<li>Artificial intelligence is interesting, but not an end in itself: we are developing AI for public welfare goals, increasing data availability, working in an interdisciplinary manner and seeking a broad social dialogue in development. </li> <li>Visitors become users: Not just looking, but finding independent access and deciding for yourself how you want to reuse the collection. We work with a CC-0 licence so that free access to culture is possible and subsequent use is legally secure.  </li> <li>Many of the objects are not accessible in on-site exhibitions and are being made digitally accessible here for the first time. We want to offer users worldwide independent access to the museum's collections from any location.</li> <li>Everyone has different interests in terms of content: We support individual access to the collections.</li> <li>We often know little about the background when we look at an object: We improve data availability and data quality with the aim of making it easier to access content-related contexts. </li> <li>So many things - how am I supposed to get an overview or access? We improve the searchability and findability of digital museum objects. </li> <li>The tool is not a single AI, 'the' museum AI or a single neural network, but offers support for individual user requirements through machine learning processes and AI models. </li> <li>The requirements for desired access to the digital collection of the users of the Badisches Landesmuseum and the Allard Pierson Museum always take centre stage. </li>

What AI strategy is behind this?

<li>To create a story, we ask you to register. You can publish your story for others. For data protection reasons and for content quality assurance, we need your registration here. Your data is safe with us. Please also note the <termsLink>terms of use</termsLink> and <privacyLink>privacy policy</privacyLink>. </li>

What happens to my data?

<li>Ethics guidelines: Our AI initiative is based on the recommendations of the German Ethics Council of 2023. </li> <li>Support through AI: Our aim is to use AI to support visitors and provide better access to art and culture. AI solutions are used in the tool to improve the ability to act in the context of the digital museum visit. We endeavour to promote a realistic understanding of artificial, or rather machine intelligence. </li> <li>Labelling of machine results: We ensure that all content generated by AI is clearly labelled as such to ensure transparency.</li> <li>Avoid misleading AI imagery: We avoid portraying AI as omnipotent in order to promote realistic expectations.</li> <li>No anthropomorphism: We avoid humanising AI in order to promote a clear understanding of how it works.</li> <li>Diversity and internationality in the team: Our international and diverse team involves different user groups in development right from the start.</li> <li>Research and development standards: We always orientate ourselves to the current state of research and development in order to ensure the quality and relevance of AI applications.</li>

What ethical guidelines were observed when using and developing machine intelligence?

<li>For cultural heritage fans and collection enthusiasts - techies and developers - co-thinkers and creatives.</li><li>For pupils, students, researchers or people with special interests in cultural history.</li>

Who is xCurator for?

<li>Wild story - better not ask. There was pizza, Club Mate and online meetings. </li> <li>There were various <dataLabLink> hackathons where wild ideas were developed</dataLabLink>. </li> <li>A group of interested AI pilots accompanied the project from the start of development and provided information on their wishes and requirements for AI-supported access to the collection. The tool development was only possible thanks to many discussions and digital volunteering - thank you very much! </li> <li>The <dreipc>3pc</dreipc> team worked on the concept and implementation in 2022 and 2023. </li> <li>The advisory board of the Badisches Landesmuseum has contributed important suggestions for the tool development!</li> <li>The team at the Badisches Landesmuseum and Allard Pierson have had many discussions about the appropriate approach to machine intelligence. </li> <li>In addition, many experts from around the world have contributed knowledge and tips. You can find the list of contributors <dataLabLink>here</dataLabLink>. </li>

How was the tool developed?

<li>You can create stories with AI support and thus help to make your most important stories accessible to others. </li> <li>You can give feedback if results are unclear or incorrect. We use the feedback to improve the AI processes in the background. </li> <li>You can take a look at the code <codeLink>here</codeLink> and make suggestions on how we can improve the tool. </li> <li>For questions or ideas, simply contact creative@landesmuseum.de </li>

How can I contribute to development and improvement?

<li>Recommendations for use in schools can be found <recommendation>here</recommendation>.</li> 

How can I use the tool in schools or educational work?

<li>These contents were generated using Artificial Intelligence. Please note that human judgment is necessary to ultimately assess accuracy. Content is often not 100% accurate but, for example, only 90% likely. Therefore, it's crucial to critically evaluate the results!</li><li>This translation was generated by a machine.</li><li>These keywords were machine-generated through an Entity Recognition algorithm.</li><li>These images are machine-recognized to provide you with more probable search results.</li><li>These connections are pre-sorted for you by an image clustering algorithm.</li><li>This text was automatically generated by a language model. Please carefully review it to ensure it is suitable for your purposes and linguistically and contextually accurate.</li>

Info buttons

<li>Discover: Your view into history: You can immerse yourself in 50,000 years of cultural history, discover connections and develop your own stories. With an experimental neuronal search, you will find exciting and surprising motifs. </li><li>Create: You can write stories for others and share them. Here you can use high-quality story templates and customise or further develop them according to your wishes. </li><li>Let machine intelligence support you: various methods of machine learning and generative artificial intelligence will help you and perhaps surprise you with a new perspective?</li><li>Help the museum: Sometimes the algorithms still have to learn from you. Do you want to confirm correct content and give feedback to others? If you notice anything, we would be delighted to receive your feedback so that we can improve the AI training in the long term! </li>

What can I do here?

<li>Artificial intelligence (AI) is a field of computer science. It involves getting computers to act in certain ways that humans perceive as intelligent. <ideaLink>The idea is over 100 years old</ideaLink> and today's technology can be used in various applications. You can find artificial intelligence in your smartphone or in your music recommendations. What has been new in recent years is the development of generative technologies that enable images or texts to be created automatically. This is why it is often said that artificial intelligence can also be 'creative'. You can find many exciting <learningLink>learning resources here</learningLink>.</li><li>Artificial intelligence refers to various technologies, such as artificial neural networks for image recognition or language models for translating or creating texts. </li><li>Incidentally, we prefer to talk about machine intelligence and have already discussed this a lot. <readMore>You can read about it here</readMore>. </li><li>In xCurator, machine intelligence is used to make the digital collections more accessible, personalised and interest-driven, and to help you make connections and create your own stories. </li><li>This sometimes works quite well and is helpful, but sometimes we perceive machine-generated results as 'wrong' because many calculations are probabilities or approximate values and a 100% accuracy rate can only be achieved with human help. We must therefore learn to deal with these approximate values, to critically assess machine-generated results or to process them correctly ourselves. This is sometimes not so easy to bear. </li><li>Perhaps you like collecting errors or helping us to improve the algorithms? You can use the feedback functions to help assess and train machine-generated results and improve the processes.</li><li>Please note that machine intelligence is not human intelligence - your own judgement is always most important. </li>

What does artificial intelligence mean?

<li>Wherever you see this symbol<imageLiSmall></imageLiSmall>, we use machine learning or show results with the help of generative artificial intelligence:</li><li>Intelligent search: How creatively can you search? The system suggests objects that match your specified interests and searches. On the homepage, museum objects are sorted into so-called clusters, which help you to access the variety of objects in the collection more quickly and coherently. The intelligent search includes a visual search based on similarities and pattern recognition and algorithms pre-sort the variety of content for you. Just give it a try! </li><li>You can use intelligent filter options to expand or narrow down the search by colour, time, material and location.</li><li>Machine-generated keywords and intelligent enrichment: New keywords were generated from the image and text information to help create new connections between things. Help the museum by assessing the quality of the keywords and automatic links. </li><li>Similarity search and recommendations: The images of the objects have been analysed and made accessible via an image search, which means that colours, motifs, objects and themes in images are recognised. Images are sorted into groups by the system so that you are shown content and visual connections. When you create a new story, you will receive recommendations based on your search and title details as to which objects might be suitable. </li><li>Personalised settings allow you to find more exciting things more quickly and discover your individual approach to the past. Your personal data is safe with us and is not processed by artificial intelligence processes. </li><li>Generative texts: The tool helps you to create texts and recognise connections between things. You can use xCurator as a co-pilot to develop new, creative and exciting contexts. We have worked a lot on the data basis so that the language model doesn't hallucinate so much, but instead draws on the collection data and really new and exciting content. However, you can give the language model hints as to what it should do for you. But be careful, you have to judge the results of the language model yourself! Simply use the museum texts to help you. </li><li>For security reasons, we have preset the prompts. If you have any good suggestions on how the prompts can be improved or if you would like a specific prompt for your story, the museum would be happy to hear from you!</li>

Where is artificial intelligence here? 

<li>xCurator was developed as a reusable and customisable open source software solution, which can be understood in detail here:<gitLink> https://github.com/Badisches-Landesmuseum/xcurator</gitLink> </li> <li>xCurator uses various machine learning methods to support you in indexing, understanding and your own production process. xCurator uses methods from the fields of natural language processing, entity extraction and entity linking, image recognition, intelligent search, multimodal methods and generative language models. </li> <li>Solutions and models from Elastic Search, Open Clip, DeepL, Spacey, <elasticSimilarLink>Elasticsearch Similarity</elasticSimilarLink>;<elasticSearchSpring> Elasticsearch Queries with Spring Data</elasticSearchSpring>;<mappingTypes> Alternatives to mapping types</mappingTypes>;<elasticSearchRelation> Managing Relations Inside Elasticsearch</elasticSearchRelation>; <dslQuery>Query DSL</dslQuery>, <elasticSearchDoc>Py ElasticSearch Documentation</elasticSearchDoc>, <elasticSearchDoc>Py ElasticSearch GitHub</elasticSearchDoc>,<entiyLinker> Radbound Entity Linker (REL)</entiyLinker>; <flairLink>flairNLP (Beta)</flairLink>,<openClip> OpenClip </openClip> a solution from <pixSolution>https://github.com/pixolution/embeddings-grid</pixSolution> are currently used. </li> <li>We are currently testing the quality of a <qualityTest> Named Entity Recognition and Entity Linking System</qualityTest> together with the Staatsbibliothek zu Berlin - Preußischer Kulturbesitz (https://staatsbibliothek-berlin.de/) from the QURATOR - Curation Technologies project (https://qurator.ai/) and are improving the hit rate for you: </li> <li>We have discussed, researched and tested a lot: The tool currently also uses the language model solution from OpenAI. Why? Because language models characterise our present day and offer opportunities as well as risks. Because it is important to actively engage with the effects of forms of artificial intelligence such as language models - including in protected spaces and through culture. Here you can practise creating content with the help of language models on a solid factual basis. Your new super skill will be prompting and fact-checking, I promise! </li> <li>In the <gitLink>technical documentation </gitLink> and the <dataLabLink> Datalab</dataLabLink>, you can find out more about the technologies and background and also an experimental place that is also exciting for curious people, researchers and developers. Simply join in and become part of the cosmos! </li>

Which AI processes and models were used exactly?

<li>Artificial intelligence should be used ethically and for the common good and support human authorship and responsible behaviour. With xCurator, we worked with the public and experts to find ways to make the digital museum more accessible. </li><li>Machine learning and deep learning can be used to take a fresh look at what already exists. The ability of artificial neural network processes to reorganise archives or exhibits in collections can make new connections visible. </li><li>In the Creative User Empowerment project, various methods were investigated and developed to create new access to digital collections using machine intelligence. The focus was always on the question of how the processes can best support users and what role and image we want to attribute to artificial intelligence. We know from the history of ideas that ideas about abstract concepts change over time. Our understanding of artificial intelligence must therefore always be understood historically. <dataLabLink>To the project documentation and experimental space</dataLabLink>. </li>

Why do we use Artificial Intelligence?

<li>Yes, of course, you can <museumLink> go to the museum </museumLink> in analogue form or find other digital offers <digitalLink> here </digitalLink>. You're sure to discover other topics and stories there. The museum looks forward to seeing you!</li>

Is it possible without Artificial Intelligence?

Here you will find all the answers to your questions about the xCurator.

About the xCurator

No title

You can export the user data here by giving the user id.

Export user data

Here you can manage the reported objects.

Manage objects

Here you can manage the reported stories and objects as well as the LLM (Large Language Model) templates.

Admin

Here you can manage the reported stories.

Manage stories

Edit LLM templates

Manage notifications

Here you will find all the information about the object

Help us by rating the quality of the keywords and automatic links.

Object details

By the way, did you know that you can change the language settings and the titles and object texts will then be displayed in the selected language? There is also machine intelligence behind this. The central text content is translated and provided in all the languages used for the project using a <li>DeepL integration.</li>

People, places or organisations are often mentioned in the descriptions of museum objects. In order to facilitate your research, we have enriched our datasets with additional information using machine intelligence. For this purpose, all text information of the museum objects was analysed to identify named entities (persons, places, organisations). This process is known as entity linking. The terms found were then linked to external sources such as <wikipedia>Wikipedia</wikipedia>, <wikidata>Wikidata</wikidata> and the <gnd>Common Authority File (GND)</gnd> and marked in the texts to provide direct access to further information. In addition, the results are used to supplement the metadata of the museum objects. The identified terms are displayed as purple keywords above the images. This also allows you to recognise which content is AI-generated: the purple font and the stars always indicate AI-generated content.

The search results are clearly displayed on the canvas, the digital workspace. The museum objects appear in so-called clusters, i.e. groups of objects that show content-related and visual connections. These clusters help you to access the variety of objects in the collection more quickly and coherently.

The best match to your search query is placed in the centre. The surrounding objects are the next best matches in descending order of similarity. The further you move away from the centre, the less the results match your search query.

Discover

Canvas and Cluster

Error detected?

Computer vision is a field of computer science that deals with the ability of computers to analyse and interpret visual data such as images and videos. Research centres on human perception and the question of how human perception can be simulated in computer systems. Computer vision encompasses a wide range of applications that enable machines to analyse and classify information from visual data and to detect patterns. The most common applications of computer vision systems include face recognition, object recognition, optical character recognition (OCR) and semantic segmentation. Today, computer vision applications can be found in many areas of everyday life. They are used to unlock smartphones, automatically tag photos on social media platforms and in autonomous vehicles.

Convolutional Neural Networks (CNNs) are at the heart of machine vision. CNNs are  special types of <li>artificial neural networks</li> that are particularly well suited to processing and analysing visual data. They are designed to recognise patterns in images by applying multiple layers of filters that identify different features such as edges, textures or colours. These features are then combined to identify more complex structures such as shapes and objects.

Use the neural search of xCurator and discover the digitized collections of the Badisches Landesmuseum and Allard Pierson Amsterdam with the help of filters.

Curious about the AI features in the xCurator? The Deep Dive icon will take you to more detailed information.

The xCurator gives you access to a wide range of objects from the collections of the Badisches Landesmuseum and the Allard Pierson Museum in Amsterdam. In 'Discover' mode, you can browse through 50,000 years of cultural history and discover a wealth of exciting content. In order to make the digitised museum objects more searchable, the data records in the xCurator are enhanced with the help of <li>artificial intelligence</li> - a process also known as 'data enrichment'. This involves enriching the existing data records with new information generated by AI. This additional information includes aspects such as colour mapping, the recognition of patterns and motifs in images as well as the analysis and translation of object texts. Machine intelligence also makes it possible to analyse large volumes of image data and thus identify themes, similarities and correlations between different data sets. Ideally, this opens up new perspectives on the collections.

To help you make the most of this enhanced data, we have developed a user-friendly interface that allows you to intuitively explore the museum collections and discover new connections between the objects.

A particular highlight of the 'Discover' mode is the intelligent search. It not only returns results based on your search term, but also makes suggestions based on visual similarities.

To enable an intelligent search, image data must first be converted into so-called embeddings. <li>Embeddings</li> are digital representations (numerical representations) of data such as images or text that enable the system to process complex information. Pre-trained models are used to generate embeddings. CLIP embeddings are used in xCurator. <li2>CLIP</li2> (Contrastive Language-Image Pretraining) is a model developed by OpenAI that projects both image and text data into the same dimensional space. This allows a multimodal search.

Multimodal means that the search does not only access one data type (e.g. text), but several data types (e.g. text, image, video, audio) simultaneously and combines them. For each modality, specific features are extracted (feature extraction):<ul><liItem>Text: Important keywords, semantic meanings or entities (people, places, organisations).</liItem><liItem>Images: Objects, colours, patterns, faces.</liItem><liItem>Audio: Speech patterns, pitches, words and phrases.</liItem><liItem>Video: Visual scenes, objects, movement patterns, along with the speech contained in the audio tracks.</liItem></ul> These features are then brought into a common visualisation form so that they can be compared and combined. This leads to more precise and comprehensive search results. The generated embeddings are used in the xCurator for various search and exploration functions to provide you with relevant and contextualised results. Detailed information about the models in the xCurator can be found <li>here.</li>

What are embeddings and multimodal methods?

Sometimes machine-generated results appear 'wrong' because they are based on probabilities and approximate values. Sometimes AI systems (especially language models) invent information. This phenomenon is known as <li>AI hallucinations</li>. Approaching 100% accuracy is only possible with human assistance. Therefore, it is important to critically examine machine-generated results and, if necessary, revise them yourself. This can sometimes be challenging. Remember that machine intelligence is not the same as human intelligence - your own judgement is always crucial.

Metadata is data that provides information about other data (such as museum objects). In the museum world, metadata refers to the descriptive information that characterises an object and enables it to be managed, searched and presented. When an object is digitally recorded in a collection, a variety of metadata is recorded - including basic information such as title, maker, period of origin and material, as well as more specific details such as condition and historical significance.

Metadata plays a crucial role in the digital management of collections and is the key to searching, categorising and correlating objects in a variety of ways.

Each object is assigned a record containing all this information and is stored in a museum database. This structured record makes it possible to manage and research the objects not only physically, but also digitally.

You can use various filters to expand or limit the search for colours, time periods, materials and locations. However, only colour recognition is supported by machine intelligence. The other filters are based on keywords and <li>metadata</li> from the museum database. For colour recognition, computer vision methods are used. This involves AI-based image segmentation (image decomposition) to separate the object in the image from the background. The pixels of the object are then grouped to create a colour palette for the museum object. For more information, click <li>here.</li>

However, the 'training' of the AI models is crucial for the accuracy and performance of the system. The more extensive and diverse the training data, the better unknown images can later be correctly classified. Careful selection and precise annotation of the data is therefore essential, as incorrect or inaccurate training data can falsify the results.

In order for AI models to correctly classify images, they need a solid database - and this is where training data and annotations (labels) come into play.

As commercial AI applications are often trained on the basis of data sets from the internet, which are often characterised by social and historical biases, it is important to take a critical look at the data base. The fact that much of this internet data comes from an affluent society often leads to a one-sided view of the world. The classification systems and terms used to annotate images can also contain and reinforce prejudices. In addition, the annotation and classification of data is essentially interpretive work, which is often outsourced to untrained and unfairly paid workers. There is often a lack of clear annotation standards or guidelines, resulting in highly individualised and often incomprehensible annotations. Another problem is that the biases present in existing datasets are rarely documented or verified. This is particularly critical as many AI models are pre-trained on the basis of the same data sets. Further information on this topic can be found <li>here.</li>

For example, in order to learn how to classify images, the AI models must first be trained with thousands, if not millions, of images. This is the only way to recognize and correctly classify the motifs in the images. The data, which is fed to AI models during the learning process, is referred to as training data. However, in order for the AI to 'know' what can be seen in the images, they must first be annotated. Annotations are labels that mark relevant objects or motifs in the image. This enables the model to learn what a dog, a tree or a building is.

Our intelligent, AI-supported search makes it easy for you to browse through our collections. You can search for objects or themes that interest you and discover surprising connections between museum objects. You can also search specifically for motifs (e.g. floral elements or animals) or be guided by emotional terms. Just give it a try!

What can I search for?

An intelligent search is a search that understands the context and intent behind a search query. Instead of just looking for exact matches of keywords, it also considers synonyms, related terms and the semantics (meaning) of a search term to provide more relevant results. Thanks to <li>natural language processing</li>, an intelligent search can also understand unstructured data and process queries in a natural, colloquial form. This enables the machine to answer even complex search queries precisely. The xCurator also supports a visual search, i.e. image recognition methods (<li2>computer vision</li2>) are used to analyse and sort images. The system recognises colours, motifs, objects and themes in the images and identifies similarities between them. The combination of image and text analysis is referred to as a multimodal search. It makes it possible to search and explore the contents of our collection in a diverse and efficent way, providing a comprehensive overview of tens of thousands of objects from our collections.

What is an intelligent search?

The process of identifying and linking entities in texts is known as 'Named Entity Extraction and Linking'. An entity recognition algorithm is used, which automatically recognises and categorises certain entities in the texts using natural language processing methods. his is a complex process in which machine learning methods are used to identify terms and then automatically link them to existing records.

You can read more about the process used <li>here.</li>

In the xCurator, machine intelligence helps you to search for similar objects in the collection. The function for identifying visually similar images is made possible by computer vision, a technology that analyses images pixel-by-pixel. Using methods such as image segmentation and pattern recognition, similar image content is identified and images are reorganised. This enables a comprehensive examination of the entire collection that would be almost inconceivable with the naked eye.

The xCurator uses a variety of AI technologies to support the user in discovering, understanding and creating content. Various machine learning methods are used, including <li1>image recognition</li1>, <li2>natural language processing</li2>, <li3>generative language models</li3>, <li4>intelligent search algorithms</li4>, <li5>multimodal methods</li5> and <li6>named entity extraction</li6>. In addition, xCurator uses <li7>deep learning algorithms</li7> to automatically categorise and classify objects. These algorithms analyze both visual and text-based features and assign the objects to suitable categories. This makes it easier for you to find similar objects and navigate effortlessly through our extensive collections.

You can find the technical documentation for the xCurator <li>here.</li>

This may sound complicated at first, but don't worry - we will guide you through the process step by step and explain the individual procedures as you use the xCurator.

Here you can make adjustments to the license assignment. The CC-0 license is preselected by default. You can find information on CC licenses here: https://de.creativecommons.net/was-ist-cc/

Once you have saved a few objects in your digital exhibition, the ‘Edit exhibition’ function gives you the opportunity to expand and refine your exhibition. The 'Recommended' section automatically suggests more objects that match the theme of your exhibition, based on your previous searches and title details. Just like on the canvas, this function is made possible by <li>computer vision</li>. The properties oof the selected objects are analysed and compared with similar or complementary objects in the collection. This allows you to expand your exhibition in a targeted manner and discover new, exciting connections between the museum objects.

Edit exhibition

Using Generative Language Models

Choose your prompts

Edit title

What does 'AI for the common' good look like ?

Still not quite sure how artificial intelligence processes work and what an algorithm is? Then take a look at the <li>AI-Campus</li>, the learning platform for artificial intelligence. Here you will find lots of easy-to-understand learning materials on the topic. You can attend online courses, watch videos or listen to podcasts according to your preferences. Let's go

The Harvard Art Museum provides an exciting example with the AI Explorer. It analyses how various computer vision systems from commercial providers see and classify their collection. Find out more <li>here.</li>

<li>Algorithm-Watch</li> is a non-profit, non-governmental organisation based in Berlin and Zurich. It is committed to ensuring that algorithms and artificial intelligence (AI) strengthen justice, democracy, human rights and sustainability instead of weakening them.

Want to know more about machine intelligence? There is so much more to report and the technological developments in the field are rapid, which is why we cannot claim to be exhaustive here. However, if you want to find out more, here are some interesting links and resources to help you.

You can find an interesting article on the topic <li>here.</li>

What if we will soon no longer be able to tell which texts were produced with AI and which were not?

And if you prefer to consume information in the form of podcasts, <li>here's</li> something for you.

Wanna know more about the carbon footprint of AI?

The <li>'EU AI-Act'</li>, a European regulation governing the use of artificial intelligence, was adopted in May 2024. It is the first comprehensive regulation of AI by a major supervisory authority worldwide.

The Recoding Art project by the two artists Bruno Moreschi and Gabriel Pereira in cooperation with the Van Abbemuseum investigates how machines perceive and evaluate art. Interested? Then watch the video documentation of the project <li>here.</li>

What is the legal basis for the use of AI?

The xCurator provides you with a wealth of information, some of which is mixed with AI-generated content. Machine intelligence is based on mathematical principles, logic and probabilities, whereas communicating cultural communication takes into account meanings and perspectives. That's why it's important to always check all information critically. Carry out a fact check to ensure that the data provided is reliable. Compare the AI-generated content with other trustworthy sources (e.g. object texts) and scrutinise any discrepancies. Your own critical assessment is essential.

Would you like to delve deeper into these topics? Our latest Deep Dive article provides further information and helpful links.

However, responsibility in dealing with AI goes beyond mere fact-checking. In order to use AI technologies responsibly, it is important to develop a comprehensive understanding of the underlying technology (see Deep Dive), its legal framework and ethical issues (see FAQ). The emergence of biases and prejudices in the use of AI should also be consciously reflected upon. A previously often neglected but important topic also concerns the environmental impact caused by the training and operation of such technologies.

You can find out more about generative language models and prompting <li>here</li>.

In xCurator we use the language model solution from OpenAI, known as ChatGPT. We chose it because Large Language Models (LLMs) like ChatGPT are shaping our present and offer enormous potential. At the same time, it is important to deal with the risks and opportunities of these AI models in a protected environment.

In xCurator, you can use this technology to create content on a secure fact base (our museum database). The aim is for you to learn how to use such models by developing your prompting and fact-checking skills. These skills are becoming increasingly important as language models such as ChatGPT are increasingly used in different areas of our lives.

Large language models are trained on huge amounts of text in order to understand and respond appropriately to human speech. They are based on artificial neural networks, which, during the training, develop statistical models that predict the likelihood of a particular combination of words, phrases or text fragments occurring. They can generate text that is often surprisingly well phrased and contextualised. They use the patterns learned during training to generate appropriate responses or texts from your input - the prompts. We often encounter large language models in the form of chatbots or autonomous agents that assist us on company websites.

The xCurator provides you with a large generative language model (LLM) to supports you in creating texts and discovering new contexts. The model acts as a creative co-pilot, giving you suggestions for object descriptions, introductions and conclusions. To do this, it draws on the extensive collection data to provide you with truly relevant and exciting content. To ensure that the AI-generated text is as accurate as possible, we have worked intensively on the database. This reduces the risk of the model 'hallucinating' content - i.e. providing inaccurate or irrelevant information. For security reasons, we have pre-set the prompts for you. However, the results of the generative language model should always be evaluated critically. This is because AI only calculates statistical probabilities and is not orientated towards objective truth or the real world. Therefore, use the existing museum texts to check the AI-generated content and ensure that it corresponds to your requirements and the reality.

Natural Language Processing (NLP) is an area of artificial intelligence that enables computers to understand, interpret and respond to human language. NLP includes various techniques that enable systems to analyse and categorise text and spoken language and extract meaningful information from it. Applications of natural language processing can be found in everyday technologies such as voice assistants, automatic translation services and text analysis tools. The goal of NLP is to make the interaction between humans and machines as natural and intuitive as possible.

Die Verarbeitung natürlicher Sprache (Natural Language Processing, NLP) ist ein Bereich der maschinellen Intelligenz, der es Computern ermöglicht, menschliche Sprache zu verstehen, zu analysieren und darauf zu reagieren. Durch verschiedene Techniken können Systeme Texte und gesprochene Sprache erfassen, strukturieren und daraus relevante Informationen gewinnen. NLP kommt in zahlreichen Alltagsanwendungen zum Einsatz, wie z. B. bei Sprachassistenten, automatischen Übersetzungen oder Tools zur Textanalyse. Das Hauptziel von NLP ist es, die Kommunikation zwischen Mensch und Maschine so natürlich und intuitiv wie möglich zu gestalten.

In the field of AI, prompts are the inputs given to a generative language model such as ChatGPT to get a response or text. They are essentially the instructions that tell the model what kind of content to generate. A prompt can be a question, an instruction or simply a few keywords - anything that helps the model understand the context and formulate an appropriate response.

Choose one of the various prompt templates to test the underlying language model. Note how subtle differences in the prompt affect the output text. Try it out and compare the results!

The influence of prompts on the language model is enormous. The quality and clarity of the prompt determine how well the model performs. A precise and well-formulated prompt usually leads to a precise and relevant response. The ability to formulate precise instructions to generative AI models is known as prompt engineering. Incidentally, prompt engineering has developed into a professional field in its own.

Edit story

Log in and save your favorite objects from the digital collections of the Badisches Landesmuseum and Allard Pierson.

Favorites

Favourites

With the xCurator, you can immerse yourself in the collections of the Badisches Landesmuseum Karlsruhe and Allard Pierson Amsterdam.

Start

Your AI co-pilot. Dive into collections, discover unexpected connections and design your exhibition.

In the case of AI-generated content on our portal, it is generally permitted to copy, print and distribute the texts and documents attached to them. AI-generated content without human contribution is currently in the public domain (12/2023). If they are protected by copyright, they can be used and distributed under the free license specified in each case, subject to the conditions.

Privacy policy of the Badisches Landesmuseum.

Declaration on the accessibility of the Badisches Landesmuseum.

The license for the picture is in the data record.

Realization

Impressum

Adjust your preferences selected in onboarding here.

Onboarding

Preferences

Create your own story with the xCurator now or continue this one!

Here you can view the story in presentation mode.

Find this story inappropriate? You can report it to our admins here.

Presentation

Do you really want to delete your account?

Attention: All your stories and favorites will be deleted and cannot be restored.

Here you can customize your personal details such as name and email address.

Username

Profile settings

Take a look at the stories of other xCurator users.

Under 'Create' you can easily put together your own small exhibition. Browse the extensive collections for exciting topics and choose the objects that fascinate you the most. For the objects you choose, you can write your own text and add it to your exhibition. Or you can use machine intelligence in the form of a <li>generative language model</li> to help you write.

You can choose to keep your exhibition private or make it public. See 'All exhibitions' for an overview of the exhibitions that have already been published. You can easily share your exhibition with others and pass on your discoveries.

The arrangement of the images on the canvas is created by a special algorithm. This algorithm converts a list of images into a visual representation. A <li>self-organising map (SOM)</li> is used for this purpose, which sorts and arranges the images according to similarity. A self-organising map (SOM) is an artificial neural network that is inspired by the way the human brain works. In particular, the way in which sensory stimuli from neighbouring sensory organs are processed in adjacent areas.

A self-organising map is implemented in xCurator to cope with the flood of images in our collections and to arrange the images in a two-dimensional grid. Detailed information about the algorithm in xCurator can be found <li>here.</li>

The main mechanism of the SOM is to project high-dimensional input data onto a low-dimensional map (usually 2D). This projection is done in such a way that similar input data is located in neighbouring areas of the map in the output space. This allows complex data sets to be simplified and relationships to be visualised. During the learning process, the map adapts dynamically by recognising the patterns and similarities in the input data and reorganising itself. Typical applications for self-organising maps are classification, visualisation and clustering of high-dimensional data and data mining.

Create your own story in xCurator and use generative AI technologies such as the large language model ChatGPT.

All stories

My Stories

You have the option to publish your posts on xCurator. Please always remember that your posts are publicly accessible as soon as you set them to 'Public'.

You are the author of your story. If you decide to publish, the stories and content will be available to the general public for free use and distribution. For this purpose, the license is set to Creative Commons CC0 by default. This means that the story can be used freely and without conditions by the general public as open content. You can also choose a different Creative Commons license for a story yourself. 

My stories

All stories