Google doesn’t appear to have a plan or model in place to ensure that creators of content used by its AI, Bard, are paid.
Google Bard’s vice president of engineering, Amarnag Subramanya, was asked about plans to pay creators of content used by Bard at a launch event for the next version of AI.
The next iteration was published this week. Upgrades to Bard include 40+ languages, expansion to 59 countries and regions, the ability for Bard to understand pictures, the ability to speak answers to users, and more.
Responding to Bard’s question about paying users for content, Subramanya said Google is still in the early stages of rolling out generative models.
“I don’t think we’re very clear about the proper business model for these models,” he says.
“Like any large language model, we are very transparent that the models we use are trained on content that is publicly available on the web.”
Subramanya said publishers, policy makers, regulators, experts and users need to come together to make sure everyone is “doing it right.”
He said the company has a history of this approach.
But Google’s largest online search tool faces additional challenges from the government and press industry both in New Zealand and abroad. They want to pay for news hosted and monetized by tech giants.
Google recently agreed to start paying a number of news outlets, including Stuff, to contribute content to its News Showcase Platform, a site that will launch internationally in 2020.
But news publishers are still waiting for government legislation to require companies like Google and Facebook to enter into payment agreements for news published on major search platforms.
Asked if Bard would start disclosing the sources of information, Subramanya said Bard’s aim is not to plagiarize others’ work, but to create new content.
“That said, there are certainly cases where Bard generates content and text similar to those found in other sources,” he said.
“In those cases, we’re obviously already citing those sources.”
Byrd mentioned when questioned about a number of recent news stories, including alleged bullying against Kiri Alan, Auckland bus driver strike, Prime Minister Chris Hipkins’ visit to China, recent political poll results, etc. There was only one answer. The first response was a news site.
When Byrd then asked where he got the information used in his answers, the AI almost always cited multiple news sites.
In the final search, the AI refused to provide sources even after being questioned, stating that it was outside its capabilities.
A privacy commissioner warned in late May that companies are bound by privacy laws just as they are AI tools, urging anyone using AI systems that might capture personal information to create new content. He said he expects government agencies to consider the consequences.
Victor J. Blue/Bloomberg
Google is one of many technology giants competing in the generative AI space, including Amazon and OpenAI.
Subramanya said users can control their interaction usage by going to Bird’s “My Activity” link and deleting certain conversations or interactions over a period of time.
Users can also set automatic deletion of interactions for 3 months, 18 months, or forever.
He said the privacy policy clearly states that some interactions are used to improve the quality of the bard model and that there are automated systems that attempt to remove personal information.
“We also have a human review process to ensure that some of this private data is not included in model training.”
What’s New in Bard
When Bard launched about three-and-a-half months ago, the response was somewhat disappointing, with some emphasizing that its features and coverage were on par with, or less extensive than, its competitors.
Subramanya said the tool started as a limited experiment in the US and UK, with Google intentionally starting with a “slightly lightweight large language model” to get feedback and ensure the tool’s safety. Stated.
This update fills many gaps in Bard’s functionality.
After the update users will be able to import images into Bard and Bard will understand what the image represents and be able to manipulate it.
The new system works with Google Lens, an image recognition technology, to introduce generative AI that can understand images, identify objects and places, and talk to users.
Subramagna said she recently used the feature to declutter her desk, allowing Bird to recognize what was on her desk and how to organize it.
This feature is initially only available in English, but will be extended to other languages.
Users can also save and label conversations so they can return to the same discussion and resume it at a later date.
Users can also change how Bard responds, shorten or lengthen responses, or request a more relaxed or formal tone.
Also, Bard is no longer a text-only response system, users can now tap the small speaker icon to have Bard speak the answer.
“It’s great to read something, but often it’s better to listen to it,” Subramanya said.
“For example, if you ask Bird to write a script, you probably want to listen to it rather than read it.”
Subramanya said the audio system could help people understand how to pronounce words.
In its biggest expansion to date, Bard will be available in 40 new languages, including Arabic, Chinese, German, Hindi and Spanish.
Bard was already available in English, Japanese and Korean.
“We feel this is a big step towards democratizing both knowledge and imagination,” Subramanya said.
The system will be expanded to a further 59 countries and territories, including all EU member states, India, Indonesia, Thailand, etc.
Coding has also become a major use of Bard, and updates have improved the ability to export certain types of code.
With this update, Python code, a general-purpose programming language, can now be exported to Replit (a code collaboration platform) in addition to Google Colab.
Subramanya said Bard’s goal is always to generate secure code, but it’s a collaborative tool and you should always check if the code is fit for purpose.
There will also be a new sharing feature that will allow users to share Bird’s answers with friends and family.
