A secret memo leaked on the Internet a week before the annual Google I/O conference set the scene: without a serious shake-up, nothing guarantees Google’s survival. Unsurprisingly, therefore, the big boss Sundar Pichai presented his response on stage Wednesday morning, in the form of an artificial intelligence (AI) embedded in most of Google’s products that already exist.
The big new thing is called PaLM 2 and takes the form of a generative AI similar in what it allows its users to do to OpenAI’s ChatGPT. His speed of reasoning has been multiplied to better understand natural language and up to twenty computer programming languages. She can correct faulty bits of code and solve complex puzzles in seconds, then explain her approach in a clear way like a student in the middle of a ministry exam.
PaLM 2 can be used by third-party publishers of all kinds of applications via traditional Web services integration tools called APIs.
All kinds of apps
Google exemplifies the versatility of PaLM 2 through a medical research-focused variant that can scan an x-ray image of a human body part at a glance. It can spot abnormalities that the most seasoned radiologist might to not see. She can produce a readable and understandable summary of medical reports painstakingly written by health specialists.
It’s so efficient and consumes so little, says Google, that it can stand relatively autonomously on a next-generation smart phone alone.
This second generation of a technology that encompasses research and development in AI made since 2012 by Google will be immediately grafted to around twenty applications used every day by its many customers around the world. This includes Gmail, the Google Workspace office suite, the Android mobile operating system and the new Bard chatroom.
Bard is also expanding: this web tool, halfway between an automated chat tool and a search engine, will now be present in 180 countries and will include around forty languages, including Japanese and Korean.
The explanations, answers and text produced by Bard can be sent in one click to the recipient of their choice by email or on Google Docs. In fact, it will be enough to make the request to him for Bard to take care of writing an entire email or text document.
In a few weeks, Bard will take care of the images, which we can submit to him or that he can generate to help answer visual questions (“What are the historical sites to visit in Old Quebec ?”).
Code of ethics
All these novelties do not fail to raise questions related to ethics and the proper management of information and sensitive data, among others. Especially since Google lost not ten days ago the services of Geoffrey Hinton, considered the grandfather of modern AI, and who slammed the door of Google labs to better resonate his fear that this new technology is misused.
At Google, a spokesperson recalled that “our policy and our code of ethics are known to the public” and that these principles “dictate how we develop our AI and integrate it into our other products”.
“We know that such tools can be used for malicious purposes and we do what we can internally to ensure that our tools are only used in the right way. However, we welcome the public conversation on this subject, we believe that it is essential for society as a whole to be able to deal with this technological evolution. »
The rival out of nowhere
It remains to be seen if all of this will dissipate that strange cloud hanging over Google I/O 2023, a shadow suddenly dreaded by some of Google’s top executives. According to them, new open-access generative AI threatens its business model. These generative AIs are similar in their operation to ChatGPT, from the Californian OpenAI laboratory, but they are free of rights (open source) and can be used for free.
Worse still, if they turn out to be a little less powerful or precise than ChatGPT or even than Bard and PaLM — the two faces of a single AI from Google — they are more agile and amply suited to the needs of the general public, autonomous programmers or start-ups devoid of the corporate rigidity of a multinational like Google. Some can even be installed and live comfortably on a smartphone as easily accessible as a Samsung Galaxy S23.
Ironically, concludes the now not-so-secret note, the main beneficiary of this massive adoption of AI rivals to those of Google and OpenAI could be Meta, to which we owe their existence in large part.
Unless Bard, PaLM 2 and the other novelties shown on stage in Mountain View on a sunny day in May were enough to convince Internet users and companies to pay a premium every month to adopt similar technology already integrated into productivity tools. like the Google Workspace suite.
This report was produced thanks to the invitation of Google Canada.