At its I/O 2022 consumer keynote on Wednesday, Google announced LaMDA 2 as its new conversational artificial intelligence (AI) model that can help make conversations through computing devices easier and more efficient. LaMDA 2 is billed as state-of-the-art conversational AI and is the successor to the original LaMDA (short for Language Model for Conversational Applications), which was introduced at the I/O developer conference last year. In addition to LaMDA 2, Google also announced AI Test Kitchen as its dedicated app to help test AI models, including LaMDA, with people outside the company.
Similar to the original model that debuted last year, LaMDA 2 is designed to help develop better conversational applications.
Google CEO Sundar Pichai said in his keynote that LaMDA 2 has “incredible conversational capabilities” that can help make conversations with computers more natural. The new AI models use massive datasets from a variety of sources to communicate with users on different topics.
Pichai demonstrated the scope of LaMDA 2 with a stage demo, in which a user asked an AI model to describe a scene in the Mariana Trench in the western Pacific. The model responded by providing detailed information about the creatures living in the trenches. It even generated follow-up questions and touched on topics such as submarines and bioluminescence, which it wasn’t specifically trained to answer.
“You can have a conversation anywhere you want. Maybe you’re curious about how scent works and you want to dig deeper. You get a unique response, too,” Pichai said.
He acknowledged that despite LaMDA 2’s design for greater safety, the model could still “produce inaccurate, inappropriate or offensive responses.” For all of these reasons, Google has built the AI Test Kitchen app, which experts like AI researchers, social scientists, and human rights experts can access to help improve ongoing AI development, including LaMDA.
The AI Test Kitchen app contains three modes, “Imagine it”, “Talk about it” and “List it”. The “Imagine” mode allows LaMDA to describe a place – real or imagined. On the other hand, the “talk about it” mode allows the AI model to talk about a topic by detailing its various aspects, while the “list it” mode allows the user to hand any task or topic to the model to get its breakdown into relevant and useful points . The demo showed that the app can also generate hints for the points it provides related to a given topic.
Google also offers the option to send feedback through the AI Test Kitchen app, allowing its engineers to improve AI models.
“One day, we hope we can answer questions on more topics in any language you speak, making knowledge more accessible in search and across Google,” Pichai said.
The AI test kitchen will initially be limited to the US and not all users will be able to use it. Its purpose is essentially to improve the model Google built for testing. The app will also continue to add other “emerging areas of artificial intelligence” to improve various existing and new models.
All of this could ultimately help the Mountain View, California-based company enhance products including Google Search, Google Assistant and others that use software algorithms to understand user preferences and communicate with the machines in our rooms (reading computers, laptops, smart screens and smart speakers) even in our hands (read smartphones).
Google is also developing a technique called Mind Chain Hints to help describe multi-step problems as a series of intermediate steps to help AI models answer even tough questions, such as how many hours are in a month.
When Google’s new technology is combined with its recently launched AI model, the Pathways Language Model (PaLM), it can lead to “promising results” for users, Pichai said.
“In the Mind Chain prompt, we gave the model a question-and-answer pair, but this time, explained how the answer was derived,” he said.
In addition to new AI-focused conversation and language model development, Google also announced the launch of its Machine Learning Center for Google Cloud customers during this year’s I/O consumer keynote. Pichai claims that it is the largest public machine learning center in the world.
The Machine Learning Hub comes with 8 Cloud TPU v4 pods, customized on the same network infrastructure that supports Google’s largest neural model.
“In total, they provide nearly 9 exaflops of computing power — giving our customers an unprecedented ability to run complex models and workloads,” Pichai explained.
The purpose of the Machine Learning Center is to enhance experience in fields such as medicine, logistics and sustainability.
Google is indeed not alone in the field of artificial intelligence and machine learning updates, and its competitors including Amazon, Meta and Microsoft are also vigorously developing their businesses in these emerging areas. However, Google’s new approach aims to be different from its competitors. The company also has a range of products that can be used to take massive datasets and convince large numbers of people to eventually help it test ongoing development in an efficient way.