Google has launched a new Android app called ‘AI Test Kitchen’ to let anyone play with some of its experimental artificial intelligence (AI) projects such as LaMDA (Language Model for Dialogue Applications). Participation in the program will be opened in phases to select group of people who have registered for it.
The Alphabet company said that the goal of the app is to let a wider set of people and developers use some of its AI projects and then use their feedback to improve them further. The first three demos available under the AI Test Kitchen will explore the capabilities of LaMDA 2.
LaMDA is a conversational AI model, which can be trained on data sets to have open-ended conversations with humans. It is built on a neural network architecture, called Transformers, which is designed to produce a model that can be trained to read words, pay attention to how they relate to one another and think of a response that is in line with the context of the conversation.
In June, a software engineer at Google sparked controversy by claiming that LaMDA had become sentient and was responding to conversations on rights and personhood. The claim was dismissed by Google and the engineer was suspended.
Being Sentinet is a state where a machine can experience feelings and sensations like humans or animals. Though AI has been shown to be capable of it in sci-fi movies, most experts believe that AI sentience is still many years away.
“While we’ve made substantial improvements in safety and accuracy in the latest version of LaMDA, we’re still at the beginning of a journey,” Google said.
According to Google, during the demos, the participants can see how LaMDA generates quick responses in different scenarios. It is one of the core strengths of the program, however, Google warned that generative responses can be inaccurate or inappropriate too.
Though Google has assured users that any conversation or feedback on the project will be collected and stored in a way that it is not linked to their Google account, it cautioned them to not share any personal information during a conversation with LaMDA. The conversations can be seen by human reviewers. If users don’t want that they can delete a conversation during the demo.
The first demo will revolve around imagination, where users can name a place, for instance, and LaMDA will suggest ways to explore it. In the second demo, users can list a goal or topic and LaMDA will break it into subtasks. The third demo is more open-ended and can venture into any area users want.