Artificial Intelligence in the SDG-iLevel Project: The Online Visibility Booster
The Online Visibility Booster, whose development will be led by ACEEU, is a tool which allows university staff to quickly create and publish social media posts (smart texts and appealing images) to effectively promote their individual SDG contributions. It focuses on two main objectives which include ensuring an optimal user experience and call to action, and also ensuring accurate, qualified, varied, and engaging textual and visual messages of individual SDG contributions.
In this article we first provide an outlook on how the Online Visibility Booster will work and then delve into the technical aspects behind it.
How does the Online Visibility Booster work?
Generative AI algorithms are strongly embedded in the development of the booster, for example in the "text generator" algorithm and the "text-to-image" function. A user, mainly academics, will be asked to answer several questions about their role at the university, their academic field, and their current assignments describing, for example, a project they are working on in terms of the project title, aims, the problem they are addressing, and the innovative element of the project. Based on this input, the text generator algorithm will generate up to five formulations (one-sentence texts) explaining the person's contribution to specific SDGs. The user can then select which option of text to post and edit and/or translate the formulations using an integrated online translation tool (e.g. API-based interface from DeepL.com). In addition, social media images of various sizes (Facebook, X/Twitter, LinkedIn) will be created with this text using the text-to-image function by the user’s description of the image they would like to create. The user will be able to instantly create and publish social media posts by using an integrated script. Each post will be hashtagged #mySDGcontribution, creating a shared identity and making it easier to track the impact of the project.
What is Generative AI?
Generative AI, a form of artificial intelligence technology, possesses the capacity to generate diverse types of content, spanning text, visuals, audio, and synthetic data. The recent excitement surrounding generative AI has stemmed from the emergence of user-friendly interfaces that empower users to swiftly create high-quality text, graphics, and videos.
It's important to note that generative AI isn't entirely novel. Its origins trace back to the 1960s when it was first introduced in chatbots. However, it wasn't until 2014, with the advent of generative adversarial networks (GANs) – a category of machine learning algorithms – that generative AI gained the ability to craft
convincingly realistic images, videos, and audio featuring actual individuals.
On one hand, this newfound capability has opened doors to opportunities such as improved movie dubbing and enriched educational content. Simultaneously, it has raised concerns about deepfakes – digitally manipulated images or videos – and posed cybersecurity threats to businesses, including deceptive requests that closely imitate a company employee's superior.
Two other recent advancements, which we will delve into shortly, have played pivotal roles in pushing generative AI into the mainstream: transformers and the revolutionary language models they facilitated. Transformers represent a type of machine learning that eliminates the need to pre-label all data, allowing for the training of increasingly massive models. Consequently, new models could be trained on vast amounts of text data, resulting in responses with greater depth. Additionally, transformers introduced the concept of attention, enabling models to trace connections between words not just within individual sentences but across pages, chapters, and even entire books. Furthermore, transformers extended their capacity to analyze code, proteins, chemicals, and DNA, thanks to their knack for tracking connections.
The rapid progression of large language models (LLMs) – models boasting billions or even trillions of parameters – ushered in a new era where generative AI models could spontaneously generate engaging text, render photorealistic images, and even craft moderately entertaining sitcoms. What's more, innovations in multimodal AI have empowered teams to produce content spanning various media formats,
including text, graphics, and video. This underpins tools like Dall-E, which can automatically generate images from textual descriptions or create textual captions from images.
Notwithstanding these breakthroughs, we are still in the early stages of utilizing generative AI to create coherent text and lifelike stylized visuals. Initial implementations have grappled with issues related to accuracy, bias, and the tendency to produce hallucinations or peculiar responses. Nevertheless, the progress achieved thus far suggests that the intrinsic capabilities of this AI genre could have a profound impact on businesses. In the future, this technology could contribute to coding, drug design, product development, process optimization, and the transformation of supply chains.
How does generative AI work?
Generative AI initiates its creative process with a prompt, which can take the form of text, images, videos, designs, musical notes, or any input comprehensible by the AI system. Subsequently, various AI algorithms generate fresh content in response to this prompt, which may encompass essays, problem solutions, or even authentic-looking creations derived from images or audio recordings of individuals.
In the earlier iterations of generative AI, users had to transmit data via an API or navigate through a complex procedure. Developers were required to acquaint themselves with specialized tools and craft applications employing programming languages like Python.
Today, innovators in the field of generative AI are working on enhancing user experiences by allowing requests to be expressed in everyday language. Following the initial response, users have the option to further tailor the results by providing feedback on aspects such as style, tone, and other elements they wish the generated content to embody.
Generative AI models
Generative AI models integrate a diverse range of AI algorithms to represent and process content. To illustrate, when generating text, multiple natural language processing techniques are employed to transform raw elements like characters (including letters, punctuation, and words) into components such as sentences, parts of speech, entities, and actions, all of which are converted into vectors through various encoding methods. Similarly, for images, various visual elements are converted into vector representations. It is important to note, however, that these techniques can inadvertently encode biases, racism, deception, and promotional exaggeration present in the training data.
Once developers determine how to represent the world, they apply specific neural networks to produce fresh content in response to queries or prompts. Methods like GANs (Generative Adversarial Networks) and variational autoencoders (VAEs) – which consist of both an encoder and decoder – prove effective in generating lifelike human faces, synthetic data for training AI, or even simulations of specific individuals.
Recent advancements in transformer models, such as Google's Bidirectional Encoder Representations from Transformers (BERT), OpenAI's GPT (Generative Pre-trained Transformer), and Google's AlphaFold, have further expanded the capabilities of neural networks. These models can not only encode language, images, and protein structures, but also create novel content.
What are use cases for generative AI?
Generative AI finds applications across a wide spectrum of use cases, enabling the creation of virtually any type of content. Recent advancements, such as GPT's adaptability for various applications, are making this technology increasingly accessible to users of diverse backgrounds. Here are some of the use cases for generative AI:
- Implementing chatbots for customer service and technical support.
- Employing deepfakes for the replication of individuals or specific persons.
- Enhancing the quality of dubbing in movies and educational contentacross different languages.
- Automatically generating email responses, resumes, and academic papers.
- Crafting photorealistic artwork in distinct artistic styles.
- Elevating the quality of product demonstration videos.
- Offering suggestions for novel drug compounds to undergo testing.
- Assisting in the design of physical products and architectural structures.
- Optimizing the creation of new chip designs.
- Composing music in specific styles or tones.