Automakers Take the Onramp to Machine Learning Strategies
Automaker experiments with vehicle tech offer established businesses lessons for managing machine learning and DevOps influences on products and operations.
I remember an episode of Jerry Seinfeld’s Comedians in Cars Getting Coffee complaining about the heat and ventilation controls in a Triumph TR6, saying “You couldn’t make it more confusing if you wanted to.”
Jerry may also be speaking about the tech influence on today’s cars. Industries are seeing their products and services evolve by software, and the automotive industry, facing the rise of autonomous transportation on the near horizon, is not immune.
The marketplace lessons the auto industry is discovering are no joke for other industries seeking machine learning opportunities. Companies with multi-featured products offer examples to learn how to manage the education process. The automotive industry has the inherent potential to yield excellent ideas for firms confused on machine learning results or baffled at how to balance data between statistics and unstructured business concerns.
Think of software as an engineered product rather than text meant for a server, and you can reimagine the features that are in a dataset. An engineered product like a car has a lot of features, but focus on particular features depends on the business unit involved. An advanced engineering team may be concerned with functionality, while a product engineering team would be concerned with how a part fit within a system of parts. This is what I learned as an engineer years ago.
The user interface for entertainment systems offers a clue as to how distractions can be factors. Automakers are competing with each other based on the tech auto buyers want. Automakers have had a few years of interfaces in car interiors, so each brand has learned which controls should be a traditional design and which are enhanced with newer interfaces. For example, Honda reintroduced a radio knob in the latest Accord and CRV, its most popular vehicle, after auto journalists criticized the lack of a physical switch in earlier models. Cadillac also was criticized for using capacitive switches instead of standard radio and A/C controls.
But others have found success. Audi received industry acclaim when it launched Human Machine Interface (HMI), a virtual cockpit in which a high-resolution screen can show several display configurations for vehicle operation, navigation, and entertainment. BMW recently introduced the most intriguing cockpit feature among the automakers, a series of gesture controls where the driver controls radio and air conditioning selection based on hand movement.
All of these examples recalibrate the understanding of driver interactions with a vehicle as a gateway to machine learning. Vehicle controls provide inputs noting intentions and reactions to vehicle and environment condition. Those inputs influence key statistical testing frameworks, specifically Type I and Type II errors.
Type I (False negative) errors are essentially an answer that states something is true when it really is not. In contrast, Type II (False positive) errors are a condition of not recognizing when something exists. In other words there is no alert for a condition that is considered a failure. These frameworks make up the accuracy and precision of a model.
Statistical testing frameworks are a valuable decision-making aid. The aid can now describe a null hypothesis with respect to vehicle dynamics and conditions to which the driver and passengers will act. So imagine the BMW hand gesture – what gestures according to the vehicle are considered noise, a Type I error, or a Type II error? Both Type I and Type II are useful because they eliminate blanket comments as valid statements of a condition, using accuracy and precision for guidance. Testers will have to think about how to classify reactions or observed instances with accuracy and precision in mind.
The opportunity implies the strongest benefit that automobiles bring to the technology space. Vehicles are literally rolling beds of sensors that can provide metrics for understanding human behavior and activity associated with feature functionality. The metrics can help brands develop machine learning models, be it for predictive features in a current vehicle or for an autonomous vehicle. In either case, the data can teach an algorithm vehicle conditions and potential factors from people and objects along a vehicle’s path. The results can give marketers a better understanding of how vehicle controls link to the type of vehicle experiences that raise questions to be solved for the benefits of customers.
The promise is sky high, but so are the present challenges, especially with driverless vehicles on the horizon. Firms ranging from traditional automakers to tech startups are experimenting with how sensors recognize roadside objects that move while at speed. Infrastructure like 5G will be essential for fast software updates and, consequently, fast data to feed machine learning where applied. Thus the challenge of coordinating real world activity with quickly needed decisions.
Scott Brinker, chairman of the MarTech Conference stated in a podcast interview that “It’s hard to find a business these days that isn’t becoming a software company.”
His perspective on software’s influence is spot on as industries are seeing their products and services evolve through software. By imagining dataset features as an engineered product like a car, analysts can better plan for potential results from the new world of driverless vehicles and machine learning.
Pierre DeBois is the founder of Zimana, a small business analytics consultancy that reviews data from Web analytics and social media dashboard solutions, then provides recommendations and Web development action that improves marketing strategy and business profitability. He ... View Full Bio
We welcome your comments on this topic on our social media channels, or [contact us directly] with questions about the site.