Prospects of Artificial Intelligence in the Cultural and Heritage Field

In this article, we will explore the new possibilities that recent developments in generative AI technologies, and AI in general, present for cultural and heritage organizations.

Challenges of the Cultural Sector in a Digitizing World:

I. Better structures for handling and centralizing data

In just a few years, our world has been flooded with data. The classification systems, both digital and physical, that have been used until now present a significant difficulty known by those who manage them: tracing objects, collections, and samples in a scenario where physical objects constantly move between museums, privately organized exhibitions, and distributed warehouses, and where there is a growing variety of digital formats that contain objects of heritage interest and/or cultural relevance, with copies in different internal repositories and links that may or may not become invalid over time.

Several institutions and researchers have begun aligning their internal identifiers and information structures through Wikidata, highly promising efforts to eventually have a single global data source for these media, but the process of entering this information remains an activity that requires a high amount of human resources and has a low immediate impact in terms of the visitor experience.

II. The new cultural consumer: Experiences to enhance heritage value

While culture and heritage have value in themselves, the values of modern society have significantly shifted in many parts of the world. For better or worse, there is a profile of the new "cultural consumer" who evaluates their alternatives based on the value they will derive from the experience of visiting a site of interest. Furthermore, the offering of cultural experiences coexists with, and is therefore compared to, other types of experiences on the spectrum (broad and diffuse) that ranges from culture to entertainment. In this context, the imperative to preserve, maintain, disseminate, and promote our cultural heritage requires institutions to place the visitor experience at the forefront. These experiences must offer a level of immersion, sensation, and growth that is on par with advances in the rest of the "experiential offering" of our modern world.

III. New generations, new demands

We have already established that the value delivered by the cultural sector is based on the experiences that can be transmitted. Thus, the new generations of visitors that we want to incorporate into the world of cultural heritage are more demanding: they need immersion, personalization, variety, and multimodality. Whereas in the past a linear visit and an audio guide were sufficient, today a direct and personal relationship with heritage objects is required. This relationship should engage in dialogue and adapt the approach of its content to their own worldview and personal interests.

How to tackle these challenges with AI

These new challenges are primarily due to the cultural changes that technology has brought to our society. For this reason, we must also adapt by making intelligent and careful use of the same technologies. Now, let's explore some of the exciting new possibilities that have emerged with the recent advent of generative artificial intelligence, which is now accessible at a low cost.

I. Generative Language Models

Let's start with the "elephant in the room": the ever-increasing diffusion of the use of large language models (LLMs) made available by companies like OpenAI, Anthropic, and the European Mistral. As has already been extensively explained, these models function as a "predictor of the next word." The difference from the autocorrect on our smartphones lies in their greater complexity, as they are capable of generating much longer sequences of text than a single word, and these sequences are highly coherent with the previous sample. Therefore, they are good for following instructions, engaging in conversations, summarizing, etc. This makes them highly efficient natural language text processors. Let's see some possible applications:

Content Classification and Tagging: With LLMs' ability to "understand" human-written text, they can accurately label textual content according to predefined categories. This facilitates the automated construction of taxonomies, collection files, error detection, inconsistency detection, etc.
Semantic Search and Question/Answer Interfaces: Since they use a numerical model to process text, LLMs can understand the semantic proximity between two words or phrases. This makes them effective for performing content searches even when exact keywords are not provided, or for determining the most suitable part of a text to answer a question. This enables the implementation of question/answer interfaces for
Accessibility: With the appropriate adjustments, LLMs allow us to transform requests made in natural language into more precise instructions that machines can understand. When combined with systems that translate voice to text, they can facilitate access to and operation of infrastructure and software for people with diverse abilities.
Automatic Translation: Another factor that enhances access and dissemination of content is the impressive ability of LLMs to translate text into different languages. This is also due to the numerical representation that LLMs use in their internal processing. In fact, this very article was written in one language and automatically translated into two others. Can you guess the original language?

If you want to know how to implement these applications, leave a comment and I will write an article about some implementations that we already have working at Paw Tech.

II. Image, Video, and Audio Generation

If we add the possibility of generating images, video, and audio from natural language instructions, enabled in recent months by "diffusion models" like Stable Diffusion, DALL-E, and others, the creative possibilities increase exponentially. Let's see some possibilities:

Conversing with historical figures: With sufficient context given to an LLM and the generation of video of the character speaking, visitors can virtually interact with historical figures. By using natural language models and text-to-audio translation, users can ask questions and receive authentic and coherent responses as if they were conversing with figures from the past. This provides an educational and entertaining experience, allowing visitors to immerse themselves in historical moments in a unique way.
Multimedia Tours with Personalized Narratives: AI enables the creation of personalized guided tours and multimedia experiences tailored to the individual interests of visitors. By interpreting preferences expressed in natural language, narratives, visual content, and unique audio content can be generated for each user, based on the same core sample or collection. This transforms museum visits and exhibitions into a more personalized and engaging experience, enhancing visitors' emotional connection with cultural content.
Recreation of lost or deteriorated artworks: By using image generation algorithms, it is possible to digitally recreate lost or damaged artworks. This offers cultural institutions the opportunity to virtually preserve and share pieces that might otherwise be inaccessible to the public due to their current condition.

At Paw Tech, we accompany you towards an innovative cultural future.

The adoption of AI is a necessary milestone for the future of cultural institutions. At Paw Tech, we not only offer advanced technological solutions but also help you devise creative proposals that will move your visitors and make cultural heritage more accessible, appealing, and relevant to present and future generations. We explore the ever-evolving landscape of technologies to promote a vibrant, interactive, and accessible culture for everyone.