Recently at its I/O developer conference, Google has announced multiple things, and as usual, the company is also demonstrating new developments in AI and machine learning. The company demonstrated a new app called AI Test Kitchen in addition to its new LaMDA 2 conversational AI model. Google last year unveiled its futuristic language model for dialogue applications (LaMDA). Before you get excited about this, you should know that the feature will roll out to users residing in the USA only.
It’s understandable that the company is hesitant to make the product public. After all, big tech’s AI projects have been criticized on multiple occasions. In 2015, it was discovered that Google’s image labelling algorithm classified Black people as “gorillas.” Twitter taught Microsoft’s chatbot, Tay, to be a sexist, neo-Nazi in 2016. Zo, a Microsoft chatbot released months after Tay’s unfortunate debut, went too far in censoring. Because it was discriminatory towards women, Amazon removed an internal AI recruitment tool in 2018.
The difference between LaMDA and a normal smart speaker that says, “Here’s something we found on the web,” is that it’s designed to say things confidently. And in a conversational tone, so consumers will assume assertions to be true. That’s why Google says “major hurdles remain to be solved before these models can truly be useful,” and it’ll be in beta for a while.
What is the AI Test Kitchen app?
Josh Woodward, senior director of product management at Google explains that AI Test Kitchen, an Android app that will provide limited access to Google’s newest AI language model, LaMDA 2, to a select group of users. The model is an update to the original LaMDA, which was revealed at Google I/O last year and has the same fundamental functionality: you talk to it, and it responds. However, Test Kitchen provides a new, user-friendly interface that invites users to provide comments on the system’s performance.
In an interview according to Woodward, the goal is to establish an experimental environment for Google’s latest AI models. “These language models are incredibly fascinating,” he says, “but they’re also quite incomplete.” “And we want to figure out a method to progressively get things into people’s hands so they can both see how valuable it is and give feedback and point out places where it falls short.”
“Imagine It,” “Talk About It,” and “List It” are the three modes of the app, each designed to evaluate a different part of the system’s performance.
“Imagine It,” asks users to name a real or imaginary location, which LaMDA will then describe (the test is whether LaMDA can match your description).
“Talk About It” provides a conversational prompt (such as “talk to a tennis ball about a dog”) with the goal of seeing if LaMDA can break down any task or topic into useful bullet points.
And “List It,” asks users to name any task or topic, with the goal of seeing if LaMDA can break it down.
When can you expect to AI Test Kitchen App?
In the next few months, AI Test Kitchen will be available in the United States; but it will not be available for free on the Play Store. According to Woodward, Google hasn’t determined how it would provide access; but it’s likely to be by invitation only, with the corporation reaching out to academics, researchers, and officials to see if they’d like to check it out.
According to Woodward, Google intends to distribute the app “in such a way that users understand what they’re signing up for when they use it, knowing that it will say false things.” It’ll say things that aren’t always representative of the final product.”
This announcement and framing convey several messages: First, AI language models are extremely complicated systems, and comprehensive testing to uncover all possible mistake scenarios is something a business like Google does not believe it can perform without outside assistance. Second, Google is very aware of how vulnerable these AI language models are to failure, and it wants to limit expectations.
Future of the technology of AI Test Kitchen App
It’s difficult not to think about how technology like LaMDA may affect Google in the future; particularly its most important product: Search, after seeing it in action. Although Google emphasizes that the AI Test Kitchen app is merely a research tool, its functionality is clearly linked to Google’s services. For Google Assistant, for example, staying on topic is critical, and Test Kitchen’s “List It” mode is nearly identical to Google’s “Things to know” function, which breaks down activities and topics into bullet points in search.
In a research article published last year, Google (maybe inadvertently) encouraged such suspicion. Four of the company’s engineers proposed in the paper that, rather than typing inquiries into a search box and giving consumers the results. Future search engines would function as mediators, analyzing the substance of the results and then extracting the most helpful information. Obviously, this approach introduces new issues arising from AI models, such as bias in outcomes and systems that make up answers.
I am an Android Enthusiast and a Ninja at content creation. I love writing anything about technology and pro at reviewing smartphones. Big time expert in multiple facets of content generation