Alphabet Rolls Back AI Image Generation in Google Earth Over Policy Violations

Alphabet Pulls Back AI Image Tool on Google Earth, A Reminder of Generative AI Challenges Google, the parent company of Alphabet, has suspended its generative artificial intelligence image creation feature within Google Earth after the company discovered that some generated output was in violation of internal guidelines. In a statement to TechCrunch, Google said its policies prohibited some of the output, demonstrating another early challenge with launching a widespread generative AI tool. After discovering issues, Google temporarily blocked functionality in Google Earth’s AI photo generator.

Some outputs could have been deemed unusable or not compliant with company standards.

Alphabet said on Wednesday, however, that work on further improving this function for Google Earth, including adding stronger policies, had now removed a barrier for future rollout. Google Earth has existed in some form since June 2001. A version introduced in 2005 featured 3D graphics created from satellite imagery, elevation data, and ground-level photos of many locations around the world. The feature added was one of the newest tools within that system.

GenerativeAI can sometimes come up with errors: While the technology to create entirely new images with the use of text prompted inputs has been refined over the years and has many useful applications, the resulting images can sometimes contain erroneous elements, inaccuracies, or entirely falsified content, depending on how well the prompt is worded.

Google’s decision to rollback was part of Google’s continued approach towards providing Responsible AI services for its many users around the world. Many AI companies are using the ability of models that train upon other AI trained materials to take existing media – from text to images and beyond – and apply learning from it and put an entirely new spin on things which then creates original media. Alphabet stated that when rolling out all of its various AI enabled services, such as those already live on Android, on Google search, for Photos and other tools it uses those technologies by means of keeping human staff in the loop as an additional level of defense to any technology they release.

This level of human interaction is meant to ensure that each model meets all of Google’s safety and content guidelines in addition to simply being very functional, to support its use and not cause user issues, which sometimes includes anything in addition to false imagery including that which could violate acceptable speech guidelines on the public facing interface.