YJ Kim, from Samsung talks about the concerns and possibilities of generative AIs

A Samsung launched in January this year the new galaxy 24, during the Samsung Unpacked 2024, an event held in the city of San José, in the United States. Presented in 3 versions, all of them arrive with Galaxy A.I., the name given by the company to the new features based on artificial intelligence (AI) on the devices.

Shortly after the event, we spoke with Jun-Youn Kim (YJ Kim), executive vice president and head of the artificial intelligence team at Samsung Mobile eXperience Business, to learn more about the potential of generative AI on smartphones. And also its risks.

YJ Kim has been leading the AI ​​team since 2020 and has played a crucial role in developing the Bixby, a personal assistant present in the company's smartphones, in addition to actively working on the development and incorporation of large AI language models (called LLMs) into Samsung devices and services. He has also worked on other Samsung projects, such as the security suite Knox and payment service Samsung Pay, and now plays a central role in the development of Galaxy A.I..

Artificial Intelligence on the smartphone

Running AI language models on a smartphone like Galaxy S24 it is a complex task. Kim explains that there are two ways to do this: the first, more limited, is the “AI on device”, which consists of incorporating more simplified versions of AIs into mobile devices, so that they can be run even when there is no connection to the web.

The advantage is having more privacy with the data and being able to perform tasks, such as image corrections, text summaries or other available resources, without needing to be connected, with faster responses. At the S24, you can even limit the experience Galaxy A.I. à On-device AI, within the device settings menu.

But, for more complex tasks, Kim comments that the device needs to be connected to the internet, to delegate this task to the cloud, or rather, to data centers. Here, the request is made on the cell phone, but processed on the web server, which then returns the result to the user's phone. The same task performed with an on-device AI may return a better result if you prefer to access data center AIs.

O Circle to Search, for example, is a function provided by Google's AI – Samsung's partner in the development of Samsung solutions Galaxy A.I. – which uses Gemini, an AI located in Google data centers. With it, you can circle anything on the screen, to find out more about it using Google Search. AI identifies texts, images and even concepts of what is on the screen.

And Samsung is also developing its own large language model (LLM), the Gauss, which will allow interactions in the style of Chat GPT. Who knows, to replace Bixby itself, still present on Galaxy smartphones?

Privacy and ethics issue

Kim says that Samsung is placing security, privacy and ethics in the use of artificial intelligence as its first priority, treating the development of these technologies with transparency and responsibility.

The creation of an option to block data from being sent to data centers, as explained in the topic above, is one of these reflections of the company's policy. Users can decide whether or not they prefer to share them, in a hybrid operating model for AIs.

One of the functions that caused the most impact during the presentation of the line S24 No. Samsung Unpacked was the live translation on device for voice and text. It consists of an AI that can listen, at the same time, to the speaker and the listener on the call, translating these speeches into audio and text on the screen.

To preserve user privacy and the confidentiality of translated conversations, Samsung explains that this data is not sent to the data center, being processed only directly on the device.

“Not only does this allow AI to work without an internet connection, users can rest assured that their individual and personal information will remain on their device and not be shared with external servers”He says.

“Users can be assured that their individual and personal information will remain on their devices and will not be shared with external servers”

YJ Kim

Os Principles of Ethics in AIs from Samsung are even publicly listed in this document and translated below:

• Justice
  • We will apply the values ​​of respect for human rights, equality and diversity in AI throughout its entire lifecycle.
  • We will not encourage or propagate negative or unfair prejudices.
  • We will do our best to provide easy access to all users.
• Transparency
  • Users will be informed that they are interacting with AI.
  • The AI ​​will be explained so that users can clearly understand the technology.
  • The process of collecting or using personal data in AI services will be transparent.
• Responsibility
  • We will apply the principles of prioritizing social and ethical responsibility to AI to avoid any violation of human rights.
  • AI will be protected with strong security measures to avoid vulnerabilities to cyber attacks.
  • We will work to ensure that AI benefits society and promotes corporate citizenship.

Even Google's language models, the Gemini Nano and Gemini Pro used on the Galaxy S24, are designed to work offline and with attention to the user's privacy. In fact, the company is a partner of Samsung and co-responsible for launching the features of the Galaxy A.I., as we explain below.

Avoiding the misuse of AIs

As it is a new technology, the Generative AI It is prone to errors and can be misused by its users. For this reason, Kim explains that, in Galaxy A.I., were designed security filters that evaluate results and block problematic, sensitive or even obscene responses in the S24.

Also incorporated watermarks in images modified with the AI Galaxy S24, to ensure that this “edited content” is not confused with an original image:

The resource Generative Editing (Generative Editing, in English) of Galaxy S24, in fact, uses the model Imagen 2 from Google DeepMind, which is Google's most advanced text-to-image fuzzing technology to date.

The Future is individual

According to Kim, Samsung envisions a future where user devices, such as smartphones, will have their own complete AI systems. This means that more and more activities, even complex ones, will no longer be carried out in a data center, moving to local processing.

But, it is very likely that this will require much more processing capacity from the chips of these devices (and their NPUs or neural processing units), going further than the Snapdragon 8 Gen3 from Qualcommpresent in Galaxy S24, can offer.

Language models (LLMs) developed by companies such as Google, Samsung, Meta, Open AI, NVIDIA and others will also need to be reduced, so that they maintain their efficiency, but can be incorporated into devices with processing and storage capacity. data more limited than data centers, as is the case with current smartphones, tablets and PCs. This, in fact, is one of the biggest challenges facing the current industry.

If you are interested in the topic, access our content on artificial intelligence. We also have a complete guide on everything you can do with Galaxy AI. And, if you want to test all this, also check out our Galaxy S24 Ultra review, Samsung's most powerful smartphone today.

Samsung Unpacked 2024

Source: Soyacincau, Samsung Principles of AI Ethics, Samsung Gauus, gadgetguy.

reviewed by Glaucon Vital in 27 / 2 / 24.