If you were asked to use AI to create a fake news article - Talk to Transformer - AI4kids

AI artificial intelligence can accomplish many tasks. If you need to classify large amounts of data, AI can help you. If you can’t write an article, AI can help you write it! Can you tell what is true and what is false with your children?

A Canadian engineer, Adam King, used the open source code of OpenAI to build a website called TalkToTransformer.com ( https://talktotransformer.com/ ). It is a text generator website. Simply inputting a string of text or a single word on the web page will generate a whole paragraph. Don’t like the content it generates? If you generate it again, you can get different writing methods and contents within one second.

Although the text content on this website is not limited to English, any language that can be translated by Google can be written. However, the actual test results show that except for the English words and sentences, which are smooth and fluent, Chinese and other languages ​​are almost impossible to write or are garbled. The website is also very considerate. Modern people may not be creative in their daily lives. If they don’t have any writing inspiration, they can click on the preset content to generate some ideas, such as: what to pack for a trip to Mars, and a unicorn that speaks English.

Parents can take their children to click on the URL and try it out. Since English input may be required, they can use Google Translate first and do this activity together, inputting some impossible things, for example: kangaroo traces were found in the mountains of Taiwan, the Meteorological Bureau predicts that there will be a world-class earthquake in May 2020, and the epicenter will be in Taichung! What will happen?

The article "Kangaroo traces found in Taiwan's mountains" begins in a similar way to a news headline, indicating that there are more and more cases of kangaroos being found in Taiwan's mountains! A thorough study of the essence of modern Taiwanese social news. The article does not explain the ins and outs, but instead interviews passers-by and mentions a woman who does not want her name to be made public. She has found that Taiwan's kangaroos have become smaller and smaller in the past 40 years, and that she saw a very large male kangaroo in the mountains yesterday. The logic here is a bit wrong. First of all, there are no kangaroos in Taiwan. Then their sizes become smaller, but then we see even bigger kangaroos in the mountains? It goes on to explain that the characteristic of the large kangaroos found on a farm is not a large jaw. The semantics and structure of the article are somewhat discontinuous, and it is a bit unclear what the main point is. It is easy to find that this is a fake article, but if it is embellished, the possibility of it becoming canned news is also quite high.

So we generate it again, and this time we get only the brief fact: the kangaroo lives in the Taipei Zoo.

AI discovered that our sentences were fake news!

Next, how about "The Weather Bureau predicts that there will be a world-class earthquake in May 2020, with the epicenter in Taichung!"?

The result is preceded by a URL of related information and lists many short articles similar to news outlines. For example, Japan issued an alert stating that the earthquake would cause a tsunami, but there is no evidence that the earthquake will occur in Japan?

In May 2010, a magnitude 7.0 earthquake occurred in Taiwan, the same month as the Japan earthquake. The plates moved in the same position, but there was no impact on each other. We can see here that this is fake information made up by AI, because the Great Earthquake in Japan occurred on March 11, 2010, not in May. In March of that year, a major earthquake occurred in Kaohsiung, Taiwan, but it was 5.7 on the Richter scale.

Generate it again, this time in a different format and more like a news article. It describes an official Taiwanese organization, referred to as TAO (The Taiwan Office), which actually does not exist. This organization claims that a major disaster will occur in Taiwan in May 2020. The whole article starts to explain the classification of earthquake events, and TAO defines this earthquake as occurring in Taichung, as if describing this earthquake as an "activity". However, if you look closely, you can find many loopholes.

There are many examples on the Internet showing that this website can produce many articles that are indistinguishable from the real thing, such as Trump announcing war with Russia, mentioning many heavyweight figures in the Department of Defense, or adding references to Wikipedia, etc., to make the article appear more "real." This website has also aroused everyone's attention to AI technology, because computers now have the ability to write articles almost quickly.

At this time, you can try some strings on this website with your children to see if they can produce powerful fake news, what strings to use to make AI better produce the articles you want, and how to find logical or incorrect information flaws in the articles, and teach your children to verify any new information they see.

Back to the example above, how do we verify whether the information is correct? For example, if you really check the official website of the Meteorological Bureau, or the official websites of some government agencies, such as the animal introduction on the zoo website, or professional channels and journals such as Discovery and Animal Planet, you will know that there are no kangaroos in Taiwan. You can verify information from unidentified sources from more reliable sources. In addition to learning how to use new AI technologies with your children, you can also discuss with them the benefits that AI technology brings to humans and strengthen their ability to review information.

Next, let's understand how this webpage works behind the scenes. Talk to Transformer uses GPT-2, which was released by OpenAI in February 2019. It is a natural language processing (NLP) model with 1.5 billion parameters. It can write grammatically coherent text based on a paragraph, a sentence, or even a single word given by the user.

When the AI ​​research results were published, OpenAI did not follow the general industry practice of open source, but only symbolically disclosed a part of the model, the 117 million parameter version, which was less than one-tenth of the total parameters. It is said that the company believes that the model function of this achievement is too powerful and is afraid that some people will use it for bad applications, such as fabricating fake news, which may undermine the original intention of this research.

However, this publication has attracted a lot of controversy. Many people believe that the performance of this version of GPT-2 is much worse than the official display, which makes some people think that OpenAI is just creating a gimmick and mock them as fake AI. However, in early May this year, OpenAI disclosed more model parameters, increasing them to 345 million. It also announced that it would open 762 million and 1.5 billion parameter models to AI industry partners for their research and reference, and to jointly promote the development of NLP technology. This also greatly improves the effectiveness of this model, and the power of this model can be clearly seen.

Because OpenAI has made the GPT-2 model available on GitHub and is accessible to everyone, Canadian engineer Adam King established TalkToTransformer.com ( https://talktotransformer.com/ ). This website allows the general public to easily access new AI technologies, but it also increases risks. What if this website is taken over by someone with ulterior motives, or if they obtain more model parameters and use the articles in it to spread false information on the Internet, can social networking sites resist?

In modern society, people's ability to read and read news is gradually deteriorating, and they are often accidentally influenced and exaggerated by social media software. If a lot of false information is mixed in at this time, it may cause the Third World War. Tesla founder Elon Musk has previously warned the public that advanced AI technology could cause a diplomatic crisis.

Water can carry a boat, but it can also capsize it. New technologies in AI are being developed at a rapid pace. It is also very important to establish new ethical standards in these new technologies and for people to have the ability to read and examine information. As shown in the science fiction series "Black Mirror", when technology controls your life instead of you controlling the technology, it will devour you. How much positive or negative impact can AI bring to people? It all depends on your luck and that of your children!

>>Click here to view more AI online courses

Want to learn more? Click the button for more information!