Google shows off next-generation AI by talking to… Pluto and paper airplanes

Tram Ho

Artificial intelligence is a big part of Google’s business, and at this year’s I/O conference, the company announced a number of achievements from its research into developing language understanding. WHO. The main star of today’s event is an experimental AI model called LaMDA, which Google says could one day power its conversational AI assistants, allowing them to perform Have more natural conversations.

It’s really impressive to see how LaMDA conducts a talk on any topic, ” said Google CEO Sundar Pichai. ” The conversation is surprisingly easy to understand and interesting. But it’s still in the early stages of research, so things haven’t worked out yet.

To demonstrate the capabilities of LaMDA, the company showed videos showing two short conversations conducted by the model. In the first video, LaMDA answers questions while pretending it’s Pluto, and in the second, it switches roles to a…paper airplane. Pichai says the model can cite facts and events throughout the conversation, such as when the New Horizons probe visited Pluto in 2015 (for unknown reasons, both Pichai and LaMDA all call Pluto a planet, when everyone knows it’s actually a dwarf.)

Pretty impressive, right? Despite not being highly trained in a particular field, LaMDA can still speak from the perspective of two unrelated objects. That shows Google’s superiority in the world of AI, where it has been trying to break through the ability to understand and generate language for a long time. Most notably, Google is a pioneer in using a machine learning technique called “transformers,” which excels at language processing, and supports the work of competitors like GPT. -3 by OpenAI.

But what is the point of discussing with a machine? According to Pichai, a large part of the job of AI developed by Google is gathering information, whether through interpreting other languages ​​or understanding what users need when they search the web. If Google can create AI that is better able to understand language, it will be able to improve its core products. Google should be able to turn searching or even using your phone into a conversation – something natural and fluid.

But that is extremely difficult. According to Pichai, ” Language is an incredibly complex thing. We use it to tell stories, joke, and share ideas… The wealth and versatility of language make it one of the greatest tools in the world. mankind’s greatest challenge and one of the greatest challenges in computer science.

One notable point in the presentation was Google’s emphasis on ensuring that LaMDA always adheres to the principles of AI: for example, making sure it benefits society, without prejudice, and across multiple issues. other. This is an area where Google will have to do more research, especially after the company fired two of its best experts on AI ethics, Timnit Gebru and Margaret Mitchell.

The reason they were fired is quite complicated, but may stem from a research report by the duo (and other colleagues) about the potential dangers associated with the kinds of language models that Google has just released. announced above. Obviously, the company doesn’t address these risks on stage. Although the case has not been completely resolved, Google still seems to be so obsessed with its AI future that it forgets to anticipate the immediate dangers.

Reference: TheVerge

Share the news now

Source : Genk