AI becomes mainstream for engineers and scientists
AI will be one of the first “tools” engineers and scientists look to for innovative solutions in solving problems and building applications in 2021. It will be used to completely replace or augment traditional techniques in math, physics and engineering disciplines such as controls and signal processing. AI will continue to help bring answers to previously unsolved problems and enhance existing solutions. Reduced Order Modelling is one example where AI replaces large high fidelity and computationally expensive models with a more efficient solution.
In academia, there will be an increase in the amount of research being performed where AI is embedded in engineering and scientific applications. This can be foreseen due to the next generation of engineers and scientists being taught how to use AI, thanks to a rise in AI + “X” courses, where AI is being integrated into traditional modules where they didn’t exist before (e.g., AI + Control Theory).
AI aligns engineering, computer science, data science and IT direction
Engineers will continue to work with data scientists using AI models to enhance existing applications or discover new innovative solutions to the projects they’re working on. However, creating a successful AI-based system is more than just developing a model. It requires model lifecycle management, which includes training, deploying, monitoring and updating the model for the system in which it resides.
To do this efficiently these processes need to be automated, robust and well maintained.
In 2021, engineers will augment their workflows to include: development best practises (i.e., model and data versioning), building production pipelines with IT, etc. This will be required to support the AI-enabled systems which will operate in real world environments over years and even decades.
Model explainability will reduce the aversion to AI within safety critical systems
AI has long been considered a black box approach to modelling systems, and with it a fear that how it operates is largely unknown. As more explainability methods are being produced by research and more software vendor tools offer them, industry practitioners will more readily adopt AI innovations within their workflows.
Engineers and scientists are beginning to understand why a model is making certain decisions and the limits at which a model can operate safely. They are running experiments to explain how a model operates in a variety of scenarios and using visualizations to understand the inner workings of a model when it doesn’t behave as it should. It’s driving innovation in the verification and validation of AI within safety critical systems, with automotive, aerospace and medical standards committees, such as EUROCAE and the FDA, working on the levels needed for certification.
Simulation and testing will go 3D and become more realistic
An important step towards formal verification and validation of AI within safety critical systems is being able to test every possible scenario in which the system will operate. For self-driving cars, this step is currently being performed physically through road testing with the aid of a human driver. Physical testing vastly limits the variety of scenarios and increases the time it takes to capture all the critical edge cases.
In 2021, engineers will look to leverage recent advances in software tools to eliminate physical testing through 3D simulation. They will integrate their AI models with traditional methods for modelling physical systems (i.e., Model-Based Design), then perform automated testing against a variety of simulated 3D scenarios.
More AI models deploy to more low-power, low-cost embedded devices
The options to incorporate AI into more edge-based systems are increasing, and engineers are taking advantage of increased hardware support for more low-cost, low-power devices including FPGAs, ECUs, and MCUs. Software vendors are enabling this innovation by extending capabilities to non-chip experts who can take advantage of advanced techniques once reserved for embedded engineers. Techniques such as quantization and pruning, which reduces the size of the deployed model, and efficient pretrained models available in the deep learning community will enable efficient deployment of AI and will enable larger adoption of AI-based systems in 2021.
A note on COVID-19:
Investment in AI will not decrease
We’d be remiss if we didn’t mention COVID-19, an unforeseen trend of 2020, which is expected to continue with us into 2021. Overall investment in AI-related projects has not decreased. While some heavily impacted industries have cut back in the near term, analysts report that these have been offset by those who increased their investment above what they had forecasted. Many are using this time to invest in upskilling remote learning, with AI themed courses amongst the top sought after by the engineering and scientific community, making them primed and ready to take on more AI projects in 2021.
