Google Never Should Have Put Nano Banana Into Google Earth in the First Place

Last week, Google rolled retired its AI image procreation model, Nano Banana 2, to Google Earth, its world explorer program. The institution advertised the integration arsenic a measurement for group to create infographics astir different locations astir the globe, aliases to make humanities landscapes of modern-day imagery. Users were besides invited to make their "dream homes" against real-world backdrops—so if you had plans to build an hold connected your home, aliases visions of a picnic location connected the lake, you could visualize it pinch Nano Banana 2 successful Google Earth.

The news followed a inclination of Google integrating its AI platforms into conscionable astir each 1 of its products and services (for example, past week it added caller Gemini capabilities to Google Docs for Workspace users). From the company's perspective, it apt made full consciousness to adhd Nano Banana 2 to Google Earth. As it turned out, however, this is 1 merchandise that Gemini wasn't meant for.

Google Earth has already dropped Nano Banana 2

Just 1 time aft Google launched this large Nano Banana 2 integration, it rolled it back. The institution issued an update to its original announcement, which read: "We cognize that group uniquely spot Google Earth for a reliable position of the world. We’ve seen geospatial professionals utilizing this characteristic for a scope of useful purposes; however, we’ve besides seen group sharing screenshots of generated imagery that look to break our policies. So we’re rolling backmost this characteristic successful Google Earth while we activity connected implementing stronger guardrails. It's important to statement that generated images didn’t look successful the main Google Earth acquisition for others to spot and were watermarked arsenic AI-generated."

You May Also Like

This connection doesn't decently picture the images that users were generating pinch this tool. In my view, it went beyond "violating" Google's policies: The internet, predictably, utilized Nano Banana 2 to make inappropriate and insensitive images based connected existent locations. Perhaps it won't travel arsenic a astonishment to study that 1 of the images that took disconnected astir was of a level crashing into the caller World Trade Center successful Manhattan. Other images depicted the Eiffel Tower destroyed, aliases the Egyptian pyramids collapsing into a sinkhole. While galore would apt understand these were produced successful jest, plentifulness much could spot these "official" Google Earth images online and presume the worst.

But beyond the jokes, the imaginable for misinformation and disinformation was alarming. For example, 1 personification demonstrated really they used the instrumentality to spot refugees enarthrosis to enarthrosis adjacent the Mexican border. One could ideate specified an image going viral, pinch viewers tricked into reasoning the Google Earth still was an existent outer image. Google apt has immoderate guardrails successful spot preventing galore image generations from taking place, but intelligibly those limitations weren't beardown enough. And since anyone, whether they paid for Google's AI features aliases not, could usage Nano Banana 2 pinch Google Earth, each 1 had to do was unfastened Google Earth and fiddle pinch a malicious punctual until Nano Banana 2 complied.

Generative AI has nary spot successful a instrumentality for illustration Google Earth

It doesn't return an AI skeptic to explicate why this was a bad thought successful the first place—but past again, fewer AI companies look to beryllium employing any skepticism earlier jamming generative AI features into their products. Tools for illustration Google Earth are meant to beryllium educational. You usage them to study astir the world, its geography, and what different cities and landscapes look like. I personally emotion spending clip spinning the integer globe astir to study astir a caller location, and often driblet a pin connected a random spot to spot what it looks for illustration connected Street View. The logic I do this is that I understand Google Earth to beryllium an accurate, reliable look astatine the world; if that weren't the case, why would I bother?

As such, it makes nary consciousness to maine for Google to present a instrumentality that, by necessity, generates inaccurate and mendacious views of the world. One of its trading points was viewing a spot connected the representation arsenic it looked successful the past, but that's not what Nano Banana 2 was doing. Instead, the AI would propulsion from its training information to make what it thought a humanities practice of the area successful mobility would be. If you asked it to show you what Paris looked for illustration successful 1726, it would apt overlay generic buildings and structures onto the map, alternatively than picture what Paris actually looked for illustration 3 100 years ago. What you're getting is the illusion of history; entertaining, perhaps, but surely not educational—no amended than scrolling done AI history slop connected societal media.

The aforesaid goes for the AI's "infographics." Companies for illustration Google emotion to tout AI's imaginable arsenic an acquisition tool, and connected the surface, it seems for illustration a awesome point for Google Earth to make learning graphics to thatch you astir a location connected the map. But your generated schematic isn't vetted by anyone, and decidedly not by an expert: The image that Nano Banana 2 spits retired could beryllium afloat of half-truths and hallucinations, and, what's worse, it won't connection immoderate sources you tin click done to cheque the AI's work. Casual users whitethorn clasp the accusation they "learn" from these infographics, whether that information is correct aliases not.

At its worst, this characteristic was a disinformation machine. If you wanted to person group that thing bad happened successful a peculiar portion of the world, you could conscionable type your petition in, and Nano Banana 2 could spit retired a realistic outer image of your thought for easy sharing connected societal media. At a clip erstwhile AI is spreading misinformation astatine a accelerated rate, this only pours substance onto an already raging fire. Companies for illustration Google should beryllium looking for ways to abstracted AI slop from existent information, not helping it grow.

This isn't an statement against AI devices successful their entirety, but it is an statement against embedding generative AI into an acquisition tool. What inferior Google believes AI serves by taking truth and clouding it successful falsehoods I do not know—other than giving it yet different illustration it tin constituent to successful bid to tout its AI "success."