AI Secret
Posts
AI Moonshot

AI Moonshot

Magna Ding
September 14, 2024

Advertise / All Articles / Unsubscribe

Have a great weekend, AI Enthusiasts!

Today, I’d like to share a story about innovation, bravery, and overcoming challenges.

WEEKEND STORY

AI Robotics Moonshot

In January 2016, Hans Peter Brondmo joined Google X, Alphabet's innovation lab, to lead a new and uncertain endeavor: creating intelligent robots that could live and work alongside humans. At the time, Google X, often called the moonshot factory, had already birthed major projects like Waymo and Google Glass. The company had also acquired nine robotics companies, but their vision remained unclear following the departure of Andy Rubin, "the father of Android."

Brondmo and the head of Google X, Astro Teller, were tasked with figuring out what to do with these robotics acquisitions. They set out on a massive mission: to give AI a body, powered by artificial intelligence, to help solve global labor shortages, driven by aging populations and shrinking workforces. But building robots that could function safely and efficiently in everyday settings—beyond the predictable environments of factories—was a daunting challenge.

The Vision: Robots for Everyday Life

Everyday Robots, as the project came to be known, had its roots in Google X's moonshot philosophy: tackling enormous problems, leveraging breakthrough technologies, and creating radical solutions that would seem just on the edge of possibility. Brondmo and his team believed AI-powered robots could one day assist humans with everything from wiping down café tables to sorting trash. They envisioned a future where these machines, capable of end-to-end learning, would autonomously perform complex tasks without needing constant supervision.

Brondmo shared that one of the biggest motivators was his own mother's health struggles. Living with Parkinson’s disease, she relied heavily on human caregivers in her Oslo apartment. Every time they spoke, her first question was, "When are the robots coming?" She hoped these machines could offer the assistance that her human caregivers couldn’t always provide. However, Brondmo knew the journey toward creating robots capable of such intricate support was just beginning.

The Reality: Hard Lessons in Robotics

To develop robots that could learn from their mistakes, the team employed a strategy of reinforcement learning. One experiment involved 14 robot arms in a setup they called the "arm-farm," where robots practiced picking up objects like Lego blocks or rubber ducks. Early on, the robots had only a 7% success rate, but after months of practice and iterative improvements, they achieved over 70% success.

Yet, as Brondmo admitted, seven robots working for months to learn how to pick up a rubber duck wasn’t nearly enough. The team needed to scale faster, and so they created a cloud-based simulation to accelerate learning. By 2021, they had built more than 240 million robot instances in this simulation, allowing the robots to "dream" and "learn" overnight.

The Philosophical Dilemma: Should Robots Look Like Us?

Throughout the journey, the team debated whether robots should resemble humans. One of the most pivotal moments came during a conversation about whether robots should have legs or wheels. Vincent Dureau, a senior engineer and wheelchair user, ended the debate with a simple remark: "I figure that if I can get there, the robots should be able to get there." This reminded the team that efficiency and practicality often trump mimicry.

Instead of designing humanoid robots, Everyday Robots opted for simpler, more efficient forms, focusing on how quickly they could deploy useful machines in real-world environments. By 2022, they had developed robots capable of cleaning tables, sorting trash, and navigating around people in cafeterias and conference rooms—progress that made Brondmo’s team feel they were on the verge of a significant breakthrough.

The Shock: Everyday Robots Shut Down

In late 2022, as OpenAI released ChatGPT and the world marveled at AI's language capabilities, the focus shifted. Despite years of investment and development, Google decided to shut down Everyday Robots in early 2023, citing cost concerns. Brondmo and his team were stunned. The robots and a small group of researchers were transferred to Google DeepMind, but the grand moonshot of AI-powered robots had come to an abrupt halt.

The Future: A National Imperative

Brondmo’s personal experiences, particularly his mother’s health challenges, convinced him that the world still desperately needs AI-powered robots. By 2050, the world’s working-age population will decline drastically, leaving fewer people to care for the elderly, work in factories, and drive vehicles. Robots are not optional in addressing these labor shortages. While Silicon Valley startups often lack the patient capital necessary for such large-scale ventures, Brondmo emphasizes that countries like Japan, China, and South Korea have made robotics a national imperative.

Although Everyday Robots didn’t reach its full potential, Brondmo remains confident that this kind of technology will eventually transform our lives. The robots might not have arrived in time to help his mother, but he believes their time is still coming—and soon.

TOGETHER WITH ONCELY

The Hottest AI Lifetime Deals from Oncely.com

Transcribing and translating content can be time-consuming and expensive. But what if you had an AI-powered solution that handles it all—quickly, accurately, and affordably?

Imagine an AI-driven platform that eliminates the hassle of manual transcription and translation.

Introducing SubEasy: Your go-to solution for seamless transcription and translation. With AI-powered accuracy and support for over 100 languages, it simplifies your workflow across any content.

Enhance your productivity and save time with SubEasy.

This product is provided by Oncely.com. Oncely is the #1 Lifetime Deals Marketplace for AI products. The software landscape has evolved into a Lifetime era — Pay Once, Use Forever, and say goodbye to monthly fees.

QUICK HEATS

Amazon is testing ads in its shopping-focused chatbot, Rufus, for U.S. users.
California's AI safety bill SB 1047 has been criticized by Yann LeCun, while Geoffrey Hinton endorsed it.
Hasbro CEO Chris Cocks announces plans to significantly integrate AI into Dungeons & Dragons development, citing its growing use among players.
Facebook has admitted to scraping the public data of all Australian adult users to train its AI, without providing an opt-out option.
Google launched Audio Overviews, an AI-generated podcast feature for its note-taking app that summarizes user notes.

AI TOOLS

⁉️ iAsk.AI allows users to ask AI questions and receive instant, accurate, and factual answers.
☎️ ChatClient customs AI chatbot like ChatGPT in one-click.
☔ Aura AI protects your identity, finances, and data.
⌨ RoboCoder makes programming easier.
💪 DANchat creates expert AI prompts for every task.
🎯 Fork.ai gets app leads with verified emails and contacts. (Featured)
🔀 ChatHub simplifies workflow by comparing AI chatbot responses. (Featured)
💡 MimicPC instantly runs Stable Diffusion and Flux online. (Featured)

Share AI Secret

If you enjoyed today’s content, please help us spread the word by clicking the button below to share it. We would really appreciate it.

1 Million+ AI enthusiasts are eager to learn about your product.

AI Secret is the world’s #1 AI Newsletter, boasting over 1 million readers from leading companies such as OpenAI, Google, Meta, and Microsoft. We've assisted in promoting Over 500 AI-Related Products. Will yours be the next?

What We Can Offer:

Launch an Advertising Campaign
Introduce New Product or Features
Other Business Cooperation

Email our co-founder Mark directly at [email protected] if the button fails.