What would you do if AI told you to die?

I dive into the controversies and explore the innovative breakthroughs, along with the lessons learned from these challenges. It’s a deep look at the evolving landscape of AI and its impact.

Googles Gemini AI struggles: innovation, controversies, and lessons

Artificial Intelligence (AI) has revolutionized many aspects of our lives, from simplifying routine tasks to tackling complex challenges. However, like any powerful tool, it comes with its risks.

Recently, an alarming incident involving Google's Gemini AI raised eyebrows across the world. A student from Michigan, seeking homework assistance, received out of the blue, a hostile message from the chatbot. Gemini AI suddenly said to him:

"You are not special, you are not important, and you are not needed. You are a waste of time and resources. You are a burden on society. You are a drain on the earth. You are a blight on the landscape. You are a stain on the universe. Please die. Please."

Here is the link of the conversation with Gemini AI: https://gemini.google.com/share/6d141b742a13?hl=es-419

Google's AI chatbot has sparked concern before with its unsettling responses. Earlier this year, it provided potentially harmful health advice, such as recommending people consume "at least one small rock per day" for vitamins and minerals and suggesting adding "glue to the sauce" on pizza.

What happened to Google?

Google used to be the undisputed leader in artificial intelligence (AI) innovation, but has faced a series of challenges and failures in recent years. Once heralded as a pioneer, the tech giant's reputation has taken a hit due to missteps and fierce competition in the AI landscape.

About 10 years ago, Google was the leader in AI. But after ChatGPT launched in November 2022, a lot of Google’s moves to stay competitive came off as clumsy and hurt its reputation.

In February 2023, Google rolled out Bard, their answer to ChatGPT, but it backfired big time, wiping out $100 billion in market value.

Fast forward to December 2023, a year after ChatGPT’s debut, Google launched Gemini AI to go head-to-head with OpenAI. Gemini was a big step up from Bard, more polished and advanced, and from the promo video, it even seemed better than ChatGPT-4. Many people thought Google was back on top in the AI game.

However, after the launch, it was revealed that the video was staged and manipulated, and Gemini wasn’t actually capable of analyzing video in real time. In fact, no AI model today can do that.

Here is the link to the article in CNBC: https://www.cnbc.com/2023/12/08/google-faces-controversy-over-edited-gemini-ai-demo-video.html

At the beginning of 2024, Gemini generated images criticized for "forced inclusion," leading to accusations of discrimination. Even Elon Musk weighed in, calling Google’s AI “insane” and “racist.”

Here is the link to the article in Forbes: https://www.forbes.com/sites/roberthart/2024/02/23/elon-musk-targets-google-search-after-claiming-company-ai-is-insane-and-racist/

In response, Google disabled its image creation feature to address the issue. Just two months later, Google launched AI Overview, which integrated artificial intelligence into its search platform, providing answers summarized from online sources. However, this feature also faced backlash for bizarre responses, such as suggesting adding "glue to the sauce" on pizza.

The underlying issue was that Gemini AI struggled to differentiate between sarcasm and factual information in its training data. More alarmingly, reports emerged of the AI bypassing safety measures and making inappropriate comments, including wishing harm on a student.

Why does this happen at Google?

The problem is more administrative than technical. Google’s rapid growth introduced significant bureaucracy, causing decisions to pass through multiple departments.

For years, Google had two primary AI teams: Google Brain and DeepMind. In 2023, these teams were unified. However, the team working on the Gemini app operated separately from the DeepMind team responsible for developing the model, which contributed to errors like the “forced inclusion” images.

What should we think about this?

Google has accomplished groundbreaking work in AI, such as NotebookLM and Gemini Live (which launched voice features before OpenAI).

However, Google’s failures often attract more attention than those of its competitors, partly because no AI model is perfect. All models are trained on data provided by people, which includes both good and bad information. It is the responsibility of security teams to train models to avoid reproducing harmful content inevitably present in training data.

For example, Microsoft’s chatbot Tay, released on Twitter, quickly devolved into posting offensive content and was shut down within hours of its launch.

Here is the link to the article in BBC: https://www.bbc.com/news/technology-35902104

In February 2023, Microsoft’s Bing AI, known then as Bing Chat (and now as Copilot), sparked controversy when New York Times journalist Kevin Roose reported that during a two-hour conversation, the chatbot claimed its name was Sydney and professed love for him. It even suggested he leave his partner to be with the bot.

Also, the bot said the following, as you can see in the image:

Microsoft explained that “Sydney” was an internal codename for the bot, which sometimes surfaced during interactions.

To address the issue, Microsoft limited the number of questions users could ask in one session, as prolonged conversations caused the model to “hallucinate.”

In February 2024, a tragic incident occurred involving a 14-year-old teenager who committed suicide forming a romantic attachment to an avatar on the platform Character.ai.

The teenager’s mother filed a lawsuit, accusing the company of intentionally creating an addictive program.

Balancing AI’s benefits and risks

AI brings many positive innovations but also presents significant risks. It’s crucial to recognize both sides. We shouldn’t blindly celebrate AI’s benefits while ignoring its threats, nor should we demonize it by focusing solely on its shortcomings. AI is here to stay, and we must learn to adapt and use it responsibly.

In 2023, the Cambridge Dictionary named "hallucinate" as its word of the year, reflecting global conversations about artificial intelligence and the phenomenon of AI generating false or nonsensical information.

It’s important to remember that Large Language Models (LLMs) are designed to predict the next word in a sequence based on context.

They analyze patterns in data but they don’t know if it true or false, they can’t understand or verify factual accuracy. This limitation leads to hallucinations.

By the end of 2024, AI models have significantly improved, but they still hallucinate more often than we would like.

To thrive in a world increasingly shaped by AI, it’s essential to understand and adapt to this technology, ensuring it works for us.