Mint Primer | No way to tell a fake: AI images face reality check

Zomato has stopped using AI-generated food images. Artists and several others have also come out against such images. While there might be a compelling case for AI-generated images in some instances, platforms have to ensure transparency. Mint explains:

Why did Zomato ban AI food images?

It was in response to customer complaints about artificial intelligence (AI)-generated food photos. In a post on X, Zomato CEO Deepinder Goyal said: “One place where we strongly discourage the use of AI is images for dishes in restaurant menus. AI-generated food/dish images are misleading, and we have received numerous customer complaints on this issue.” This leads to breach of trust and higher complaints, refunds, and also, lower ratings. The platform urged restaurants to avoid using artificial intelligence for images of dishes. Zomato will start removing such images by the end of August.

Read more: India’s generative AI startups look beyond ChatGPT-like models

What is the problem with AI images?

Software has been used over the past few decades to improve images. But with AI, it’s not just about improving an image, but much more. There are many problems with AI-generated images: they can make the real look ‘unreal’, generate fakes, mislead users, create mistrust, and there could be copyright issues. They can also undermine those creating real work—from loss of creative control to ownership ambiguity. When it comes to brands, marketing, etc., where consumers make decisions based on what they see, like on e-commerce sites or food apps, AI images can be misleading.

How do you identify an AI-generated image?

It’s not easy. It’s about using good AI to track down bad AI. When experts examine an image for signs of AI, they try to zoom in as much as possible on every part of it. Odd outlines, inconsistent pixels and misplaced shapes may identify a fake generated by AI. According to Nexcess, a web hosting firm, even AI savvy people could identify AI images only about 50% of the time.

Read more: What Ola founder Bhavish Aggarwal’s AI chip ‘moonshot’ means

What is the regulation on such images?

Typically, content created by generative AI is considered to be in the public domain and copyright-free because it lacks human authorship. India does not have laws on generative AI, such as deepfakes. It has instead a series of advisories and guidelines to encourage responsible development and implementation of AI technologies. It’s up to brands not to misuse images—like making hotel rooms look better than they actually are or tourist spots look like an El Dorado. It will impact brand reputation.

What’s the future of AI-generated images?

Zomato’s ban also applies to their marketing team—they have been told not to use AI-generated images for marketing purposes. AI image generators will have to prioritize transparency and fairness in image generation processes, ensuring that visuals are inclusive, representative, respectful of diverse perspectives, identities and are real. There will be ways to help users spot an AI image (like a watermark). It’s up to image generation platforms like DALL-E, Midjourney and Nightcafe to ensure transparency.

Read more: Artificial intelligence: Not hype but an economic catalyst

 

Leave a Comment