Page 35 - EE Times Europe November 2021 final

EE|Times EUROPE 35

SPECIAL REPORT: ARTIFICIAL INTELLIGENCE
AI’s Impact on the Current and Future

Automotive Industry

By Egil Juliussen

rtificial intelligence is a misnomer. AI is neither artificial nor in advancing auto technology. The promise of deploying autonomous
intelligent. The implication is that AI is analogous to human vehicles (AVs) is primarily dependent on new AI technology. There
intelligence, but AI requires extensive human training to seems to be near consensus that neural network advances are the lead-
A function, and it exhibits completely different logic from ing approach for reaching future AV deployment success.
humans in terms of recognizing, understanding, and classifying objects The good news is that AI, and especially neural network technology,
or scenes. AI often lacks any semblance of common sense, can be easily is early in its R&D phase. This implies that advances are ahead, with
fooled or corrupted, and can fail in unexpected and unpredictable ways. breakthrough innovation expected. With extensive AI investments con-
In other words, proceed with caution. tinuing across the globe, it is a good bet that AI and neural networks
This column looks at how AI technologies are affecting the automo- will solve many more complex problems — including challenges in the
tive industry. We’ll consider: automotive industry.
• How AI solves problems
• AI’s benefits and drawbacks in automotive AI DRAWBACKS
• The unique challenges of using AI in automotive Among the challenges in developing and deploying AI technologies is
• The auto electronics segments already using AI adequate training of neural networks. In general, the more complex the
• Future auto electronics segments that will rely on AI technologies problem, the more complex the neural network model must be. That
AI development has three phases: Build AI models, train AI models implies large models. Training requires vast resources and expertise
using relevant data, and use the trained model to solve problems (the to design and test AI models that rely on large datasets to verify that
inferencing stage). models work as advertised.
Most AI models are based on multiple versions of neural net- Larger sets of training data are becoming available, but training
works and learning networks. Examples include convolutional neural remains a time-consuming and expensive task. Most training data also
networks, generative adversarial networks, deep reinforced learning, must be labeled by humans to allow the AI models to learn and become
federated learning, and transfer learning. Each brings different advan- proficient. There is growing concern that biases are creeping into
tages and drawbacks. All are evolving rapidly. training data.
The table below summarizes the pros and cons of AI technologies Then there is the black-box problem: It remains difficult to deter-
along with safety considerations and proposed regulations. mine how AI models make decisions. Such obscurity remains a big
problem for autonomous systems. Better solutions are needed.
AI ADVANTAGES Another issue involves a model’s sensitivity to minor data changes.
AI is primarily used to solve complex problems. Because the auto That vulnerability creates security concerns, including the potential to
industry has plenty of difficult problems, AI is playing a growing role hack autonomous systems and the resulting threat to AV safety.

AI in Automotive: Big Picture

Topic Key Information Other Information
• AI is primarily for complex solutions • Automotive has lots of complex problems
AI advantages • AI is viable tech only for solving AV use cases • Neural network is leading technology
• AI tech is new → Huge future advances expected • Innovation required for auto AI solutions
• AI models are becoming very large • Large resources: Design, test, verify
• Large AI training data is required • Costly and time-consuming to gather
• Model training data with bias or other issues • Training data has unwanted features
AI drawbacks: training • AI model obscurity in how decisions are made • Unknown how some decisions are made
• Sensitivity to minor data changes • Can change results → Hacking potential
• Shortage of AI expertise • In automotive and other industries
• High compute resources, costs, and power use • Emerging AI-based chips will help
AI drawbacks: inference • Many AI solutions have black-box issues • Undesirable or prohibited in auto AI

• How can AI provide required automotive safety? • More research needed
AI safety • Center for Security and Emerging Technology (CSET) • Groundbreaking report on AI crashes
• Explainable AI (XAI) can add safety • XAI is a Darpa project
• EU GDPR impacts some auto AI solutions • Mostly in-vehicle content and HMI use
AI regulation • EU-proposed regulation for AI technology • EU max penalty: €30m, or 6% of revenue
• How will regulation impact AI in automotive? • What AI regulation in other regions?
(Source: Egil Juliussen, August 2021)

www.eetimes.eu | NOVEMBER 2021