In February 2023, Google officially unveiled Bard, its own Artificial Intelligence tool that was created with the aim of competing with OpenAI's more famous ChatGPT.
In March, Mountain View's chatbot was made available only in the United States and Great Britain to a limited number of English-speaking users, and only in the coming months will access be extended to other countries and languages.
Google Bard is mainly based on a broad search language model, which is an optimized and lighter version of LaMDA, another chatbot from the American company that works via artificial neural network.
Here also lies the main difference from ChatGPT. In fact, Google's proposed AI solution uses up-to-date resources fished from the network, while OpenAI's uses a specific proprietary database.
In any case, Google urges caution in using artificial intelligence, advising users a moderate use of Bard and any AI-based tool. Let's find out together a little more about how Google Bard works and possible future implications.
What is Google Bard
Last Feb. 6, Google first announced Bard, its prototype AI chatbot that is supposed to be an answer to rival ChatGPT. The software was unveiled by the CEO of Google and Alphabet's Sundar Pichai.
As early as December 2022, Google indicated its intention to deepen its research on Artificial Intelligence following the official release of ChatGPT. Within the American company, teams of researchers specializing in AI and security have been actively collaborating to launch new products.
Google Bard is thus an innovative tool for Conversational AI powered by LaMDA, Google's next-generation language model. It aims to combine world knowledge with the creativity of the most advanced language models.
Unlike ChatGPT, Bard is structured on the information on the web to ensure recent and accurate answers, giving people the solutions they need.
Specifically, a lightweight version of LaMDA that requires little computing power is used so that Google can make the service scalable and offer it to more users to receive additional feedback. These feedback, together with internal testing, will ensure more high-quality and reliable responses.
As mentioned, ChatGPT and Bard are both conversational AI models that seem similar, but still have differences. In fact, in the case of ChatGPT, this is based on the GPT3 autoregressive language model.
The goal for both, despite using different approaches, is to make the user experience in communication and research better, but especially in Bard's case, the focus is also on integration in the Google search engine And in the Google Assistant.
Google's history suggests how the Mountain View company has been working steadily in recent years to improve its artificial intelligence algorithms.
One of the first advanced models developed by Google was BERT, which has made it possible to define increasingly sophisticated natural language processing models. Instead, for the past 2 years, MUM (Multitask Unified Model) has been introduced, which is much more powerful than BERT and has superior, multilingual data understanding.
MUM is capable of grasping information in many different languages, giving critical information, and answering complex questions that require multi-level knowledge. In addition, MUM is able to support multiple search by simultaneously matching images and text.
This was followed by the current Bard, which employs a language model for dialog applications. It is able to participate in conversations and answer questions, but Google has also pointed out its limitations, such as not being able to return detailed answers on very specific or technical topics. On the other hand, as much as AI is growing exponentially, it is still not immune to the errors.
More generally, Bard seems to have been designed more to stimulate curiosity and creativity than to give precise answers to specific questions. For that reason, many do not consider it a true competitor to ChatGPT. In any case, for now, Bard is still in the experimental and is available only in English in the United States and the United Kingdom.
The reliability of software such as Google Bard
Google and OpenAI share the fact that they are leaders in Artificial Intelligence, but both proposed models can produce false answers and language errors since information from the Internet may contain unreliable notions.
To remedy this problem, the two giants are working to verify the information provided and ensure a more secure search. For now, however the doubts and uncertainties still shroud these kinds of AI tools, including Google Bard.
As an example, the service's promotional video reveals the software's interface and an example of use, but one of the answers Bard gave to a question about discovering planets outside the solar system was wrong.
In fact, the program claimed that the James Webb Space Telescope took the first picture of a planet outside the solar system. In fact, the information was wrong since the first image of an exoplanet was instead captured 18 years earlier by the ESO Observatory's VLT telescope at Paranal.
Needless to say, how a mistake of the genera cost Google dearly, especially financially, as it resulted in a 7% loss on the stock market, equivalent to $100 billion.
Situations of this kind can undermine the credibility of AI and its algorithms, but Google itself wanted to reassure everyone by saying that the service will be released on a large scale only after it has been thoroughly tested and improved. In conclusion, we just have to wait for new developments so that we can experience firsthand the value of these programs.
Future prospects for Google Bard
In the official announcement, Google's CEO highlighted the importance of making accessible advances in AI, not only through Google products, but also through the development of programs and APIs that will make it easier to create new applications.
In addition, it was stated that starting in March, developers, companies and content creators will be able to test the new generative language interface that is based on LaMDA and other later models.
Google then plans to cooperate with the startup to increase the impact of such activities through Google Cloud's partnerships with Cohere, Anthropic and C3.ai. The star-studded company's commitment is thus evident and steadily increasing in order to hold up to the relentless pace of competition.
So the goal is to work with businesses and governments to establish effective standards and best practices, engaging experts and communities to make AI more reliable and useful.