Page 36 - EE Times Europe November 2021 final

36 EE|Times EUROPE

AI’s Impact on the Current and Future Automotive Industry

A lack of AI expertise is another big drawback in the auto and other AI REGULATION
industries, and the skills gap is not likely to be remedied anytime soon. AI and the General Data Protection Regulation (GDPR) are closely tied.
The problem-solving inference phase also has drawbacks. Large GDPR affects AI development in Europe and other regions. The regulation
models, especially for AVs, require tremendous computing resources to explicitly covers automated, individual decision-making and profiling.
crunch sensor data and support complex software. Those resources also The rule protects consumers from the legal consequences of both. Auto-
require power, which is always limited in auto applications. mated, individual decision-making in this case includes decisions made
Emerging technologies will improve capabilities and reduce infer- by AI platforms without any human intervention. Profiling means the
encing costs. They include new AI chip technology, lower-cost LiDAR, automated processing of personal data to evaluate individuals.
and sensors with increased performance. For automotive applications, this primarily affects content delivery
The biggest drawback for inferencing is the black-box problem, or AI systems and user interfaces.
explainability. AI systems remain unable to explain how they arrive at The European Union is preparing an AI regulation that would be sim-
decisions, creating a host of AI trust issues. For automotive applica- ilar to GDPR and would likely have as broad an impact. A draft proposal
tions, that’s a non-starter. representing a legal framework for regulating AI was released in April.
The EU proposal seeks to identify high-risk AI technology and its
AI SAFETY applications aimed at critical infrastructure such as transportation
Automotive AI requires much greater safety than other consumer seg- that could endanger citizens. This means AVs will be a target of AI
ments. Hence, greater emphasis on AI safety and R&D is a must. To that regulation.
end, Georgetown University’s Center for Security and Emerging Technol- Fines under the EU-proposed AI legislation could run as high as
ogy (CSET) has released a pioneering report (bit.ly/3EX8Six) examining €30 million, or 6% of a company’s global revenue, whichever is higher.
the unintended consequences of AI and the potential impact. Maximum fines under GDPR are €20 million, or 4% of global revenue.
The CSET report identifies three basic types of AI failures: robust-
ness, specification, and assurance failures. Robustness failure means AUTOMOTIVE AI
AI systems receive abnormal or unexpected inputs that cause them The table below summarizes AI technology integrated with auto
to malfunction. In specification failure, the AI system is trying to electronics. Not included are AI used in auto manufacturing, supply
achieve something subtly different from what the designer intended, chain management, quality control, marketing, and similar functions in
leading to unexpected behaviors or side effects. Assurance failure which AI is making significant contributions.
means the AI system cannot be adequately monitored or controlled Decisions generated by neural networks must be understandable. If
during operation. not, it is hard to comprehend how they work and correct errors or bias.
The report, released in July, includes examples of what unintended Neural network decisions also must be stable — that is, remain
AI crashes could look like (the authors prefer the term “accident”) and unchanged despite minor differences in visual data. This is especially
recommends actions to reduce the risks while making AI tools more important for AVs. Small strips of black and white tape on stop signs
trustworthy. can make them invisible to AI-based vision systems. That’s an example
Explainable AI (XAI) is a method for mitigating the black-box effect, of unacceptable neural network performance.
allowing better understanding of which data is required to enhance model AV applications require better technology to understand edge cases or
accuracy. XAI research sponsored by the Defense Advanced Research new driving events not experienced by previous software driver training.
Projects Agency (Darpa) seeks to develop machine-learning technologies This remains a key limiting factor for deploying AV systems in volume.
that produce more explainable models, while retaining a high level of
learning performance and accuracy. XAI would also enable human users CURRENT AI USES
to understand, trust, and manage AI models. XAI can also characterize its Speech recognition and user interfaces have been the most successful
own abilities and provide insights into its future behavior. AI-based applications in automotive. These applications leverage AI

AI Technology in Automotive
Topic Key Information Other Information
• Understandable neural network decisions • Explainable AI is required for safety
Auto AI needs • Neural network decisions must be stable • Impervious to hacked visual input data
• Learn to handle AV edge cases • Untrained driving events for AVs
• Speech recognition and user interfaces • Alexa, CarPlay, Android Auto
• Remote diagnostics service data • AI technology turns diagnostics to prognostics
Current AI use • Vision recognition • Driver-monitoring systems
• AI-based ADAS: L1 and L2 • ACC, BSD, FCW, LDW, LKA, PA, others
• Limited driving pilots (L2+) and L3 AVs • They should not be called autopilots
• OTA software update software platforms • OTA clients, SaaS, and cloud analytics
Emerging AI use • Automotive cybersecurity software platforms • Cybersecurity clients, SaaS, and cloud analytics

• Developing and testing AV use cases • Sensor fusion, vision system, software driver
• Deployment of AV use cases: software driver • Most complex AI development ever
Future AI use • Minimize software bugs in code development • Identify and correct software errors
• Expand and improve AI-based cybersecurity • Required for all auto software platforms
(Source: Egil Juliussen, August 2021)

NOVEMBER 2021 | www.eetimes.eu