Why we can't move fast and break things with autonomous vehicles

The prevailing approach to innovation across the tech ecosystem for almost two decades has been guided by one rule, which is to move fast and break things. The now famous saying from Facebook creator Mark Zuckerberg has undoubtedly proved successful for many businesses, however, for autonomous vehicles such an approach creates more problems than solutions.

Ultimately, it's hard to overstate how big of a challenge achieving autonomous mobility is, and much of the excitement around a driverless revolution has led to disillusionment in some quarters. The expectation this would be the next tech revolution has crashed because of the difficult reality of autonomous mobility.

But this isn't the end of the conversation. High-performing, safe and trusted autonomous vehicles can be created, but there are no shortcuts. Success is possible if the tech and automotive sectors collaborate.

There needs to be iterative approach that effectively processes the vast amount of data needed to power the machine learning algorithms as well as a collaborative mindset across the industry can get there.

The move fast and break things mindset, which puts a premium on coming first over anything else, will only lead us down a developmental dead end. With safety critical applications, this approach can have disastrous consequences as there is zero margin for error.

What does good look like?

Describing the task of achieving autonomous mobility as difficult is an understatement. It requires huge amounts of data that needs to be curated, annotated and refined to train machine learning models, which provide the brains of the AI “running in the vehicle”. These data streams are ongoing and dynamic -- they always change as operational domains expand.

Secondly, the task of driving is highly subjective. If we were to conduct a survey and ask people to define what a road or driveable surface is, we would get vastly different answers. In some cases, gravel is considered a drivable surface and in other cases not. Navigating such nuance and complexity with the machine learning models, which are by their nature a blunt instrument, poses significant difficulties. How can we expect a machine to get it right if we can't tell it what's right?

The goal is not perfection. The process will require the industry to learn from it mistakes but it's vital to understand that the development of autonomous vehicles will be incremental. Progress has been made. Ford's BlueCruise technology gained approval from the UK's Department of Transport.

However, the road is long to achieve fully autonomous vehicles.

The development of autonomous vehicles will require collaboration between humans and the technology itself. Ultimately, humans will decide how this technology is used, and this collaboration will enable a flywheel of automation. This approach is not just in contrast to Silicon Valley, but also in some cases large automakers, many of which are still relying on a traditional waterfall methodology that is not suitable for managing data within machine learning efforts.
 

Industrywide standardization

The industry is also crying out for standardization and regulation. This will not only ensure the industry delivers on the promise of autonomous driving but will aid consumer buy-in and confidence.

Without backing and trust from the consumers, who will ultimately be the users of this technology, it will be impossible for autonomous vehicles to realize their full potential.

There is a bright future for autonomous vehicles and mobility, but it won't be an overnight success. There is a long road of development ahead, and it that requires a different approach from those commonly adopted in both the tech and the automotive industry. However, there are undoubtedly significant benefits that can be unlocked through the development of this technology.