- Blog
Google I/O 2023: Leveraging AI for Digital Development
May 30 — 2023
Artificial Intelligence acts as a significant accelerator for development teams and digital initiatives. Capable of saving us countless debugging hours, providing insights into user behavior, optimizing our customer interactions with tools such as chatbots, and enabling us to stay at the cutting edge of technology. AI isn't merely a buzzword; it's an essential tool for enhancing our products and maintaining competitiveness in a perpetually evolving digital universe.
At the latest Google I/O conference, Google's experts unveiled several new tools that make it easier and quicker to incorporate AI models into digital projects, without requiring significant investments in model development and training. By integrating AI into many of their products and services, Google promises new advancements that will ease the tasks of developers and organizations.
In this article, we delve into the various announcements about the integration of AI into Google Cloud solutions and relevant tools for developing new features or enhancing existing ones in digital products.
Generative AI: The New Companion for Developers
Generative Artificial Intelligence constitutes a true revolution in the field of digital products. It profoundly alters the way we design and interact with digital content. Equipped with the ability to autonomously create various types of content - images, text, or music - based on its learning, it offers vast possibilities in terms of personalization and scalability for digital products. With it, content can be customized for each user, unique designs can be generated, product descriptions can be written, or even music can be created for advertisements. The extent of innovations it can bring about is considerable, making generative AI a real transformational lever in the digital universe.
The combination of traditional coding and generative AI models can lead to remarkable results in the development of innovative features. These advancements, both for web and mobile applications, have the potential to transform user interactions with these products.
Tools that Facilitate AI Accessibility in Development
Gen App Builder: Designing Generative Apps for Chat and Search
Unveiled earlier this year, the Gen App Builder offers developers, including those with limited machine learning experience, the opportunity to quickly and simply design generative applications for chat and search. It's an ideal tool for conducting experiments or swiftly constructing prototypes, like for instance, a counterpart of ChatGPT, but utilizing our own data.
Four New Models Introduced in Vertex AI
Vertex AI is Google's flagship tool for researching, creating, testing, deploying, and evaluating AI models. The introduction of generative AI models expands the selection of models that can be integrated into our projects via APIs. This tool simplifies and democratizes the integration and adoption of personalized intelligent models into applications.
1. PaLM 2 for Text and Chat
PaLM 2, the updated version of one of Vertex AI's foundational models, serves as the basis for several other tools. A foundational model like this is more versatile and can be adjusted to perform a multitude of different tasks. This improved version of PaLM is specifically designed to understand, generate, and translate more than 100 languages, solve complex mathematical problems, and write code in various programming languages. PaLM 2 is the technology that powers over 25 features and products in the Google ecosystem.
2. Codey for code completion
Codey is an AI model specifically designed to transcribe text into code. It assists developers by suggesting code snippets based on comments and analysis. It's adaptable to the unique programming context of different projects. Codey's ability to accelerate the creation of new features is a significant advantage, as is its potential role in facilitating automated testing.
3. Imagen for text-to-image
Imagen is a text-to-image model that generates images from text, facilitating automated image creation. It also supports the editing of pre-existing images, and can be used as a starting point for creating new images. This model provides a range of useful functions:
- Starting from the original image of an item, Imagen can generate new images of the same item in different colors.
- The model can upscale and convert an image to achieve better resolution.
- Imagen can automatically create a caption for images, facilitating the integration of descriptions, tags, and labels into images. As a result, the accessibility of a website or application can be improved.
- When integrated with content moderation tools, it helps increase safety in production environments.
- All images generated by Imagen can be used to create marketing content and can be shared publicly. Additionally, original images uploaded to the model for the purpose of generating others are protected by Google's governance policies and cannot be used or shared with other Imagen users.
4. Chirp for speech-to-text
Chirp is a text-to-speech model that enables the development of functionalities involving voice interaction between the user and the selected website or application. For example, developers can use the text generated by Chirp as input for a chat model. It can also be used to caption videos. The voice instructions generated as a result can be used to converse with a virtual assistant or to execute voice commands.
Embeddings APIs for Text and Images
Accessible via Vertex AI's Model Garden, the new Google Cloud embeddings APIs for text and images offer the ability to develop features such as recommendation engines, classifiers, Q&A chatbots, sentiment analyzers, and other functions requiring semantic understanding of texts or images. Each of these features can be set up to use data from a specific project, thereby achieving more personalized results.
Reinforcement Learning from Human Feedback (RLHF)
The new RLHF feature in Vertex AI facilitates the integration of human feedback into the optimization of AI models. When a model is deployed in a feature, the collection, filtering, and preparation of human feedback on the model's generated results are complex processes. The new tools integrated into Vertex AI simplify these processes, allowing for quicker improvements. Despite automated protections, human feedback remains a valuable tool for preventing hate speech, bias, and restricted content.
Mediapipe: Low-Code Machine Learning Solutions
MediaPipe is another tool developed to integrate AI into our daily lives. It consists of frameworks that enable the execution and deployment of Machine Learning models directly on mobile devices, the Web, connected devices, and other gadgets, without needing to go through the cloud. These models are capable of analyzing images, understanding the semantics of text, and hearing and recognizing sounds.
It's worth noting that, despite the optimization of models run directly on mobile devices, these remain more rudimentary and less precise than the more sophisticated models previously presented that operate on machines in the cloud.
Responsible and Ethical Use of AI
The importance of responsible AI development and data protection - both the data used to refine models and the data generated by users - was highlighted during the presentations. These fundamental principles are found in all of Google's new AI product announcements.
To ensure these principles are adhered to, and for the ethical use of AI in compliance with the latest standards, it is advisable to work with a team of digital product development specialists like Mirego. Our team is ready to advise you so that you can provide your users with an ethical, innovative, and sustainable digital product.
Mirego and AI: A Dynamic Duo!
All these new advancements and features constitute powerful tools for designing new solutions or improving existing functionalities. However, they are not a one-size-fits-all solution that could be used in every context. Each tool has a specific scope, optimized to excel in a particular task. More generic foundational models, such as chat models, excel in communication with users, but often require the support of other specialized models for more specific tasks.
One thing is certain: the future of AI is incredibly promising. For technology enthusiasts like us, these unprecedented innovations and tools are not only stimulating, but they also open up a world of opportunities for the creation of outstanding new digital solutions. At Mirego, our design, strategy, and development teams continue to work in tandem with our clients to find the most effective ways to integrate AI into their digital products.