Shared my thoughts on Trusted AI in a LinkedIn article.
When talking about Trusted AI, people implicitly assume that AI tends to be anti-human without a certain penalty provided by humans.
I have a different thought. Machine learning as the core technology of AI is merely a mathematical technique to capture certain patterns hidden inside data. It should be neutral in nature to gender, race, sexual orientation, or any attributes of the samples. If so, we can think of another research direction for Trusted AI: AI as a tool for democratizing infrastructures.
“Frugal” Innovations for Africa
A project named “Frugal Innovations for Africa” is one of the projects I led before based on such a concept. The project’s immediate goal is to help Nairobi City, Kenya, develop a city-wide traffic monitoring system. The catchy word “frugal innovation” was something Prof. Robert Morris, the former head of Global Labs, IBM Research, brought to us.
Traffic jams were one of the most serious problems of the infrastructure of Nairobi City. As the first step towards a solution, they were interested in developing a city-wide traffic monitoring system. Typically, such a system needs specialized hardware such as induction loop detectors embedded under the road and purpose-built communication lines. Obviously, it requires a massive investment in addition to big maintenance costs to keep the system running. It is not a viable solution in most African cities.
The chart below is what I, as the project lead, always used when introducing the project.
The challenge we took on is something like this: Can we replace a multi-million dollars intelligent transportation system with a cheap web camera network?
The idea is to leverage images of internet web cameras to precisely estimate the city’s traffic flow. As a machine learning problem, it was to precisely estimate the traffic flow on a road network in which extremely low-quality image data is provided at a limited number of graph edges.
The team developed really cool machine learning algorithms as described in:
T. Idé, T. Katsuki, T. Morimura, and R. Morris, “City-Wide Traffic Flow Estimation from Limited Number of Low Quality Cameras” IEEE Transactions on Intelligent Transportation Systems, 18 (2017) 950-959 [link, slides for related paper].
Liberate AI for Social Good
In some sense, the project was an attempt to make expensive infrastructures, which have been exclusively used by the rich nations, available for developing countries with the aid of AI. It is a democratization of infrastructures to help people in need. Let’s liberate AI, instead of penalize it, to put it to work for social good.