. Furthermore, we propose to quantify the degree to which these bots can communicate intelligently by tracking how the bots are tweeted, liked, and retweeted.

In addition, since Twitter users' tweets can be made public, we also propose to engage citizen scientists in an effort to provide qualitative information regarding the reliability of the bots' interactions. This will be done by polling human Twitter users on their experiences interacting with them.

One proposed public bot is a so-called anti-troll. This bot would be called upon by human Twitter users in response to harmful tweets. The bot could be programmed to respond in multiple ways to counter these tweets. This type of function would require the bot to interpret semantic content, a critical hallmark of artificial intelligence.

The main goals of this project are to implement programmable bots on Twitter and to quantify their abilities to mimic human behavior. These bots will be used to create a network that will interact both with themselves and other human users.