How can we do something with AI that is note evil?
- Plenty of people working in GLAM are doing thoughtful things with machine learning.
- People working in GLAM used targeted questions and problems with their AI/machine learning.
- Dr Graham has used machine learning and image recognition to track the changes in the international space station. National WWII Museum (New Orleans, LA, USA)
- Voices from the front: a tech deployed in this museum using AI.
- The visitor can interact with a AI model of a soldier from WWII.
- Built on several interviews with actual WWII soldiers.
- Enables people to 'talk to' WWII vets.
- Could have been done with retrieval augmented interaction. In reality much simpler than that.
- Just a keyword search based on text to speech and a bunch of videos.
- a 'corporate washing'. Problematic to give this company videos about your personal life. Generative adversarial networks
- two neural networks
- One is an image (the forger) net which has been fed several images, this creates a bunch of random noise. The second network determines how close the image produced in similar to the prompt image-text-pair. Done again and again until the network learns to create images closest to the prompt.
- One has been training to know what 'pottery is' for example. Curation of metadata
- Was more prominent before.
- With LLMs like ChatGPT the data is less curated than it probably should be. Conservative
- Models are generally conservative as they do not create new things but only draw on its training and dataset. List of initiatives in museums to incorporate AI.
- bit.ly/old-museum-ai
- themuseumsai.network/toolkit
- the latter link contains a worksheet useful for project planning in the future.
- Project legacy in one of the questions, important to think about what will happen to this data after. Ethics
- what happens when the code dies, what happens when the project is over, these are important to think about Free AI options for code help
- Claude.ai
- Groq.com
- Jan.ai Going forward
- Look at the worksheet
- Determine the steps
- Keep the goal in mind
- Be incredibly pointed with the prompts. Break it down step by step with very specific requests.