Bard: get to know Google’s ‘ChatGPT’ features and controversies – 07/13/2023 – Market

Launched by Google to face the resounding arrival of ChatGPT in the area of ​​artificial intelligence chats, the Bard program has versions in Brazilian Portuguese and in 40 other languages ​​that are available to the public from this Thursday (13)

The tech giant had a slightly more cautious stance on AI development, but has been changing its approach as rivals advance in this field.

Microsoft invested US$13 billion (R$62.4 billion) in OpenAI, the ChatGPT company, which attracted 100 million users in just two months of existence.

Chatbots like Bard and ChatGPT answer a variety of questions and perform tasks ranging from writing an apology to putting together programming code.

These models have been impressing for interacting with users in a language considered “more human” than in previous technologies.

Experts, however, are concerned about the consequences of the widespread and unrestricted use of artificial intelligence programs.

A manifesto with more than a thousand signatures of prominent figures in the area even called for a pause in the development of some systems.

Google tries to emphasize in its ads that Bard is an experimental stage project and may occasionally share incorrect or false information.

In the same week that the arrival of the chatbot to new countries was announced, Google found itself involved in controversies for its practices in this field.

The launch of Bard in Brazil is important considering the relevance of the country as a market for the company (more than 90% of searches by Brazilian users are via Google) and the possibility that chatbots may become, in the future, the main instrument for obtain information on the internet, replacing search engines.

Bard’s new features

Bard can be accessed from both desktop and mobile through this address. You need to login with Google account to use.

The company suggests that the user asks for tips on places to visit and itineraries for outings, advice to improve productivity or start routine practices (“how to read more?”), explanations of complex topics and even action plans to solve life problems.

Bruno Pôssas, Google’s global vice president of engineering for search, warns that the system is still subject to “hallucinations”—that is, when the program gives an unexpected response, out of line with what the programmers plan.

“Although the model’s responses seem realistic, in some cases they will not be anchored in facts”, he says.

Among cases of hallucinations reported in recent months, a Microsoft AI system advised a columnist for The New York Times to “end the marriage”.

Also announced are the following features and updates to Bard:

• the program will give audio answers, not just written ones, which can help in case you know the pronunciation of words for sure;
• images can be used in task requests for Bard, with the aid of the Google Lens tool;
• the answers given by the system can be adjusted according to tone and style in five options: “simple”, “long”, “short”, “professional” or “casual” (the company says that this feature will be available soon in languages Besides English);
• Bard conversations can be pinned to a sidebar for easy reference;
• sharing conversations with the program to the user’s network;
• Export programming codes to more locations.

Google’s controversies with its AI

In recent days, Google has been the subject of controversy over the way it trains its artificial intelligence systems – among which Bard is one of the flagships.

On Tuesday (11), a class action lawsuit in California sought compensation from the company for “stealing everything that was ever created and shared on the internet by hundreds of millions of Americans”.

Models like Bard are fed a huge amount of data and then trained to give answers from that base.

However, there are accusations that copyrighted works are used — without financial compensation — to power a system that will be commercially exploited later.

The lawsuit also states that people’s sensitive information also ends up in this data repository.

Google, in a statement sent to CNN, said the company has made it clear for years that the information used is from public sources and that it is in line with its “AI principles”.

“US law supports the use of public information to create beneficial new uses.”

At the launch of the Brazilian version of Bard, Google did not clarify how it will monetize its chatbot or whether it will use user activities for commercial purposes, such as selling personalized ads and shopping suggestions. But he said that the Brazilian LGPD (General Law for the Protection of Personal Data) will be followed.

Already on Wednesday (12), a report on the Bloomberg website stated that, according to internal company documents, humans who work to train systems like Bard receive low wages, have an excessive workload and face the stress of completing complex tasks in up to three minutes.

The work of these humans is critical to preventing Google’s AI models from providing offensive, false, or problematic content responses.

The company said it “does extensive work to build our AI products responsibly, including rigorous testing, training and feedback processes that we’ve refined over the years to emphasize truthfulness and reduce bias. [nas respostas]”.

This text was published here.