Google LLC today revealed that it has invested $300 million in Anthropic, a startup that researches and develops “reliable and controllable” artificial intelligence systems, as part of the search giant’s foray into generative AI, one of the hottest areas of technology today.

As part of the deal, Google will receive about a 10% stake in Anthropic. The company will also purchase cloud-computing resources in a “preferred provider” partnership with Google Cloud.

Additionally, Google Cloud intends to build out its own large-scale, application-specific AI-accelerator tensor processing unit and graphics processing unit clusters that Anthropic plans to use to train and deploy its AI models on.

Generative AI refers to artificial intelligence algorithms that can generate text, images and other media from prompts from users in plain language. The technology has captured the hearts and minds of millions of people and generated headlines following the launch of OpenAI LLC’s ChatGPT last year, which is capable of answering questions almost like a person would.

Anthropic was founded in January 2021 by a team of researchers, engineers, policy experts and others with experience with the AI breakthroughs such as the large language model such as GPT-3, which is what powers ChatGPT. There’s also reinforcement learning from human feedback that allows machine learning models to learn quickly to become more conversational based on positive or negative feedback.

The company’s focus is on AI safety and its objective is working on building reliable and steerable AI systems that are more predictable. The hope of the collaborative team is to eliminate bias in the AI model and overcome some of the limitations that current conversational AI has.

“At Google, we believe it is imperative to pursue AI boldly and responsibly,” said James Manyika, senior vice president of technology and society at Google. “We are committed to developing and delivering useful and beneficial applications, applying responsible principles grounded in human values and safety.”

Through this partnership, Google will gain access to Anthropic’s flagship conversational assistant AI product named Claude. It’s an intelligent agent chatbot that can understand and react to natural language, similar to how conversational AI models such as ChatGPT operate, but it’s currently only available in closed beta through a Slack integration.

Anthropic describes the process it used to create Claude as “Constitutional AI,” a learning model for AI that uses both reinforcement learning and supervised learning to critique its behavior in order to produce a more “harmless AI assistant.” According to the research paper on the subject, it uses self-improvement to identify harmful outputs and fine-tune itself to weed them out.

That will become increasingly important as generative and conversational AI becomes more common. AI depends heavily on the type of data it’s trained on and how dependent people are on it for their day-to-day use. In the past, machine learning algorithms used in smart agents displayed problematic racial and gender biases, repeated slurs and spread misinformation. That’s still possible even with the new innovations such as ChatGPT, and OpenAI needed to add guardrails to protect against it.

Competition is hot in the conversational AI space, with Microsoft Corp. investing $10 billion into OpenAI, the maker of ChatGPT and adding its service to Azure as well as integrating GPT-3.5 into its Sales Viva platform. Chinese search giant Baidu Inc. this week announced plans to launch its own ChatGPT rival in China’s markets based on the machine-learning model Ernie. Further competition could come from Character AI, a neural network AI founded by former Google researchers who worked on the company’s internal LaMDA AI system.

Image: geralt/Pixabay