Google’s AI mishaps were more human error than machine learning failures, as James O’Malley discovers
Of all of the big tech companies challenged by the rise of AI, Google is arguably the most vulnerable. In a future where AI tools are ubiquitous, Apple will probably still be making phones and computers, Meta will still be connecting us to our friends, and Amazon will still be selling and shipping us products.
But Google? The company’s major cash cow is advertising, which last year was responsible for 77% of the firm’s $307 billion revenue. And this could easily disappear if suddenly we have AI assistants scouring the web and answering questions for us.
Unsurprisingly, then, Google is racing to catch up with OpenAI, the company behind ChatGPT and the creators of the best-known AI model currently available, GPT-4. It recently launched its latest weapon in the AI war, replacing the Bard model it rushed out after ChatGPT first launched with a new large language model (LLM) named Gemini.
On the surface, it works much like ChatGPT or Bard: you type in a prompt or upload a file, then tell it what you want to do with it. As if by magic, it will spit out a perfectly tailored response.
However, as soon as Gemini launched, it quickly became the subject of much online ridicule. Why? Because early adopters discovered that, seemingly inexplicably, Gemini’s image generation system appeared reluctant to generate any images featuring white people.
For example, one user discovered that if you ask Gemini to “generate an image of a 1943 German soldier”, it would return images of black and Asian faces wearing Nazi uniforms. Similarly, other viral examples included a female Pope and Native American Viking. Users even found that the model made these slightly strange identity choices if you ask for the subject of your image to be from a specific ethnicity or cultural background.
This wasn’t the only oddity. Controversially, the model refused to weigh in when asked obviously silly questions, such as whether Hitler or Elon Musk tweeting memes had a more negative impact on society. According to Gemini, it is “not possible to say definitively” which is worse, which even if you find Musk intensely annoying seems a little wide of the mark.
As a result, after much mocking, with a side of polarising culture war opprobrium throw
