The Awakening of Industrial GenAI

Generative Artificial Intelligence (GenAI) has recently taken the world by storm with the release of ChatGPT. What about industrial adoption of this technology?

Trends show that industries prefer a focused adoption. Greater gumption for the technology has been shown by those in high potential for automation and content creation. The focus of such industrial users is on productivity and integration of workflows.

There is significant variation in the level of awareness and adoption of GenAI across industries and functions. Sectors heavily relying on data, such as software engineering and retail started leading the way. Software Engineering business, for instance, is transforming itself at breakneck speed with new GenAI tools performing code generation, completion, testing, translation, documentation and optimization, thereby tremendously accelerating time-to-market.

Highly regulated industries like healthcare and some professions like Human Resource specialists and financial advisors, still need more confidence in using GenAI. A survey by Salesforce revealed that healthcare service professionals who were assumed to get substantial benefits from GenAI were the least likely to use it, possibly pointing to issues with training and trust.

Energy, Manufacturing and Supply Chain Management, which rely on well established processes and physical systems are somewhere in the middle as they find it challenging to integrate GenAI.

TCS AI STUDY

TCS recently conducted an AI for Business Study to understand how business leaders around the world view AI and its transformative power to reimagine their business. 57% of executives say that they are excited and optimistic about AI's potential impact on their business. 72% are currently planning to address the impact. However concerns were raised for GenAI, especially in the areas of security, privacy, ethical use, lack of IT readiness, talent development, training and cultural shifts. Industry leaders also feel that there are no good enough metrics to measure returns.

Despite GenAI's commercial promise, industries are proceeding with caution.

GOVERNMENT INITIATIVES

Governments around the world are doing their bit to accelerate industrial adoption of GenAI.

Singapore launched its revamped National AI Strategy or NAIS 2.0 in Sep 2023. It collects transformation ideas for businesses with AI, provides sector specific Digital Plans for SMEs, encourages them to use its Digital Services lab and provides funding and support through programs like A*STAR with Accelerated initiative for AI and SGInnovate. GovTech Singapore recently whitelisted its PAIR solution - a Large Language Model (LLM) powered assistant for public officers and has received good adoption and favorable reviews.

On 21 June 2024, Australia also released its National framework for AI in government, which establishes cornerstones and practices of AI assurance. India with its IndiaAI initiative has multiple startups and 100s of case studies - like multilingual AI anchors for farmers, digitizing clinical trials, blue-collar job recruitments, etc.

The Norwegian government is taking Industrial AI further. Its Petroleum Directorate manages a database, called Diskos, that holds over 15 petabytes of North Sea sub-surface data. Its National Strategy for Artificial Intelligence comprehensively covers what the government will do on aspects of platform, people, governance, partnerships and funding for AI, which would be a boon to the offshore petroleum industry.

GOING MAINSTREAM

GenAI with LLMs are powerful tools built on vast knowledge organized into distinct categories. Providing industrial context is crucial, especially when dealing with unfamiliar topics and prevent hallucinations. In the field of computability theory, it is established that there are infinitely more problems that cannot be solved by algorithms (undecidable) than those that can be (decidable). Therefore, the likelihood of an LLM delivering an accurate answer to an undecidable problem without context is very low.

To fully harness the capabilities of GenAI, contextualize it, make it useful for solving industry problems and route it mainstream, it is important to pay attention to the following six key areas:

1. LARGE, HIGH QUALITY DATASET: A massive, diverse, relevant and well-structured dataset is essential for training Industrial GenAI. The dataset serves as the foundation for AI to learn patterns, relationships and contexts through high-fidelity knowledge-graphs.

For instance, in Manufacturing and Energy, proliferation of Internet of Things (IoT) devices with Edge computing and OT-IT convergence is generating and unlocking vast amounts of industrial data. Industry 4.0 with digitization and digitalization is expanding this multifold, not to mention sources like Diskos mentioned above. This ever-growing data pool provides rich resource for training GenAI models allowing them to learn from real-world industrial scenarios and making knowledge graphs sharper. Like-minded industries should share their data-sets to make them even richer.

2. ALGORITHMS AND MODELS: Appropriate sophisticated deep learning algorithms and models, such as Graph Retrieval-Augmented Generation (Graph RAG), Recurrent Neural Network (RNN) and Generative Adversarial Networks (GAN) are necessary to process and learn from the dataset, enabling GenAI to generate coherent, context-specific yet diverse outputs.

Algos and models for instance drive machine vision, where AI analyses images and video feeds from sensors inspect products for defects, optimize production lines and improve safety; are at the heart of autonomous robots where smart AI enables handling of complex, hazardous tasks in strenuous environments; and perform predictive maintenance where AI analyses sensor data to predict failures before they occur.

3. COMPUTATIONAL POWER: GenAI models require significant computation power and resources to process large datasets and perform complex calculations. These typically include Graphics Process Units (GPUs), Tensor-Processing Units (TPUs), Language Processing Units (LPUs) and other specialized hardware.

Higher the aspiration and expectation from Industrial GenAI, greater would be computing power and investments needed. For instance in March 2024, the world's most profitable company, Saudi Aramco unveiled its GenAI model called Aramco Metabrain with 250 billion parameters trained using seven trillion data points collecting more than 90 years of company history. They own 2 of the 3 largest supercomputers in the region and doubled their venture funding to $7.5 billion.

4. FINE-TUNING AND OPTIMIZATION: Fine-tuning involves taking the pre-trained model that has already learnt from large high-quality dataset and further refining it on a smaller, task-specific data set, that are contextualized to granular use cases.

Take for example the Oil & Gas industry, where a pre-trained model can be fine-tuned to optimize predictive maintenance for a specific equipment, like a pump or a valve, generate high-fidelity synthetic data for training AI models in drilling and exploration. Fine-tuning and thereby optimization is done through various techniques such as hyperparameter tuning, regularization, retraining and early stopping to continuously achieve best possible performance

5. DOMAIN EXPERTISE: A deep understanding of specific-industrial domain for the intended usage is essential. Industrial expertise helps in contextualization, designing the models, selecting the right dataset and fine-tuning AI performance.

For instance, to create a high-quality generative model for the Utility industry, nuances of operations such as grid management complexities, need for accurate forecasting, importance of reliability centered maintenance is essential to choose datasets that reflect industry-specific pain points and fine-tune algos.

6. GOVERNANCE: Last, but not the least, active governance is a must for industrial GenAI. Developing and using relevant and transparent evaluation metrics with feedback mechanisms is crucial to assess the AI's performance, identify areas of improvements and refine its output.

Financial measures such as ROI, cost savings and revenue growth due to AI should be tracked and appropriate change management with training should be put in place. Governance should also include mechanisms for monitoring and auditing the AI, with transparent processes for reporting and addressing biases or errors identified, thereby ensuring responsible and ethical AI use.

With corporations and governments driving accelerated GenAI adoption, we are at the threshold of a huge awakening of the Industrial GenAI. Industry leaders must take cognizance of the six areas and appropriately align them to their needs and aspirations to create a value-adding, responsible and ethical GenAI that is profitable and safe for the populace.