Artificial Intelligence Gets Real

Elon Musk’s latest brainchild, Neuralink Corp., promises to speak—and think—for itself. The California-based startup wants to use artificial intelligence (AI) to help treat disorders such as epilepsy and depression through a concept known as “neural lace,” which would implant tiny electrodes into a person’s brain to enhance memory and thought capabilities. While the technology is described as being in an embryonic stage without any guarantee of success, it underscores AI’s growing importance and incredible potential.  

Having long been used in computer gaming, facial recognition and internet search engines, AI is now becoming increasingly sophisticated and spreading to complex finance, medical, manufacturing and military applications as well as smartphone apps, such as Apple’s Siri virtual assistant. So-called deep learning, a branch of AI that specializes in teaching machines to learn by themselves, is also transforming the capability of speech recognition, computer vision and analytical algorithms used in autonomous vehicle technologies. 

Carmakers and traditional suppliers are rushing to partner with tech firms to develop these capabilities, with much of the leadership to date coming from Japan and Silicon Valley. In late 2015, Toyota created a new subsidiary—Toyota Research Institute (TRI)—to develop artificial intelligence and advanced robotic systems. Working with researchers at Stanford, MIT and the University of Michigan, TRI has been allotted a $1 billion budget through early next decade. The carmaker also is using its new Concept-i car to help in-vehicle AI systems learn about driving patterns, schedules and frequent destinations. 

To keep pace, Honda recently formed a new Center X research and development unit in Tokyo. Researchers there will initially focus on AI systems that work cooperatively with humans, including technologies designed to understand and relate to emotions. And BMW is working with IBM to integrate the software company’s Watson artificial intelligence technology into future cars. This includes uploading vehicle owner’s manuals to IBM’s cloud platform and Watson, which would allow motorists to ask questions and receive answers about their vehicles in a conversational style while driving. 

Denso and Toshiba, meanwhile, are working together on an AI technology to help self-driving cars more quickly identify and analyze images, including pedestrians, road markings and environmental conditions. The partners expect the resulting deep neural network algorithm to significantly improve accuracy by enabling vehicle computers to extract and learn the characteristics of various objects as they are identified rather simply comparing images against a preloaded database. 

Chipmaker NVIDIA’s Drive PX 2 in-vehicle supercomputer uses AI and deep learning to quickly teach itself safe driving techniques from simulations, millions of miles of on-the-road videos, real-world testing and input from engineers and professional drivers. With the ability to process 24-trillion operations per second, the system continually gathers and integrates information from dozens of sensors while the vehicle is driving to monitor its surroundings and handle the nearly infinite number of variables it can encounter.

In addition to being the brains behind autonomous vehicles, next-generation AI technologies are expected to yield significant improvements in the quality, efficiency and cost of manufacturing—including providing human operators better and more information, automating menial and safety-critical operations, and optimizing machine maintenance. 

The technology is advancing rapidly. Whereas early AI systems relied on software engineers to code in strict rules for computers to follow, deep learning enables the machines to do the work themselves. Using a variety of sensors, historical data and other inputs, computers write and update the algorithms that allow for advanced problem solving far beyond human capabilities. It’s to the point now where human programmers don’t even understand how or why AI-based systems make the decisions they do. 

This could be a big problem if something goes wrong that leads to a car crash or other foul-ups that can’t be explained. As a result, companies are working to make the AI decision-making process more transparent through reverse engineering and highlighting how information is prioritized. The European Union may require this type of capability for future AI applications. Whatever the answer, it’s sure to be a (deep) learning experience.