I recently attended a panel talk hosted by Big Motive’s Rebecca Walsh as part of Belfast Tech Week. The panel discussed the role a AI in research in relation to health. One point I found particularly interesting was the idea of “Synthetic Data”. These are data sets that either create or expanded by AI. While these can be useful during the early stages of projects, there has been a rise on the over-reliance on them. As someone in university, AI can be a great tool for replacing parts of projects that wouldn’t be possible in the setting, but I think the ease we have to ask AI for things could become an issue. As designers, we are designing for humans, and over-relying on AI to fill gaps will just detach us further from our users. I think while it can be a useful tool to expand data, the idea of creating data through AI is fraught with trouble.
What is synthetic data? - MOSTLY AI
Another interesting idea raised was the concept of “air-gapping” AI. Air-Gapping is the process of using something completely offline without connection to the internet. In research we handle large amounts of personal data, and when we process that data using online AI we may not be following GDPR guidelines. Air-gapping AI would essentially let you run the AI models locally, ensuring data doesn’t leave your computer. With a lot of popular AI models being cloud based, this can be difficult. New tools like Ollama, an open source AI model that can be downloaded a ran locally. This fixes an initial issue of ensuring data doesn’t leave machine, but with the number of people who would need to use something like this to follow GDPR, it’s not a given that their machines will be powerful enough to run an AI model locally. I could imagine that this will lead to new legislation soon, as we are still in the infancy of AI.
How developers are using offline AI tools for air-gapped security
Overall this was a really interesting talk that introduced me to some new ideas that I wouldn’t have normally been exposed to. This will let me more carefully consider when I get my data, and how I process it with regards to AI.