Generative AI Unleashed for Sustainability

Written by Andreas Feiner, Saskia Becke, Nimanshi Jha (PwC) and Dr Laura-Marie Töpfer (Microsoft). The fusion of technology and sustainability is poised to revolutionize the global landscape, ushering in an era that transforms businesses, driving increased revenue, optimizing resources, all the while safeguarding the environment. In this dynamic landscape, Generative AI emerges as a groundbreaking catalyst that challenges the conventional norms of artificial intelligence and brings forth an unprecedented opportunity for sustainable transformation. By delving into the question, how does the distinctive nature of Generative AI set it apart from traditional AI, and what specific advantages does it offer in driving the incorporation of sustainable practices within businesses?

This paper navigates the uncharted territories of technology-driven sustainability, not only offering a breakaway from traditional standards, but also ushering in a revival of corporate methodologies for sustainability. The unique capabilities of Generative AI not only redefine the boundaries of traditional AI models but also offer a strategic catalyst for embedding sustainable practices at the core of business operations, promising a future where innovation aligns seamlessly with environmental responsibility. As the narrative explores the uncharted territories of technology-driven sustainability, the focus sharpens on Generative AI’s transformative potential and its implementation. To lay the groundwork, this paper first initiates an exploration into the concept of sustainability and its paramount importance for the corporate organizations.

Drivers for Companies regarding Sustainability

Within the business landscape, corporate sustainability signifies a comprehensive approach that balances economic, environmental, and social goals while focusing on meeting the needs of future generations just as it is capable for the present one. For companies, adopting sustainable practices optimizes risk management, saves costs, engages employees, fosters stakeholder relations and enhances reputation. As today’s world grapples with an unprecedented environmental crisis, the call for a shift towards sustainable practices reverberates globally. What’s nudging companies towards this shift? Increased focus by companies regarding sustainability is driven by three key factors.

1. One of the factors being regulatory pressure towards compliance both in existing as well as future regulations, where non-compliance to such regulations carries both legal and reputational risks. Continuous adaptation is thus necessary for being assured of adhesion to the most updated requirement.
2. A second more important integration driver of environmental, social and governance aspects in companies can be seen in risk management. This will motivate the company to always better manage environmental and social risks, increasing its resilience while protecting its reputation and financial stability.
3. Finally, it is economic potentials that arise out of a sustainable corporate orientation and take care of new business opportunities ensuring the long-term value creation of a company.

Building upon the conceptual understanding about sustainability and its paramount importance in the contemporary business environment, the narrative now directs attention to the current challenges faced by businesses in their pursuit of the journey to achieve net zero goals.

Current Challenges for Companies regarding Sustainability

Despite growing awareness, organizations encounter various challenges in integrating sustainability into their operations. Sustainability covers a range of interconnected issues, such as climate change, depletion of resources, environmental degradation and social fairness, which adds complexity. The existing landscape of sustainability is characterized by the following challenges for businesses:

Navigating these challenges demands a strategic and adaptive approach that can be accelerated by GenAI. As organizations strive to stay ahead in the ever-evolving realm of sustainability, the commitment to innovation, education, and a harmonious integration of principles becomes imperative. By acknowledging and addressing these challenges head-on, we pave the way for a more resilient and sustainable future.

The use of AI in sustainability is not only widely discussed in the literature but is also successfully applied in practice. In the unfolding narrative of artificial intelligence, GenAI stands as a compelling protagonist, pushing the boundaries of conventional models. While the foundation models are not entirely novel, the distinctive feature lies in their generative capabilities. To comprehend the efficacy of GenAI, it is essential to scrutinize the constraints inherent in conventional AI methodologies. Traditional algorithms demonstrate proficiency in effectively handling structured data organized in tables, they grapple with contextual understanding, often confined to singular tasks. However, sustainability challenges often present unstructured data in the form of text documents, images, and graphs, eluding the analytical grasp of conventional AI.

In contrast, GenAI exhibits versatility, seamlessly transitioning between tasks such as creating comprehensive reports, developing social media strategies, and automating workflows across diverse domains. A key distinction lies in the alignment of GenAI systems with human values. Unlike earlier models that exhibited misalignment, GPT-3+ (Generative Pre-Trained Transformer and its successors) have been trained on positive reinforcement learning, effectively mirroring human preferences and aspirations. This alignment stems from the incorporation of reward engineering and inverse reinforcement learning techniques, enabling AI systems to internalize human values and pursue objectives aligned with our societal norms. Furthermore, the sheer magnitude of GPT-3’s neural network, with over 175 billion weights, where weights refers to its parameters associated with connections between the neurons, enables its remarkable ability to process and generate complex information.

How GenAI Surpasses Conventional Approaches

GenAI, with its ability to excel in unstructured data analysis, model complex relationships, handle ambiguity, and provide possibility of human AI collaboration, stands as a transformative solution to previously mentioned sustainability complexities and challenges:

In the rapidly evolving landscape of artificial intelligence, it is advisable for companies to take a holistic approach that combines the benefits of traditional AI and generative AI. To make the right decision, it is necessary to comprehend the unique abilities of each approach and their suitability for specific organizational requirements. Traditional ML excels in tasks that demand precise pattern recognition and predictive analytics demonstrating more accuracy in task specific applications. Following strict and precise rules it proves invaluable in applications such as fraud detection, risk assessment, and customer segmentation, where the emphasis is on accuracy and efficiency. On the contrary, GenAI thrives in environments that require flexibility, adaptability, and the ability to generate creative solutions – novel ideas, concepts, and content. Its applications span diverse fields, including product design, marketing campaigns, and scientific discovery, where innovation and exploration are paramount.

Use Cases Across Key Sectors

In the pursuit of sustainable business practices, organizations across various sectors are increasingly turning to GenAI solutions. Lets explore some diverse case studies across various industries showcasing the transformative impact of generative AI on sustainability initiatives.

Potential Applications of GenAI in Environmental, Social, and Governance Dimensions (E, S and G)

Having explored these insightful case studies of Generative AI’s real-world impact across diverse sectors, let’s have a look at the potential applications of Generative AI within the ESG dimensions.

Engaging in sustainable business practices not only represents an offering but also functions as a distinct competitive advantage, setting forward-thinking businesses apart from their competitors. We at PwC guide you towards operational excellence for your individualized sustainability solutions based on your needs. Generative AI has an impact, beyond being a buzzword. It plays a pivotal role in optimizing operational efficiency, requiring minimal executive involvement.

We stand on the verge of witnessing carbon footprints evolving into a commonplace metric, much like how we perceive billing data. Sustainability reporting (e.g. Corporate Sustainability Disclosure Directive, short CSRD) is at the same level as financial reporting. This shift marks a pivotal moment in our journey towards sustainability, where we recognize it not as a mere moral obligation but as a gateway to prosperity – sustainability is the new profitable.

Stay tuned for more insights! Together with Microsoft and the University of Oxford we are working on a paper on Data & AI for Sustainability. In this study, we explore the transformative potential of AI within the realm of sustainability and corporate profitability. Some sustainability pioneers have recognised the financial potential of Data & AI for Sustainability and are successfully realizing huge scale cost-saving potential and unlocking new revenue streams. This integration redefines the competitive landscape for visionary business leaders. Data & AI for Sustainability represent a revolutionary paradigm shift, fostering sustainability as a driver of corporate prosperity, with implications for global-scale collaboration and continuous progress toward a more sustainable and profitable future.

Generative AI is increasingly becoming a powerful force in driving the progress of sustainability initiatives. Its ability to automate various tasks, identify new possibilities, and improve decision-making processes has proven to be instrumental in accelerating the realization of these projects. But how can companies realize this potential? This chapter will shed light on the implementation approach.

Technical Understanding of Generative Models

Selecting the appropriate Generative Model and Type for a particular sustainability application is crucial since it can have a considerable influence on the accuracy, efficiency, and overall effectiveness of the model. The decision regarding the type of model to be used should be based on various factors, including the characteristics of the data, the desired results, and the computational capabilities at hand.

Types of Generative AI

GenAI spans various domains, encompassing text-based models like LLMs, image generators like DALL-E, VAEs and GANs, music and speech generators like tacotron, art and faces generators like styleGANs, and even code generation models like Codex. The diversity in Generative AI models, including GANs (Generative Adversarial Networks), VAEs (Variational Autoencoders), auto-regressive models, and flow-based models, opens avenues for multifaceted applications. Text generation involves crafting reports and content for sustainability reports and policies, while image generation extends to visualizations where graphics can be created for sustainable design ideas and promotional content around sustainability such as arts and videos that convey powerful messages about environmental conservation, raising awareness through visually striking and emotionally resonant pieces, sustainable product concepts, and impact images. In a similar vein, music and audio, video, 3D modeling, speech, faces, and code generation offer unique capabilities for innovation across sustainability. Where StyleGANs can be used to generate diverse and inclusive visual content, promoting representation and awareness of sustainability across different communities and cultures and code generation can generate code snippets that calculate and report sustainability metrics within software applications.

Foundation Models and beyond

The foundation models, be it for text, time series, image, or other domains, serve as the basis for Generative AI. These models are trained on extensive amounts of data and use techniques such as next-token prediction and masked language modeling to generate outputs that are contextually relevant and coherent. The architecture includes embeddings, attention mechanisms, and encoding methods that empower GenAI’s contextual understanding.

• Embeddings refer to the numerical representations of words or other inputs, typically learned from large amounts of data.
  
• Attention mechanisms, on the other hand, are a type of neural network that allows models to focus on the most relevant parts of the input data.
• Encoding, meanwhile, serves as a means to convert the input data into a format that the model can efficiently handle.

The computational constraints of large models are tackled by model compression techniques such as quantization, pruning, and distillation.

• Quantization is the process of converting the weights of the model from higher precision data types to lower-precision ones. It can be applied to weights as well as activations. Weight quantization focuses on decreasing the precision of the weights, while activation quantization deals with reducing the precision of the network’s activations in order to reduce the memory footprint of a neural network.
  
• Pruning is a technique employed to eliminate unnecessary or insignificant connections within a neural network,to reduce the size of a neural network.
• Distillation, on the other hand, is a method used to train a smaller and less complex neural network to mimic the behavior of a larger and more complex neural network.

To further improve their performance, optimized architectures and training methods like KV cache, group query attention, and benchmarking are employed.

• A KV cache or key-value cache, functions as a data storage system that holds key-value pairs. The purpose of utilizing KV caches is often to enhance the performance of neural networks,as they can store frequently used data and avoid having to re-compute it each time it is needed.
  
• Group query attention is a technique that allows attention to focus on a group of words by acquiring a sentence representation that encapsulates the interconnections among the words.It shares single key and value heads for each group of query heads. Benchmarking refers to the assessment and analysis of a model’s performance.
• Benchmarking is important for ensuring that models are performing as expected and for identifying areas where they can be improved.

Foundation models, model compression, optimized architectures, and benchmarking form the cornerstone of Generative AI, enabling its wide-ranging capabilities.

PwC offers a wide range of services from consulting specific to your sustainability needs to building an end-to-end Generative AI solution to solve your sustainability challenges. PwC is uniquely positioned to advise clients and stakeholders on this transformational technology to build trust in their business and drive sustained outcomes. PwC possesses a wealth of internal and external expertise, along with cutting-edge data resources, which can be effectively utilized to harness the vast capabilities of Generative AI. We offer a unique approach to AI that includes a standards-based approach to enterprise GenAI architecture, a factory model for scale, and a repeatable approach to use cases to deliver increased value for clients. Please feel free to reach out to our expert team for more information.