GeologyOracle是一款免费的AI助手，专门针对地质科学领域，旨在帮助地质研究人员、学生和地质旅游者。它可以通过文本、图像或音频等形式接收用户输入，并实时生成文本回复。该系统基于GPT-4，但其性能是否优于原生GPT-4尚不明确。文章还探讨了GeologyOracle在数据来源透明度、用户数据隐私和评估工具的伦理问题等方面的一些潜在问题，并呼吁公开发布训练数据、基准测试问题和代码，以促进地质科学领域的开放科学。

🤔 **GeologyOracle是一款免费的AI助手，专门针对地质科学领域，包括地质研究、教学和地质旅游。**它通过模拟对话的方式，接收用户输入并提供实时文本回复，支持文本、图像和音频等多种输入形式。

⚠️ **GeologyOracle的回答没有引用来源，这对于科学AI工具来说是个问题，难以验证答案的准确性和识别作者。**这与联合国教科文组织关于人工智能伦理的建议相悖，也可能导致知识产权问题。

🔒 **GeologyOracle基于GPT-4，用户上传的文本和图像可能会被OpenAI用于训练数据，这可能导致用户数据隐私问题。**文章建议用户不要上传机密内容，并呼吁GeologyOracle明确说明其数据使用政策。

🤔 **GeologyOracle提供的“评估地质科学文本”工具可能存在伦理争议，其评分标准和使用场景需要进一步讨论。**文章认为该工具的推广需要谨慎，避免对教学和学习产生负面影响。

🔎 **GeologyOracle的性能是否优于原生GPT-4尚不明确。**初步测试表明，GeologyOracle在某些方面并未表现出显著优势，这需要进一步的研究来验证。

GeologyOracle has been released today, free to use. “It is the first AI assistant specifically trained for geoscience. A typical GeologyOracle session resembles an instant messaging chat, i.e., the human user inputs information, which can be in the form of text, images, or audio, and the AI system simulates the conversation by generating a real-time text in response to the user input.”. Link here: https://geologyoracle.com/

I congratulate the university researchers on their noble intent and hard work to create this system to support geoscience researchers, students and geotourism to support UNESCO Geoparks. They have some genuinely innovative ideas and it is a useful contribution to this area. There is no registration process, so no personal details are captured which is good.

I reached out to the authors and am awaiting details on whether they intend to openly release under FAIR principles (1) the earth science “training” data they used to configure the system, (2) the 152 questions used to benchmark the performance of the system, and (3) open-source the code/prompts written on top of GPT-4o. These would support open-science for the whole geological community.

I have some ethical concerns on this first release, which may be supported by UNESCO Geoparks, which targets geoscientists. These include:

(1) None of the answers cite their source. For a scientific AI tool this is problematic, providing no easy way for answers to be verified or authors to be recognised. This would also go against UNESCO’s own recommendations on AI Ethics. Considering this is in part likely a Retrieval Augmented Generation (RAG) system, it should be easy to do this.

(2) The system uses GPT-4. Nowhere does it state on the website that the text and images you upload may be used by OpenAI for training data which could find their way into commercial models. So confidential content should not be uploaded, which may not be an issue for geotourism, but might be for other geoscientists. Geoscientists should be made aware.

(3) Of the 6 tools provided, I feel the ‘Grade a geoscience text’ tool for teachers and students that provides a score on a scale of 1 to 10 for your work, may not be something that is necessarily ‘good’ to be promoted in UNESCO’s name. One for ethical debate.

From a performance point of view, I compared answers, both text and images, to results from using the generic GPT-4o app. My quick tests appear to indicate tGeologyOracle did not perform as simply using native GPT-4o (which has no additional special geological training). So it is unclear to what extent GeologyOracle improves on what is already available in terms of AI/Large Language Models (Including vision). This was not tested in the paper either, perhaps an area for further research.