Detecting AI generated content

With AI generated content starting to get embedded in our daily life, Generative AI is bringing a new challenge for the researchers.

Can AI help to separate wheat from the chaff?

Deluge of AI generated content brings new challenges

Students have started using chatGPT for their university assignments. Software programmers are taking assistance from code generators and avoiding frequently reaching out to their senior colleagues for help. Marketing companies are planning to replace their content creators with AI based image/video creation tools. Web searches now have started showing synthetic content more often.

As we have started using AI to provide answers, instead of raw information, this can influence our decisions and we do need to trust it much more deeply than we have before. This is the same level of trust that we have with Google Search providing most relevant results on the first page, and Wikipedia providing usable information even though the information could be edited by almost anybody.

Increased use of Generative AI is bound to raise some questions about how does someone differentiate between truth and untruth, fact and fiction, real and fake? AI-based platforms are being used to create deep fakes that can be used to sow the seeds of discord in society and create chaos in markets. Microsoft CEO, Satya Nadella recently talked about the AI hallucinations? Watch Satya Nadella’s Big bet on AI on WSJ’s podcast.

In this post, we aim to take a look into the challenge of identifying AI generated content and whether AI itself can help us to overcome this challenge.

AI content detection to the rescue

AI content detectors can be used to identify AI-generated content by analyzing various aspects of the content and identifying patterns or features that are indicative of the type of content. These are a few ways in which AI content detectors can help:

Text Analysis: AI content detectors can analyze the text of AI-generated content to identify features such as tone, language, vocabulary, and structure, which can help determine the type of content.

Image and Video Analysis: AI content detectors can also analyze images and videos generated by AI algorithms to identify the subject matter, style, and content type. This can include features such as color, contrast, composition, and subject matter, which can help identify the type of content.

Comparison with known content: AI content detectors can compare the AI-generated content to a database of known content to determine similarities and differences.

Available methods and tools to detect AI generated content

OpenAI’s GPT-3: This is a language model that can generate text, answer questions, and even write articles. It can also be used to detect content generated by similar models.

Writer.com: It provides a free (with a word limit) as well as paid access and is a powerful tool that accurately identifies AI content, providing insight into the originality of the content.

Sightengine: Sightengine is an AI-powered image and video analysis tool that uses computer vision algorithms to analyze visual content and detect various features and attributes. It can be trained to detect specific types of content or features based on specific use cases or applications.

The Giant Language Model Test Room: GLTR is a  It is a collaborative effort between the MIT-IBM Watson AI lab and Harvard NLP is a enables us to detect the presence of AI-generated content in text. Using a variety of natural language processing algorithms to analyze the text, including a giant language model, it can detect whether the text has been produced by AI.

Clarifai: Clarifai is an AI-powered visual recognition tool to help detect a wide range of visual concepts, including objects, scenes, and text. It can be used to detect a wide range of visual content, including AI-generated images and videos.

This is just an indicative list of tools available and the list is growing rapidly.

Decade Forward for AI generated content and how is industry is evolving

The two technology giants are indirectly locking horns in AI space, i.e. Microsoft and Alphabet(Google parent company) launching their own adaptations of AI in a matter of weeks, and shows that the space is starting to heat up for what even the AI can’t imagine. It is going to be very important for the organizations competing in Generative AI space to apply controls, principles and validations so that the generated content is trustworthy.

Until humans learn to objectively evaluate online content, AI content detection tools have to be our assistant to separate the wheat from the chaff. Consumers of the online content will require methods to detect AI generated content to avoid misinformation, protect authentic ideas and to limit intentional information bias.