Brute Force vs. Smart Design: The Choice Between Raw Computational Power and Care in Mobility AI

Original article can be found here (source): Artificial Intelligence on Medium

Brute Force vs. Smart Design: An Important Choice Between Raw Computational Power and Care in Mobility AI

When designing mobility technology — or anything else — we’re faced with many decisions. One decision is this: Do we prioritize raw computational power to process as much data as possible or Artificial Intelligence (AI) design to be smart about what data to collect and process in the first place?

Unfortunately, many designers and data scientists start with the assumption that the more data, the better. They often lack a background in social science and aren’t able to hone in on what data (or patterns within) are most important. So, they design technology solutions that enable massive amounts of data collection. Typically, this requires a lot of raw computational power.

Mobility AI requires much more design than brute force data mining to make sure it effectively meets social needs.

However, this oftentimes isn’t the best way to go about designing data approaches. It’s more important to collect, access, and utilize the right data than it is to collect the most data. Data scientists should begin by asking: what is the shortest sustainable path for us to deliver the best solution to a problem? From there, they can prioritize what data to collect and design an AI solution that does not require networks with immense bandwidth to transfer the data into the cloud and back but can run algorithms on modest resources, already available within Internet of Things devices and Connected Vehicles.

Data collection for the sake of data collection is not going to solve our most pressing social problems. We need to be smart about data, about AI and about limited computational and network resources we may still have, especially in developing countries.

Of course, this approach isn’t appropriate for all circumstances. Sometimes, researchers do not know what data to prioritize. In those cases, it really is helpful to collect massive amounts of data first. Later on, in the solution development process, we can closely examine the most important data and then device an algorithm that will be significantly more efficient limiting the data requirements at collection phase.

Among others, NYU’s Center for Urban Science and Progress is conducting important research that takes the right approach to design. CUSP is collecting data in New York City in the hopes of developing solutions that solve major problems. For example, its Sounds of New York City (SONYC) initiative is taking on the problem of urban noise pollution. CUSP is collecting these data with the intent to design technological solutions that reduce it.

We need to see more of this kind of thinking — data collection with the aim of solving a particular problem. This thinking will help us design algorithms that have a very clear idea in mind rather than brute force data mining to derive patterns that may have little meaning in real life.

Data collection for the sake of data collection is not going to solve our most pressing social problems. We need to be smart about data, about AI and about limited computational and network resources we may still have, especially in developing countries.