Abstract
Computers and applications of computers into our world have changed dramatically during the past five decades, from early days of minimal central processing unit capacity, limited memory and without advantage of global networking. In this article, the author highlights the application of predictive artificial intelligence in use globally over the last 40 years in process industries. It discusses the novel application of process automation and robotics in health clinical high-volume laboratory use that began as a Canadian innovation initiative and followed by similar innovation extending to other countries.
Introduction
The author is chronicling milestone accomplishments applying advanced Artificial Intelligence (AI) including international work across multiple countries and her work with Shell International (1988). She also focuses on the healthcare domain through her work with MDS Laboratories renamed LifeLabs (1993-1995), with the Provincial Health Services Authority in British Columbia (BC) (2003-2005), and with the Mayo Clinic Platform, a partnership with Google (2023).
In the landscape of AI, the spotlight often shines on the latest generative AI marvels, captivating the public with their limitless creativity. Yet, quietly behind the scenes, predictive AI modelling has been silently shaping our world for over four decades, orchestrating monumental changes often unseen by the public eye. As we navigate this terrain where innovation meets apprehension, it becomes evident that not every AI endeavour is met with open arms; concerns for safety and reliability loom large.
With each new advancement, there is a simultaneous balance between optimism and trepidation. While the promise of new AI tools ignites hope for a brighter future, it also stirs anxieties within today’s workforce, fuelled by fears of obsolescence and the encroachment of AI-driven processes into the workplace. The notion of machines offering advice or even sharing physical spaces with humans evokes a cocktail of apprehension and uncertainty.
Public trust in AI is a fragile construct, easily fractured by instances of failure or deception. The recent case of Air Canada’s flawed generative AI application, which led to a lawsuit by disillusioned users, serves as a reminder of the repercussions of AI gone awry. In a world where trust is paramount, the stakes could not be higher for ensuring the reliability and integrity of AI applications.
“In a groundbreaking move, the European Union (EU) has launched its bid to set the new comprehensive standard for the regulation of artificial intelligence with the European Parliament passing the EU AI Act on March 13, 2024. This legislation, set to come into effect in the coming years, ushers in a new era in AI regulation and stands as a testament to the EU’s commitment to ensuring a subtle balance between safe, ethical, and innovative use of AI.” 1 At the same time, Canada's federal government is working on Bill C-27 that includes the Artificial Intelligence and Data Act for its citizens.
Amidst these debates, speculation is high regarding the inevitable milestone when machines equipped with advanced software will surpass human capabilities. From triumphs in game theory where computers outmanoeuvre chess grandmasters to the fictional realm of Dr. Susan Calvin’s positronic brain, envisioned by Isaac Asimov, the boundaries between human and machine blur with each technological leap. However, it is crucial to remember that despite these advancements, today’s robots remain bound by their programmed algorithms, devoid of true intelligence or self-awareness.
In this unfolding narrative of human ingenuity and technological prowess, the road ahead is fraught with both promise and peril. As we grapple with the ethical and existential implications of AI, one thing remains certain: the trajectory of our future will be profoundly shaped by how we navigate this new world of artificial intelligence.
Advanced process control: Shell’s multi-variable optimization data model and controller
It is fascinating to delve into the development and implementation of advanced process control systems like Shell’s Multi-Variable Optimization Controller (SMOC). SMOC, along with other controllers like Setpoint’s proprietary IDCOM-M controller and Charlie Cutler’s Dynamic Model Predictive Controller (DMC), represented a significant leap forward in the field of Model Predictive Control (MPC).
The integration of predictive AI models into control systems has indeed revolutionized the optimization of industrial processes, particularly in complex environments like refineries. The ability of these systems to handle non-linear processes and adapt to various constraints is crucial for improving efficiency and quality in production.
The author’s experience with SMOC’s development both in Houston, Texas, at Setpoint’s headquarters and at Shell International’s offices in The Hague, including the challenges faced in convincing the safety committee at the Shell Godorf refinery, underscores the importance of not only developing sophisticated control algorithms but also effectively communicating their functionality and benefits to stakeholders. Clear articulation of how these systems optimize control and enhance operational performance is essential to gain buy-in and ensure successful implementation. Subsequent Shell International projects that the author participated included successful integration of SMOC at operating plants in England, Brunei, and Australia.
The SMOC application has the following two components: (1) input of historical process data sets captured from the local processing plant and resultant data mining application that constructs the optimized matrix using Gaussian elimination to simplify the matrix of equations involving independent and dependent control variables and constraints that comprise the predictive data model; and (2) closed-loop controller that references the predictive data model and manages input of setpoints for independent variables and monitors dependent variables to ensure optimal results that are guided by the historical behaviour. Both Shell’s SMOC and Setpoint’s IDCOM-M controllers represented the third generation of the family of model predictive controllers that continue to evolve today with wide adoption across the control industry globally.
Figure 1 highlights the genealogy of the family of predictive AI controllers across four generations of maturity. Genealogy of MPC algorithms.
The article in the International Journal of Engineering Trends and Technology provides additional context on the historical development and advancement of MPCs over the last four decades, emphasizing its transformative impact on industrial operations. 2
MDS automation program, beginning in Toronto and expanding across North America
The expansion of the MDS (AUTOLAB) automation program across North America in the early 1990s marked a significant milestone in streamlining laboratory processes and improving efficiency in specimen analysis. The collaboration between MDS Inc., MDS Laboratories, People and Process Solutions (PPS), and Labotix demonstrates the interdisciplinary approach required for successful automation initiatives.
The author was included as a member of project team due to her prior experience with automation and process control in manufacturing processes and advanced process control and familiarity with the SETCIM control application product from Setpoint, contributed expertise to the team.
The integration of various applications and technologies into the MDS laboratories automation system highlights the complexity of the endeavour and the need for robust management and coordination. Applications included in the automation project were the MDS LIS/LIMS, SETCIM Process Computer, Gensym G2 graphical expert system, work cell Labotix robotics PLC Controllers, and Johnson control automated storage and retrieval. Each of these solutions played a critical role in optimizing MDS’ laboratory processes, first in their Toronto, Ontario laboratories, and followed soon thereafter at their Burnaby, British Columbia laboratories.
The article by Dr. Stephen R. Middleton on “Developing an Automation Concept That Is Right for Your Laboratory” provides valuable insights into the considerations and best practices involved in implementing automation solutions tailored to the workflows of high-volume clinical laboratories. 3 Process redesign and implementation was focused on chemistry and haematology testing, but there were also process reviews on potential for redesign of microbiology/virology testing.
These efforts towards automation not only improved operational efficiency but also enhanced the quality and reliability of laboratory results, benefiting healthcare providers’ effectiveness and improved patients’ outcomes. MDS AUTOLAB was organized to further expand clinical laboratory automation for health organizations in United States.
Visual automation applied to cytopathology Pap smear screening and future diagnostic outcomes
It is fascinating to see the intersection of automation and AI in the field of cytopathology, particularly in the context of laboratory integration efforts within healthcare systems like the Provincial Health Services Authority (PHSA) in British Columbia.
The author was involved as the Chief Information Officer for THiiNC Health and assumed role of project manager in supporting the transformation of PHSA’s clinical laboratories underscores the importance of leveraging advancements in automation and technology to improve efficiency and enhance patient care. Her trip to Kaiser Permanente (KP) Northwest laboratories in Portland, Oregon, was led by Dixie McFadden (the former director of these laboratories). Kaiser Permanente organization is recognized as one of many leaders in healthcare innovation. This field trip provided PHSA’s senior lab management team with valuable insights into the possibilities of automation and integration, exemplified by KP’s utilization of the updated Labotix automation for core laboratory chemistry serum assays and visual digital automated screening of pap smear specimens in cytopathology. Labotix’ robots that automated specimen handling including decapping, pipetting, aliquoting, and recapping had been reduced in size from the decade earlier MDS implementation and were now hidden from view by KP’s laboratory staff.
Despite the strides made in integrating laboratory services and implementing common Laboratory Information Systems (LIS) across PHSA’s clinical labs, PHSA management and pathologists were reluctant to embrace automation in cytopathology screening process. This highlights the challenges facing recent technologies and workflows in healthcare, particularly in specialized areas like diagnostic pathology.
However, as highlighted in the article by Aziza R. Alrafiah, the application and performance of AI in cytopathology offer promising opportunities for enhancing diagnostic outcomes. 4 Moving beyond screening to achieving diagnostic-level interpretation, AI can play a crucial role in identifying early cancer concerns and improving accuracy and efficiency in pathology workflows. As healthcare systems continue to evolve, it will be essential to address barriers to technology adoption and foster collaboration between stakeholders to realize the full potential of automation and AI to improve patient outcomes in fields like cytopathology.
Mayo Clinic Platform
Dr. John Halamka was serving on the Board of Orion Health concurrent with the author’s time as National Director for Orion Health Canada (2015-2018). She and Dr. Halamka had an occasion to meet on several occasions, which included our global CEO Ian McCrae who was our vocal advocate for enabling precision medicine with focused data analytics on concise patient health information. Orion Health globally was managing over one hundred million patients’ health information in context of health information exchanges, including information held within Canada. Google was seeking opportunity partners during this time to leverage large health data sets in the United Kingdom through the National Health Service, and globally in development of health data lakes. In parallel to Google, IBM Watson made investments to acquire large health data by purchasing four significant data firms: Truven, Phytel, Explorys, and Merge. If there is a lesson learned by these organizations, value does not accrue by securing large volumes of health patient data, rather locating the right health patient data to address needs and questions.
By 2020, John Halamka had taken on role of President of Mayo Clinic Platform. Mayo Clinic and Google embarked on a decade-long strategic partnership with advanced cloud computing and AI-powered analytics at its core. By June of 2023, the two organizations showcased some of the generative AI use cases that they had been working on together.
At this time, John shared: “I have been in academic healthcare for almost 40 years, and one of the challenges with academic healthcare is that every project we do is ad hoc. That is, you get an idea, the innovator talks to your lawyers, 18 months later the contracts are signed, and work finally begins. It is a very inefficient process. What have we done over these last 3 years?” he said of the partnership. “Templated the process. We will go from idea to running code in two weeks. And how does that happen? Well, we took the entire corpus of Mayo data—structured, unstructured, -omics, telemetry, images, digital, pathology—deidentified it, moved it to a cloud container, and now it takes almost no time to bring any innovator into that cloud container to work with the data.” 5
Conclusion
Predictive AI, highlighting the model predictive controllers has been routinely deployed in process industries over the last four decades. Because of successes in other industries, automation in the health industry was an obvious market. This underscores that regardless of the industry, evolution and implementation challenges are the same universally. We have learned from industries who preceded us in healthcare and continue to apply these lessons learned as we move forward with technologies.
Clinical core laboratories were able to leverage similar solutions to the advanced automation and robotics applications that were first demonstrated at MDS facilities Toronto, Ontario, for large-scale clinical specimen core laboratory processing. Canadian innovators such as Cerner Labotix Automation have continued refinements from their first-generation robotics from the 1990s MDS Toronto project including miniaturizing their robotics to ease adoption by workers in companion with clinical laboratory automation. In recent expansion from predictive to generative AI, Mayo Clinic Platform has effectively partnered with Google to develop a health data lake of Mayo’s various clinical data sources to accelerate creation of agile new algorithms for clinical applications.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Ethical approval
Institutional Review Board approval was not required.
