On the internet, seeing is not believing

It’s time to update the old warning, “Don’t believe everything you read,” to “Don’t believe everything you see.”

Driving the news: Image editing tools that use AI to generate convincing—but fake—visuals are going mainstream with Adobe’s announcement that it is adding a “Generative Fill” feature to its popular Photoshop software.

  • Generative Fill lets users edit real photos with nothing more than text prompts, producing images that look more realistic than those generated entirely through AI in tools like Midjourney.
  • Tutorial videos posted online show people using the tool to add animals to their landscape photos, change outfits, and place people in fake background scenes.

Why it matters: Photoshopped images circulating online are nothing new, but the latest crop of AI editing tools has dramatically expanded the pool of people who can create convincing fakes. That’s already having real-world consequences.

  • Earlier this week, an AI-edited image that purported to show the Pentagon on fire went viral on Twitter and caused markets to dip briefly.
  • Filters used to change people’s appearance on apps like TikTok have become so realistic some experts warn they could have harmful psychological effects on users.

Yes, but: Most people will use these tools for non-evil, creative endeavours. The Photoshop subreddit is filled with examples of art users have created with Generative Fill, as well as more pedestrian uses like expanding photos or enhancing headshots.

Bottom line: We still don’t have any consistent way of quickly verifying the authenticity of an image (though some ideas for fixing this are floating around). Until that changes, a healthy skepticism about what you see online is warranted.