Ruomu Tan, Nicolai Schoch, Reuben Borrison, Martin W Hoffmann, ABB Corporate Research Ladenburg, Germany, ruomu.tan@de.abb.com, nicolai.schoch@de.abb.com, reuben.borrison@de.abb.com, martin.w.hoffmann@de.abb.com
Generative AI refers to artificial intelligence (AI) systems that can generate new content – such as text, images, music, or computer code – based on training data and user inputs. These systems use machine-learning techniques, particularly neural networks, to analyze patterns and features in vast amounts of existing content (ie, training data) and produce original outputs similar in style. Generative AI – devoid of consciousness, emotions or subjective experiences – operates solely on probabilistic algorithms and data and generates content based on patterns identified in its training data, not personal creativity. The main difference to established AI methods lies in the generation of original content. For example, analytical AI, the traditional approach, is limited to deriving insights, creating classifications, or making predictions or recommendations →01.
Originating in the mid-20th century, generative AI underwent a significant shift with the advent of deep neural networks in the 2010s, particularly with the introduction of the breakthrough Transformer architecture in 2017. By leveraging this neural network architecture, which processes information like the human brain, generative AI approaches, such as ChatGPT (Chat Generative Pre-trained Transformer), can create human-like outputs. Pre-training on vast datasets enables the model to understand patterns, context and nuances. Combining this pre-training “education” with the Transformer architecture produces a groundbreaking AI that is unleashed through advanced training that uses a huge amount of data. The “chat” part, ie, the interface to non-AI-scientist users, gives the technology a huge push, as it democratizes the accessibility of Generative AI to general audiences.
Generative AI for and at ABB
While the media focuses on chatbots and image generators, Generative AI will also play a vital role in ABB industrial applications so:
- Employees can speed up their daily work.
- Internal business functions can be automated.
- Product performance can be improved.
ABB has already started using Generative AI, to the advantage of customers as well as internal applications. For example:
- Intuitive user interaction with ABB Ability™ Genix Industrial Analytics and AI Suite
- Efficient engineering of control systems and PLCs by control code generation
Risks, limitations and challenges
While offering transformative potential, -Generative AI also presents risks, limitations and challenges:
- Due to its stochastic nature, Generative AI may “hallucinate” – namely, generate wholly or partially inaccurate information without indicating it is doing so.
- Copyright infringement: Generative AI can reproduce or closely mimic existing copyrighted materials, particularly when the model is pre-trained on a large body of data with incomplete copyright information.
- Generative AI’s requirement for substantial computational resources may restrict the accessibility and business potential for certain customers and use cases.
- “Deepfakes” – highly realistic AI-generated images or videos – present serious concerns for misinformation and privacy violations.
Trends in Generative AI include multimodality for handling various data types (eg, language, visuals and sounds) and more easily customizable general GPT models so non-expert users can tailor them to their particular applications.
The reliable use of Generative AI in industrial applications requires a robust governance framework, ethical guidelines and technological safeguards. ABB is inviting its customers, partners and general public to collaborate on these topics to unleash the potential of Generative AI.
