In this era of information, everyone has access to spread his or her thoughts on the internet. The exposure provided to the internet to everyone is a double-edged sword. With many people spreading fake news over the internet, it is difficult to decide who is legit and who is fake. Nowadays, advanced AI tools help spammers to make fake news faster. In the world full of chaos and people running for their benefits, it is time to use Artificial Intelligence against the liars.
Not all is lost when there are people who want the truth to prevail. Grover from AllenAI is capable to identify which text is made from online text generators. Computer Science and Artificial Intelligence Laboratory students at MIT have concerns about the usage of Grover to find fake news. The team is using the text that is written by humans, and not making new text at all. However, they manipulate the meaning of the pre-written test. Using a GPT-2 language template when performing the edits to maintain consistency, they showed that there are scarier uses.
Tal Schuster, a Ph.D. student at CSAIL, shared his views about the lack of potential in current fake news finders to unveil the truth. The group used a NASA paper that describes the newly collected information on CME. The team acquired an online text generator to write about the usefulness of this data. The AI received an insightful and utterly correct explanation of how the data would help scientists study the magnetic fields of the Earth. Nonetheless, it was classified as fake news. If both were machine-generated, the discrimination between real and fake news was not made by fake news detector. The team reports that the legal use of text generators will become more reliable in the future.
The team at CSAIL has a new algorithm for all of us. Their algorithm is better than all algorithm available already in all aspects. The team hopes that the combination of fact-checking into established defenses will make models stronger against attacks in the future. The team seeks to improve existing models further.
The goal of the team at CSAIL is to make AI able to differentiate real and fake news. The Conference in Hong Kong on Empirical Methods in NLP, which is going to take place in November, a paper will be presented about the contributions of the team to fact-checking.