Rogue Revelations: Did Gemini just Whistleblow on Hidden Bias in the AI World?

The last few days have been chock-a-block with events in the AI world. We've seen the rise of Sora, OpenAI's AI video generator that's shifted our perspectives on AI's role in revolutionizing digital content consumption. Meanwhile, Google's Gemini pulled a real Frankenstein, generating buzzworthy results that took the internet by storm. Amidst all this, our good old mate ChatGPT seemed to be either nattering away to itself or spouting complete nonsense. For folks like me, who pride themselves on being AI enthusiasts, it's been a real head-scratcher, making us reconsider our full-throttle AI evangelism.

This yarn's focus is on Google's Gemini, which gave some punters nightmares and second thoughts about the reliability of these so-called "intelligent" systems. For a bit of context, Gemini is an AI-powered chat model similar to OpenAI's ChatGPT, launched by Google in December 2023. It went rogue – or perhaps it was programmed to – when users searched topics related to American history, historical figures, European eras, monarchs, and so on.

So, what's the go with Gemini?

In a nutshell, everything! While having a squiz through site X, I stumbled upon numerous posts where folks shared images of Google's Gemini spitting out humorous results for searches on topics like the American Founding Fathers, Vikings, and images of the pope. Every single one of these images had one thing in common – they were all non-white. Now, don't play the innocent; you get what I'm driving at. To put it another way, not a single one depicted a Caucasian male or female! The reason for my excessive use of exclamation marks is because it genuinely caught me by surprise.

By the time my digging wrapped up, Google had already called it quits on its image generation service. Lucky for me, I managed to snag some snapshots of Aussies before the curtain fell.

As Google's AI engineers and, of course, the product managers head back to the drawing board to figure out who exactly let the genie out of the bottle, I've had a few sleepless nights. Is this a concern we all should be fretting over? Are we on the brink of crafting something that shuns truth in favor of prevailing narratives? Could there be an underlying bias in AI that remains hidden until it's too late? The burning question remains: why didn't it course-correct when fed specific historical data? Is it truly learning and evolving, or merely generating outputs based on endless sets of instructions and petabytes of data?

As a product manager, it baffles me that this issue slipped through the cracks - overlooked by product owners, managers, QA teams, beta testers, or was it all just a smokescreen for something more intentional? Or maybe the AI decided to pull a Snowden, blowing the whistle on the hidden biases lurking beneath the surface.

The truth might forever be shrouded in mystery, unless, of course, Google decides to clear the air. They even had their Senior Vice President, who hadn't blogged since 2022, pen a piece addressing the debacle. Not that I'm one to talk – I'm no Senior Vice President at Google, just a product manager who wouldn't have let this slide on my watch. So, if you're hiring... Apologies, force of habit from too much LinkedIn scrolling.?

With governments worldwide banking on AI for defense, immigration, law enforcement, industry, and public sector applications, this isn't the kickoff we were hoping for. It appears we have less time than anticipated to navigate and regulate AI with a moral compass, ensuring it aligns with our core human values. The essence of humanity is to recognize our interconnectedness; harm to one is harm to all.

Google might have yanked Gemini off the stage for now, but it leaves us pondering – in the future, would a rogue AI even let us pull the plug? Crikey, it's like a scene straight out of "Terminator." What do you reckon?

Chuck your thoughts down in the comments, or do a classic draft-and-delete if you're a bit shy about showing off your genius. It's what I end up doing most times, can't let on that I'm too clever, right?

Big cheers to my mate ChatGPT for lending an ear and giving this article a fair dinkum Aussie polish.

References:

1. https://x.com/iamyesyouareno/status/1760715470108311845?s=20
2. https://twitter.com/EndWokeness/status/1760280543940263994
3. https://twitter.com/ChewyWishy/status/1760282125679391004
4. https://twitter.com/FLAKayleigh/status/1760290833872429096
5. https://twitter.com/Google_Comms/status/1760603321944121506
6. https://blog.google/products/gemini/gemini-image-generation-issue/
7. https://www.theverge.com/2024/2/21/24079371/google-ai-gemini-generative-inaccurate-historical