[ad_1]
Google uses it a lot to show the annual developer conference, I / O Artificial intelligence with the wow factor. It was introduced in 2016 Google Home smart speaker with Google Assistant. 2018, Duplex it was premiered to answer business calls and schedule appointments. In keeping with that tradition, last month CEO Sundar Pichai introduced LaMDA, an AI “designed to be a conversation on any topic”.
In an on-stage demo, Pichai showed what it’s like to talk to a paper plane and Pluto to a celestial body. For each query, LaMDA responded with three or four sentences that resemble a natural conversation between two people. Over time, Pichaik said, LaMDA could be included in Google products, including Assistant, Workspace, and most importantly, search.
“We believe that LaMDA’s natural dialogue capabilities have the potential to make information and computing radically more accessible and accessible,” Pichaik said.
The LaMDA showcase provides a window into Google’s search approach that goes beyond the list of links to find out how billions of people do it online. This approach focuses on AI that can infer meaning in human language, engage in conversation, and answer multiple-choice questions as an expert.
I / On also introduced Google as another AI tool, called the Multitask Unified Model (MUM), which can take into account searches with text and images. VP President Prabhakar Raghavan said that someday users will take a picture of a pair of shoes and ask the search engine if they can be worn while the shoes are being climbed on Mount Fuji.
MUM generates results in 75 languages and Google says it understands the world better. A demo on stage showed how the MUM would respond to the search query “I’ve climbed Mt. Adams and now they want to walk the Mt. Fuji next fall, what should I do differently?” your search is different because MUM wants to reduce the number of searches needed to find an answer. MUM can summarize and create text; will know how to compare Mount Adams with Mount Fuji and travel preparation may require results in search of fitness workouts, hiking equipment recommendations, and weather forecasts.
“On paper”Rethinking the search: making amateurs experts“It was published last month by four researchers at Google Research who thought of the search as an interview with human experts. An example on paper” What are the health benefits and risks of red wine? “Today, Google responds with a list of bullet points The article suggests that a future answer looks like a paragraph that says red wine promotes cardiovascular health but stains teeth, with citations and links to sources of information.The article shows the answer as a text, but it’s also easy to imagine oral responses, now with Google Assistant as an experience.
But relying more on AI to decipher text also carries risks, as computers have difficulty understanding language with full complexity. The most advanced AI for tasks such as text creation or answering questions, known as large language models, have shown a tendency to increase bias and unpredictability or create toxic text. Such a model, OpenAI‘s GPT-3, has been used to create interactive stories for animated characters but it also has has created a text about the sex scenes in which children participate in an online game.
Of a paper and demo published online last year, researchers at MIT, Intel and Facebook found that large language models show biases based on stereotypes about race, gender, religion and profession.
Rachael Tatman, a linguist with a doctorate in natural ethics of processing, says that as the text created by these models becomes more compelling, people can believe that they are talking to an AI who understands the meaning of words. creates – in fact, when he does not understand the world sensibly. This can be a problem when creating a text that is toxic people with disabilities or Muslims or tell people committed suicide. As he grew up, Tatman recalls that a librarian taught him how to judge the validity of Google search results. If Google combines large language models with search, users will need to learn how to evaluate conversations with AI experts.
[ad_2]
Source link
