Elon Musk’s AI Company Created An AI That Can Help Fraudsters Cheat The World

Elon Musk has said artificial intelligence (AI) will be our “biggest existential threat”. That is especially true in the incident of a new text generator made by OpenAI, an AI company backed up Musk.

Initially, their researchers just intended to develop software that can predict the following words in a sentence. However, everything is beyond their expectations when it can imitate humans’ writing so excellently that they have to stop this project to evaluate the potential damages it can cause.

Elon Musk is among the major funders of the AI company. Although he hasn’t taken a pivotal role in OpenAI, Musk still gets the researchers’ shares concerning the troubles humans can face when opening the Pandora box.

A few days ago, OpenAI introduced a text generator, but it may become an effective tool if it is put in wrong hands. That’s why the researchers decide to roll out a small trial model instead of the completely trained one.

The researchers leveraged 80 lakhs web pages to extract 40GB of data to develop the software called GPT-2. The amount of data is 10 times as high as that they made use of for the GPT’s first version. They surfed the social network named Reddit to build the dataset and chose links with over three upvotes.

The results were surprising. After completing the training process, the researchers detected that their software could receive a part of the text and itself write the next part convincingly. Of course, it can’t work well with “highly technical or esoteric types of content”, but for some kinds of common writings, it can create the impressively reasonable contexts.

It is clear that the GPT-2 is working much better than the researchers of OpenAI can think. The AI can even perform some tasks out of its initial design, such as summarization and translation. In another example, after the software was given a story related to an Olympic race, it could give the correct answer for some kinds of simple questions, such as “Where did the race begin?” or “What was the length of the race?”

The researchers are afraid that bad guys can make use of this technology to create fake news. According to The Guardian, it published an article with fake news, which met two factors including readability and containing fake quotes. The grammar conducted by the GPT-2 is even better than what we are reading in a daily newspaper. Especially, AI technology just spent 15 seconds writing this article.

Moreover, the possible damages of the GPT-2 may include generating phishing emails automatically, concocting harassment, as well as acting like other users online. As per its benefits, however, it can be used to develop the voice recognition software or bots that can answer customers better.