The manufacturing industry is experiencing another transformation driven by the recent rise of Generative AI (GAI), which joins the more established Discriminative AI (DAI). Understanding these two fundamental approaches to AI is crucial for business owners and manufacturing professionals. While DAI focuses on making decisions, GAI creates new content.
DAI uses data to make decisions, such as: is this a flower, if so, what kind? Or, given these past trends, what will happen next? These two classes of decisions are known as classification and regression and have accounted for most applications of AI to date. For instance, they help us read car license plates or detect banking fraud.
The newer GAI has a different purpose. GAI creates new content. Famously, ChatGPT allows a user to seek narrative answers to complex questions. You want an essay on Oppenheimer? You got it. Or, if you want a picture of a subject that has a particular likeness – such as, what would you look like if you were the offspring of John F. Kennedy? – just ask.
AI solutions consist of an AI model that processes the external incoming data to be optimized and then makes a decision. It knows how to do this because the AI model has been trained on a training data set beforehand. But where did this training data set come from and how much of it is out there? The answer to that depends on the application.
It’s worthwhile comprehending the size of the data currently being generated in databases and what it means to industry and society at large as seen through the AI lens. A recent study by UBS Warburg estimates that the amount of data available globally is expected to grow more than 10 times from 2020 to 2030, reaching 660 zettabytes. A zettabyte is a billion terabytes or a trillion gigabytes.
This new data is generated by the existing population of humans and machines with an additional two billion more internet users and 30 billion IoT devices expected to come on line by the end of the decade. This is while data storage costs go down by 25 to 30 percent every year. Thus, we have many more devices and people adding data to ever lower digital storage costs.
Two countries lead the data accumulation: The United States and China. At the moment, China’s store is about the same as the United States’, but it is growing much faster. Most Organization for Economic Co-operation and Development (OECD) countries have a slower growth of data than emerging economies.
By 2025, half of all the data, 200 zettabytes, will be stored in the cloud. Cloud data centers will process the majority of this data. Sixty percent of all corporate data is already stored on the cloud. The majority of AI processing will also occur on the cloud. The majority of AI services will also be available on the cloud in the form of a software-as-a-service (SaaS) business model.
The critical factor here is that any form of AI thrives on more data. And the enterprise needs to define its relationship with the massive build-up of data and AI services outside its firewall.
Manufacturing and AI
AI enables the transition from manual processes to sophisticated, automated systems. This shift is not just about replacing human labor; it’s about expanding organizational capabilities with machines that can work long hours, with precision and in environments that may be hazardous to humans, reducing the scope for human error.
The impact of AI on the various stages of manufacturing processes is manyfold. Both GAI and DAI offer unique advantages and are tailored for different stages of the manufacturing process.
GAI, as mentioned before, excels in creative and design-oriented tasks. GAI’s ability to process and learn from vast data sets can lead to the discovery of new materials, design techniques and production methods. This also means that GAI relies on extensive public databases like Wikipedia and Common Crawl for training. In the design and prototyping phase, these methods can enable rapid simulation and testing of new designs, reducing the time and cost associated with traditional prototyping methods. For production, generative models can analyze the production process and provide patterns for optimized processes.
DAI’s strength, on the other hand, lies in the capabilities for precise distinctions through efficient data processing. During production, by integrating with manufacturing systems, these models can monitor the process and make real-time adjustments to ensure optimal performance. In quality control, DAI revolutionizes the way products are inspected and tested.
Through advanced image recognition and machine learning algorithms, DAI systems can identify defects and irregularities with accuracy and speed. Finally, in maintenance, by analyzing data from sensors and identifying patterns indicative of potential failures, DAI can forecast equipment malfunctions before they occur, playing a crucial role in predictive maintenance.
DAI and GAI benefit from an increase in data volume, albeit in different ways. DAI improves with more specific data, like videos of a welding pool, while GAI thrives on a variety of data. The key takeaway for enterprises is that the effectiveness of AI technologies correlates directly with the nature of the task, amount of data and quality of data available.
Quality, specificity and safety
The next topic to pay attention to is the data and the challenges that come with collecting good data for AI. High-quality, relevant data is essential for AI’s success, especially in specialized fields like manufacturing. The reliability and safety of AI models, particularly in critical tasks like robotic welding, depend heavily on the quality of the original training data. Additionally, generative models present unique security challenges, including intellectual property protection and potential misuse.
Manufacturing requires highly specific data for AI training, typically proprietary and sensitive. This specificity means publicly available data is often inadequate for effective AI application in manufacturing. A crucial consideration for enterprises is whether to share data with other users in the cloud. This decision impacts data privacy, security and the overall effectiveness of AI applications.
When deciding to share data, especially for AI applications, enterprises must carefully evaluate several crucial factors. This includes conducting a risk-benefit analysis, ensuring compliance with data protection laws, employing strong data encryption and implementing robust access controls.
For companies looking to contract third parties to develop their AI, it should be noted that they will be sharing their data with industry partners and potential competitors. This data may include various types such as operational metrics like production rates, sensor information on machinery status, supply chain figures like stock quantities and client details, among others. Ensuring the security of this data is crucial.
Comprehending the data’s structure, origin and quality is vital for high-quality, dependable results. It is standard to use different data abstraction methods instead of raw data formats. This involves practices like amalgamating data from diverse sources or generating new data sets from existing patterns. Consequently, if the data source lacks credibility, substantial time and resources might be misdirected in your pursuit of automation objectives.
For these complex considerations, working with an experienced and trusted partner can be a key to successfully addressing these challenges. Such a partner can provide expertise, resources and support to ensure that data sharing, specifically in the cloud, is both effective and secure.
SaaS for manufacturing
Another important trend that manufacturing facilities need to pay close attention to is AI presented as a SaaS offering. Some industries might find themselves unfamiliar with the trend of subscription for software in their manufacturing process, even though they already use SaaS for products such as ERP. The shift from a capital expenditure (CapEx) model to an operational expenditure (OpEx) model aligns with the nature of software, providing regular updates and scalability. This transition is essential for manufacturers to remain competitive in a digital landscape.
Looking at the statistics mentioned above, the data accumulation led by the United States and China, along with the predicted shift of half the global data to the cloud by 2025, underscores the shift toward cloud-based AI processing and services. This trend suggests a future dominated by AI services delivered as SaaS.
The main reason is that SaaS introduces unprecedented agility and flexibility. Manufacturers can access cutting-edge tools and technologies without significant upfront investments, allowing them to rapidly adapt to market changes and customer demands. Moreover, SaaS solutions facilitate seamless integration of various functions, which is particularly crucial in an era where data-driven insights are key to optimizing production processes and reducing downtime.
By moving to SaaS, manufacturers not only keep pace with technological advancements but also leverage them to create more responsive, efficient and competitive operations. The ability to continuously improve AI models and scale software solutions as needs evolve ensures that manufacturing enterprises can extend on their existing system capabilities and remain robust in a fast-changing industrial landscape. Note that historically, SaaS required developers to provide upgrades. With AI, it is the AI that does the updating in the background. This is a continuous improvement process that enhances the performance of the SaaS offering.
Cost-benefit analysis, ROI
Manufacturers must weigh the costs and benefits of adopting AI, particularly in choosing between GAI for innovation and DAI for control and precision. A recent IDC study reported that enterprise implementations of AI were returning 3.5 times the original investment. Some of the projects included GAI, but most were DAI. Seventy percent of respondents reported already using AI.
Of course, there are challenges and risks associated with implementing AI: new methods, intellectual property, safety and security, quality and reliability, cost and data privacy. Finding the technical resources to start and maintain both conventional and AI software is particularly difficult. The key to success for the enterprise is the adoption of reliable AI models through a trusted industry partner. DAI is known for its precision and reliability and offers a more controlled approach to AI applications. It has less concerns about unintended IP infringements while ensuring higher standards of quality and reliability in production processes.
Working with an established industry partner for AI development not only mitigates risks but also accelerates the integration of AI into manufacturing processes. These partners bring a wealth of experience and a deep understanding of industry-specific challenges and requirements. Their expertise in addressing the complex landscape of AI, from algorithm development to implementation, can significantly shorten the learning curve for manufacturers.
Reputable companies ensure that the AI solutions are tailored to fit the unique needs of each manufacturing operation, thereby enhancing effectiveness and efficiency. This collaboration also provides access to ongoing support and maintenance, ensuring that AI systems evolve with the enterprise and remain at the forefront of technological advancements.
The key question is, given all the moving parts in the AI landscape, what should an enterprise do? DAI and GAI models are readily available. Cloud computing is commonplace and SaaS allows users to access innovation at an unprecedented rate. Data, cleansed and organized, is also growing and is a greenfield opportunity for exploitation. That said, is there a first mover advantage to learning how to pull these disparate trends together?
Innovation like AI rarely works in isolation. There are always systemic consequences of making something faster or better in one area of the enterprise. In the Theory of Constraints, the bottleneck always moves somewhere else, allowing for further, wider improvements. The heart of the AI adoption question is, what is its distinctive competency? The prior automation of accounting, sales, marketing, manufacturing and supply chain resulted in winners and losers. The winners were the ones who envisioned correctly how the technology could help them improve and then implemented it correctly. It was their ability to map the digital inflection points into their distinctive competency.
This is the same with AI. How an enterprise interprets DAI or GAI and spreads the benefits through its ecosystem is what matters. Expect that now the corpus of data grows and AI learns as we sleep. Data about many disciplines is now being automatically transformed into knowledge and made available on tap. How are you going to respond to a digital universe where everything is everywhere and available all at once?