AI Talks

The Announcement

On December 6th, Google unveiled Gemini 1.0, its most powerful AI model to date. According to the announcement, this next-generation multimodal AI model surpasses state-of-the-art performance across numerous benchmarks, excelling at complex reasoning, understanding diverse media formats, and boasting advanced coding capabilities.

The Controversy

The same day, Google released a captivating YouTube video titled “Hands-on with Gemini: Interacting with multimodal AI.” The video delves into Gemini’s impressive capabilities, showcasing its ability to answer questions, generate diverse creative text, such as poems and code, translate languages, and even play games and solve competitive programming problems. The demo sparked a lot of excitement, reminiscent of the ChatGPT release in November last year.

I was disappointed to learn, through social media, that the content had been edited and scripted to portray a live interaction with Gemini. Although there are introductory notes at the beginning of the video, they fail to explicitly mention that the live interaction with Gemini is simulated. Additionally, Google included a disclaimer about latency in the demo, but it was discreetly placed in the video notes rather than being prominently stated. This practice could be seen as deceptive and unethical since not many viewers are likely to delve into the notes of a YouTube video.

This raises significant concerns about Google’s choice to release the video without the appropriate level of disclosure that would help viewers understand the background and motivation behind it. Why does the company consistently make mistakes in their presentation materials whenever they unveil a new generative AI product? Is it the fear of being outpaced by competitors in this fast-paced AI race? Is there excessive internal pressure on the team to perform and deliver? Who within the organization oversees and approves these materials before their public release? Google claims that it is committed to develop AI responsibly and with trustworthiness. However, the irresponsible use of fake interactions in a video to promote a product will surely undermine these claims.

Under media scrutiny, Google admitted they created the video with the aim of inspiring developers. However, the presentation was perceived to be misleading and approached manipulation. Such misrepresentation, whether accurate or perceived, could adversely affect the company’s reputation and potentially harm investors by acting as a drag on the stock’s long-term performance.

Recommendations

I’d like to propose several recommendations for Google to prevent similar situations in the future. While I would be surprised if the company doesn’t already have numerous safeguards in place for product announcements and launches, considering the controversy surrounding the Gemini announcement, it’s still appropriate to present these suggestions :

• Establish a high-level committee, including the head of ethics, to review and approve product launch materials.
• Conduct live demos to showcase the product’s performance under normal conditions.
• Clearly and explicitly disclose, at the beginning of product presentation videos, any editing or scripting that might present the product’s performance differently from user expectations. Avoid incorporating “inspirational” elements that exaggerate the product’s capabilities.
• Let a select group of external independent users test the product and get an honest feedback on their experiences.

If the company adheres to these recommendations or its own internal guidelines (which I am confident are in place), there won’t be a cloud of controversy to overshadow the next product unveiling and steal its spotlight.

Additional Readings

Google DeepMind, “Welcome to the Gemini Era”

Google for Developers, “How it’s Made: Interacting with Gemini through multimodal prompting“, December 6, 2023