Special thanks to Allison Earlbeck (CEO, Earlbeck Gases), David Schaer (President, Computers Unlimited) and Dave Beltz (CIO, Trinseo) for their contributions.
It seems as if 2023 was the year the “world” first heard about Generative AI (Gen AI). It was clearly one of the most impactful stories driven by remarkable technological progress that matched or surpassed human performance in some intricate tasks, such as answering complex medical exam questions, generating persuasive political messages, and even choreographing human dance animations to match diverse pieces of music.
This year it appears to be the year companies and organizations began using, scaling and attempting to derive value from this new technology. The influx of investment into this space is well into the billions with companies such as Microsoft, Meta, Google and many others both investing and starting to incorporate some of its capabilities into their business tools. In a recent McKinsey study, the technology is expected to herald a new age of efficiency, for example in manufacturing and supply chain alone, it could reduce expenses by up to half a trillion dollars! Is the hype warranted? Well, that’s to be seen…but it is clear that its potential is something that hasn’t been seen before. As a result, we will take a deeper look at Gen AI starting with a brief history, global and industry-specific trends driving the hype, and some best practices to employ to ensure successful implementation and adequate ROI on your investment. The full article was published in the November Global Print edition of Gasworld magazine and is contained in this LINK.
Definition and History
Generative artificial intelligence (Gen AI) describes deep learning models or algorithms (such as ChatGPT) that can be used to create new content, including audio, code, images, high-quality text, simulations, and videos in seconds. It has a fairly short history, with the technology being initially introduced during the 1960s, in the form of chatbots. One of the first primitive Gen AI solutions was ELIZA, which was a text chatbot created by Joseph Weizenbaum. ELIZA was one of the first examples of Natural Language Processing (NLP) and mimicked the work of a psychotherapist and could communicate with humans in natural language. But despite some advances, the computational power and data resources needed for systems like this to flourish weren’t yet available. So, it wasn’t until 2014, when the concept of the generative adversarial network (GAN) was introduced, that Gen AI evolved to the point of being able to create images, videos, and audio that seem authentic recordings of real people. Another breakthrough in Gen AI and in the development of NLP was the introduction of GPT (Generative Pre-trained Transformer) models. In 2018, the first version of GPT was created by OpenAI, and the rest we can say is history….
Trends
The excitement around Gen AI and its massive potential value has energized organizations to rethink their approaches to business itself. “The broad range of capabilities are both tremendous and unique,” states David Schaer, President of Computers Unlimited. “With large scale offerings such as ChatGPT, Co-Pilot, Gemini, coupled with a rapid build of new compute power, it’s easy to see the potential for the technology to be transformative across business processes.” Organizations are looking to seize a range of opportunities, from creating new medicines to enabling intelligent agents that run entire processes to increasing productivity for all workers. Many organizations say they are seeing value from their early Gen AI pilots/tests and those successes are driving more investment. As a matter of fact, in a recent Deloitte survey, two-thirds of organizations said they are increasing investments in Gen AI because they have seen strong value to date. However, a raft of new risks and considerations are on the horizon, none more challenging than the significant power demand of AI requirements as a whole. Goldman Sachs stated in a recent report, an average ChatGPT query needs nearly 10 times as much electricity to process as a Google search, which means there is a significant wave of electricity demand on the horizon. Consequently, they are forecasting global power demand to increase 160% by 2030, largely due to AI requirements (inclusive of Gen AI), with most electricity grids already operating at or near capacity.
But despite the challenges that lie ahead, there are several exciting trends that are enhancing the capabilities, functionality and ultimately the cost of Gen AI including:
• Conversation – Gen AI is making Conversational AI tools more intuitive, dynamic, and capable of handling complex interactions seamlessly, but there is plenty of room for improvement. They are enhancing advanced NLP and machine learning (ML) techniques to further improve conversations, by better understanding context to generate more coherent and relevant responses, and better personalize conversations based on the user’s history and preferences.
• Creativity – OpenAI’s Dall-E tool launched in 2021, introduced numerous unexpected capabilities, marking the first instance of an AI generating artwork from minimal inputs. While its initial version struggled to produce high-quality art, its subsequent versions (Dall-E2 & Dall-E3) have significantly improved, closely aligning with input requests and producing accurate images. Expect dramatic improvements to continue…
• Hyper-Personalization – one of the significant Gen AI trends is the use of AI algorithms to analyze vast amounts of data to predict and adapt to user preferences. For example, in the commercial pharma space, it is starting to be used to create hyper-personalized content for healthcare professionals by analyzing vast amounts of data to tailor messages and materials to individual preferences and needs. This enables more targeted and effective communication strategies, ultimately improving engagement and outcomes in the healthcare industry.
• Smaller Models – If 2023 was the year of large language models (LLMs), 2024 has witnessed the power of small language models (SLMs). LLMs are trained on massive datasets containing terabytes of data extracted from billions of publicly accessible websites, which comes at a significant cost. SLMs on the other hand, are trained on more limited datasets such as textbooks, journals, and authoritative content. These models are smaller in terms storage and memory requirements, allowing them to run on less powerful and less expensive hardware.
• Multimodal – The ambition of Gen AI is rapidly expanding beyond single-domain performance to embrace multimodal models that can process and interpret multiple types of data (i.e., text, audio and images). One example recently shard by IBM, was a model receiving a photo of a landscape as an input and it generates a written summary of that place’s characteristics. Or vice versa, it could receive a written summary of a landscape and generate an image based on that description.
• Regulatory – There are rising concerns over privacy and biases with the technology. The ambiguity surrounding regulations may impact the speed of adoption of all AI-related technology, as businesses may hesitate to invest amid concerns that future laws could render current investments outdated or unlawful. The Artificial Intelligence Act proposed by the EU aims to regulate AI and promote transparency, particularly for high-risk systems. In contrast, in the U.S., the primary hub for AI innovation, regulatory efforts are still in flux, despite endeavors to set standards for AI usage in government and developers’ pledges to uphold ethical practices. This is something to keep a close eye on…
The pace of change in this field is astonishing,” states Allison Earlbeck, President and CEO of Earlbeck Gases. “Staying informed about advancements is crucial, as AI’s capabilities evolve at an unprecedented rate. While AI offers many benefits, it also presents risks, such as deepfakes and security vulnerabilities, which must also be understood.” Dave Beltz, CIO at Trinseo also adds “there’s a lot of hype similar to other recent technology revolutions such as Cloud and Blockchain; however, Gen AI feels a bit different with plenty of weekly/monthly improvement, so much so, the level of accumulated improvement at the end of 12 months is expected to be extremely large.
Use Cases / Industrial Gases
Although Gen AI introduces a wide range of innovative features its use cases are not limitless. It has constraints for which traditional AI and other solutions are better equipped. Nonetheless, Gen AI has several use case that are being piloted/scaled across many industries including:
• Personal Productivity – semi/full automation of repetitive tasks like email drafting, scheduling appointments, summarizing documents, preparing/customizing slide decks, preparation of web meeting notes, creating personalized to-do lists, providing real-time translations, and so much more. In a recent study of the impact of Gen AI on the productivity of white-collar workers using OpenAI’s ChatGPT-4 model was spearheaded by researchers from Harvard, Wharton, and MIT. The study observed consultants from the Boston Consulting Group in a series of tasks, which resulted in a 40% increase in performance across the board. Now imagine if that can be scaled across a single working group or department…
• Enhanced Customer Engagement – although customer service/support, sales and marketing functions have been leveraging automation, data analytics, and ML solutions over the last several years, it is believed that Gen AI will enable even greater productivity and growth. It is helping businesses anticipate the customer needs and respond rapidly with meaningful hyper-personalized experiences. “As an Independent Distributor, I see numerous applications for Gen AI, “states Allison of Earlbeck Gases. “An example is the potential for real-time insights into customer trends, such as identifying declining volumes, at-risk clients, and new upsell opportunities.” In a recent McKinsey survey of Commercial leaders, more than 85% who have deployed Gen AI in their organizations report that they’re “very excited” about the technology, they point to improved efficiency, top-line growth, and customer experience as among the most important benefits.
• Production Optimization – Gen AI is enhancing process control systems to further optimize the gas/chemical production and filling processes (i.e., air separation, hydrogen production, packaged gas fill plants, etc.), by better predicting optimal operating conditions, to reduce energy consumption and improve yields. It is also being used to enhance maintenance models for manufacturing assets by generation of a predictive maintenance plan. By leveraging past maintenance history and capture of real-time asset performance from sensors, they are better able to predict equipment failures, enable proactive maintenance and reduce downtime.
• Supply Chain and Logistics Optimization – Gen AI’s ability to autonomously generate solutions to complex problems, has the potential to further enhance every aspect of the LB and PG supply chain landscape. It will further simplify, streamline automate processes from demand forecasting to route optimization, inventory management and risk mitigation, leveraging data and conversational interfaces. The Majors view this area as proprietary for obvious reasons, but you can confidently assume these capabilities are being explored/piloted/implemented based on the progress being seen in other tangential industries with complex supply chains (i.e., chemicals).
In addition, “AI computer vision can be used to help automate Accounts Payable (A/P) by reading and processing vendor invoices,” states David Schaer of Computers Unlimited. “It could also be used to capture, record, and validate cylinder markings for data accuracy and integrity, and a third use case is the use of machine learning to optimize and maximize bulk gas deliveries.” Dave Beltz of Trinseo also adds “there are plenty of use cases, but the real power is its ability to help you/your business see things you otherwise don’t see. By using both structured and unstructured data it helps answer the question, ‘am I missing something,’ and it is capable of being used to support activities in multiple areas within a company.
Best Practices
As with any new technology transformation, there is typically a higher failure rate at the beginning since early adopters are first to pilot and implement. However, the good news is that some Gen AI early adopters are delivering successful prototypes and impactful products using some common critical success factors. A few of the best practices from these players are include:
• Strategy clarity – if you are already using AI tools, you may need to “tweak” your AI strategy to incorporate Gen AI. If this is your first foray in AI then it starts with how does this support your business strategy and answering some basic questions such as how does this capability support or better enable your business strategy, what problem(s) or pain point(s) are we trying to solve, what value does this capability bring to the business (efficiency, productivity, revenue growth)? Then start with a clear and well-defined use case that targets the specific problem you are trying to solve. For some companies this may be a difficult step in “seeing the potential of Gen AI,” so there may be a need for an external consultant/company to assist and support this step.
• Data, Data, Data – it appears that “data” is one of the critical success factors for most every business process or application, but due to the unique nature of Gen AI things like data life cycle management, privacy, security, bias, etc., it takes on increased importance. Identify and address any shortcomings in data availability, quality, and integration, ensuring that data is clean, comprehensive, and accessible. Across leadership, define clear criteria for data quality, including accuracy, completeness, consistency, and timeliness. Document these standards to serve as a reference for data collection and processing. By maintaining a focus on data quality, it will lead to more accurate predictions, better decision-making, and ultimately, improved business outcomes for our companies.
Data is absolutely crucial to the success of Gen AI,” states David, “Data capture methodology, quality (both accuracy and depth), security, and building a digital footprint of your business processes are all key elements to maximize the potential benefits of AI.
• Use Case Selection – In parallel with creating a sound data foundation, identify discrete use cases for Gen AI and then experiment with the technology. The goal should be tangible, quick wins that will naturally build momentum. A good way to target the highest-value use cases will be to look for those that will have a direct impact on revenue, costs, risk, or other important outcomes, and are reliant on high volumes of data, insight, and reasoning.
• Culture – Any company’s use of AI tools is likely to be viewed by employees with some level of apprehension and sometimes outright fear. So, it is important to share how your business strategy will leverage the technology, and it is leadership’s intent to develop the necessary skills and expertise of the organization to take advantage of Gen AI. Investments to train existing employees to stay current with developments in the AI field will be important, as well as becoming more open to hiring ML engineers and other Gen AI experts from outside your own industry. Some companies often designate a single person to coordinate Gen AI activities across the company, inclusive of ensuring development of basic Gen AI literacy across the workforce. Another approach is to form a cross-functional team to serve as a liaison between senior leaders and employees, to help accelerate the integration of the technology while bringing cultural sensitivity to the undertaking. “Engaging employees early and sharing your AI vision can ease job-related concerns,” states Allison. “At Earlbeck Gases, we involve our team through role-specific training sessions, demonstrating how AI can be integrated in ways that align with our core values, mission, and goals.”
• Choose Partners Wisely – As companies start to develop their Gen AI literacy and capabilities, many will want to find external partners to help them begin forming, executing and accelerating their Gen AI strategy. Look for partners with extensive Gen AI experience and expertise. The list of experienced companies is not very long…
Focus on areas of transactional intensity,” states Dave. “Large volumes of repetitive activity are great fits for Gen AI and can lead to greater value and payback.
Next Steps
In Deloitte’s quarterly survey of companies (who have active Gen AI pilots in the works), 68% of the respondents said their organization has moved 30% or fewer of their Gen AI pilots fully into production. This isn’t necessarily surprising, even though there’s been rapid and impressive advances in Gen AI’s capabilities, its applications are still relatively new, and organizations are figuring out what it can (and can’t) do well. Despite those hurdles, the appetite for Gen AI capabilities has not slowed. Whether you are looking to defend your position against competitors, differentiate products/services from your peers, or looking to develop something complete new not seen in the marketplace, Gen AI is a technology that can assist you in each of those scenarios. So, why wait, it’s time to dive in!
