The blue wins
2020 is a phenomenal year.
Covid-19 sweeps out the world. Economy is trembling. The US finally has a new President (Hooray!). And natural disasters such as wildfire, flood, super storms and hurricanes break all historical records about intensity, severity and frequencies. Yes, I am talking about climate change which is ironically known only by daily weather news.
It’s obvious to me that the physical environment and personal health has an interconnection. Let’s talk about skin care - one of the most active communities I have ever been. The message is very persistent with three key learning:
- No good product. No bad product. It is about experiential learning.
- True beauty comes from the inside with a strong foundation of personal health.
- Environment matters a lot to sustain a healthy and shinning look.
But building a technology for skin does not sound as easy like that. One of the most critical challenges is privacy. How hard is it? Well, a few days ago I was contacted by a senior medical advisor with specialization in business and law from a medical school. He invited me to attend a webinar of a newly-published review on Nature in a similar topic.
I was surprised by his outreach until reading a section called social and ethical considerations with four main points to sum up a concept of a trustworthy AI system (Here I am just highlighting privacy.)
- Privacy: Confidentiality, sharing and retention of data through a collection privacy-preserving process with a loop with all stakeholders, e.g. unauthorization access and de-identify. A list by 6 known computation methods such as differential privacy, face blurring, dimensionality reduction, body masking, federated learning and homomorphic encryption (Image)
- Fairness is reviewed by two aspects: dataset bias and model performance.
- Transparency is about interpretation, e.g. prediction, description and relevance.
- Research Ethics is about human protection, independent review and public benefit.
As an AI technologist with a background in privacy and security research, I still find it very difficult to find a good solution to meet a long list of requirements and regulations. The common direction now is solved by business development and traditional methodologies like homomorphic encryption. Federated learning, a new promising technology, is still at the early stage with a risk of reversing models and leaking information. Nonetheless, there is a lot of nuance when talking about privacy issues.
Personally I do agree that we need a framework to progressively accelerate innovation in healthcare between all stakeholders. At least, covid-19 is a wake up call for an entire global community about prevention, protection and early treatment, especially at the very first level of awareness, evidence-based communication.
Tagged #Nature, #summary, #ai.