A CTO’s Perspective on Gaining Value from AI
Author: Didrik Pinte, M.S., CTO
AI – Leading While Lagging
Artificial intelligence has never been more widely applicable. Given the rapidly increasing rate of data generation, it can be thought of as never so under utilised. Many science-based organizations globally are investing in AI and digital tools, without always getting the benefits promised. According to MIT, “40% of organizations making significant investments in AI still do not report business gains from AI. As with technology advances in the past, technology alone isn’t the answer to value.” 
This is a significant problem. Scientific advances that would otherwise be greatly accelerated by AI (and other commonly available technologies) are in fact inhibited.
Despite the emphasis on AI (and its sub-categories of machine learning and deep learning), few organizations are leveraging the full value of this technology. There are other critical elements of the issue that get less attention (domain expert understanding of AI’s potential) or more (today’s version of ‘…you need a data lake’), and need to become part of an integrated approach. To make uniform progress on today’s scientific challenges this blog post will consider three critical elements;
- Digital Strategy
- Digital Skills
- Data & Data Flow
As time passes, gaps that appear when these are not adequately addressed widen – most young, agile software companies will iterate and develop quicker than the mature, large organizations they sell into. Organizational maturity in digital capabilities progress is disjointed, and over time, the strategy, skills, and data gaps become increasingly large.
Digital Strategy – the Gap Between Ambition and Execution
The Digital Strategy gap exists where technology has been adopted in an ad hoc manner. Successful implementation of any digital tool requires strategic intent and direction, developed in the context of the business challenges of the individual organization. Why are we adopting this technology? Is it delivering business impact to our organization or laboratory and its strategic goals and objectives? Answering these questions requires an underlying digital strategy, which many organizations have not dedicated enough effort.
The data on AI adoption supports this, showing a significant gap between ambition and execution. An MIT Sloan and BCG global review found that “almost 85% believe AI will allow their companies to obtain or sustain a competitive advantage. But only about one in five companies has incorporated AI in some offerings or processes. Only one in 20 companies has extensively incorporated AI in offerings or processes. Less than 39% of all companies have an AI strategy in place. The largest companies — those with at least 100,000 employees — are the most likely to have an AI strategy, but only half have one.”
Enthought’s experience of organizations with limited digital strategy is that teams within the organization will often enjoy project-by-project successes as a result of the adoption of AI and other digital tools. However, these successes do not translate across, or scale within, the organization.
Digital Skills – Where AI Does (or Doesn’t) Meet Domain Expertise
Getting value from AI requires a corresponding investment in developing skills, ones that can be very challenging to estimate. One part is developing the skills to leverage AI-based tools and technologies. The other, very often overlooked, is developing deep experts with an understanding of the potential of AI. Said another way, they may not know how to develop it, but they sure know what it can do, and are able to guide those who do know how to develop AI.
Without this domain expert understanding, there will be incremental advances that have the organization believe it is realizing the full potential of AI. In reality, its orders of magnitude possibilities are being missed. Sam Ransbotham, professor of information systems at the Carroll School of Management at Boston College says, “AI talent can be a particularly difficult limitation. Once armed with technology and infrastructure, many organizations find that they don’t have the AI skills they need.”
Tesla is an organization setting the standard in hiring for digital skills, recognizing that today’s education programs may not necessarily meet the needs of the rapidly evolving digital landscape. Tesla CEO Elon Musk said in a tweet; “A PhD is definitely not required. All that matters is a deep understanding of AI & ability to implement NNs [Neural Networks] in a way that is actually useful (latter point is what’s truly hard). Don’t care if you even graduated high school.”
Musk highlights an important point – implementing digital tools in a useful way is not easy. What is not highlighted, and is equally important, is domain expertise, a deep understanding of the underlying science of the problem being taken on. Domain expertise is critical to be able to apply AI to real-world problems.
Ransbothan further highlights the need for digital skills, “According to the 2019 MIT SMR-BCG Artificial Intelligence Global Executive Study and Research Report, organizations that actively help their existing workforces gain AI skills are more likely, by 40 percentage points, to generate value from AI than companies that have not focused on reskilling.”
Data: The Third–and Foundational–Gap
Digital consultancies and software development houses (described by one Pharmaceutical CIO as “bright but very young children”) produce sophisticated digital tools and data infrastructure technologies. The efficacy of these tools in isolation is impressive. Yet the execution in large organizations is where these technologies meet barriers. And thus, we meet the third gap through which these technologies often fall; the data gap.
AI is not natively intelligent. Algorithms and models learn by analyzing data. The outputs of any deep learning or machine learning-based technology will only be as strong as the inputs, i.e., the data fed into them. Many research labs today produce significant volumes of data, increasingly through automation. The capture, flow, storage, and accessibility of these data are where problems present. Where scientists cannot access and provide fit for purpose data (meaning of suitable quality), AI models will not produce quality outputs.
The mirror image challenge is where there are small data sets. In many natural sciences and engineering disciplines, large volumes of data can be hard or expensive to generate. The reality is these datasets are often limited in size, poorly curated, and bespoke to particular problems. Coupling domain expertise with an understanding of the potential of AI and other digital tools like modeling and simulation is the mandatory key to success.
Last but not least, companies tend to want to develop AI programs to replace manual experimentation. Such a transition is only possible by taking many more data points to train the future AI models, more than likely through automation. Many managers refuse to accept that the path to AI requires investing in their experiment generation and data capture programs.
Research shows those organizations leading in the industry have a more innate understanding of the value of data, as compared to their peers who lag behind similar sectors.
Outsourcing – A Key Ingredient
Entire sectors are emerging as technologists see an opportunity to help organizations make the most of digital technology advances and form companies to meet the need. Ransbothan writes, “…when an organization doesn’t have the necessary talent to produce results with a technology like AI, it can look outside for help. Unfortunately, outsourcing delivers AI-related value to only 12% of organizations that report using this approach.”
To be truly effective in a digital journey, working with external teams can require a multi-year collaboration, with dedicated outsourced teams working more like an extension of the organization, than an outsourced service provider. In Enthought’s experience, the most successful digital transformation initiatives have occurred where the engagement is 3-5 years, with a keen focus on skilling client scientists with new digital techniques.
The remote working and virtual experiences of the COVID-19 period should be an accelerant of value through external collaboration in advanced scientific computing.
And Finally – People and Culture
Organizations – and science-driven labs in particular – understand that AI capabilities must advance in parallel with those other, interdependent areas. One way to frame it; it is about technical experts making better decisions earlier in all workflows, delivering value continuously to the business. To do this, AI must become integral in the processes and workflows of the technical community, to be able to execute quickly, using speed as an advantage over competitors.
Ransbothan writes, “Acquiring the right AI technology and producing results, while critical, aren’t enough. Instead, to gain value from technologies like AI, the company needs to focus on those employees who will consume the AI results that the organization produces.”
There is a foundational need for establishing a data culture, one that complements – rather than competes with – the deep experience of the existing scientific community. But that is a topic for another time.
About the Authors
Didrik Pinte, M.S., CTO at Enthought, holds an M.S. in bio-engineering and an M.S. in management from the Catholic University of Louvain (UCL) in Belgium. He is an expert in artificial intelligence, data management, and software development. He served as a research assistant at UCL, developing Python-based integrated water resource management applications.
Sam Ransbotham (cited throughout this post) is a professor of information systems at the Carroll School of Management at Boston College and the MIT Sloan Management Review guest editor for the Artificial Intelligence and Business Strategy Big Idea Initiative.
Making the Most of Small Data in Scientific R&D
March 9, 2023|Life Sciences, Materials Science, Transformation Making the Most of Small Data in Scientific R&D For many traditional innovation-driven organizations, scientific data is generated…
ChatGPT on Software Engineering
Recently, I’ve been working on a new course offering in Enthought Academy titled Software Engineering for Scientists and Engineers course. I’ve focused on distilling the…
What’s in a __name__?
if __name__ == “__main__”: When I was new to Python, I ran into a mysterious block of code that looked something like: def main(): …
3 Trends for Scientists To Watch in 2023
As a company that delivers Digital Transformation for Science, part of our job at Enthought is to understand the trends that will affect how our…
Accelerating Science: the Classical Mechanics Perspective
When thinking about enhancing R&D processes, Newton’s second law of motion provides the perfect framework. Classical mechanics teaches us that putting a body into motion…
Retuning the Heavens: Machine Learning and Ancient Astronomy
What can we learn about machine learning from ancient astronomy? When thinking about Machine Learning it is easy to be model-centric and get caught up…
Announcing Enthought Academy
Dear Students and Friends of Enthought, I am pleased to announce Enthought Academy—the culmination of over twenty years of teaching Scientific Python. Since our founding…
Extracting Target Labels from Deep Learning Classification Models
In the blog post Configuring a Neural Network Output Layer we highlighted how to correctly set up an output layer for deep learning models. Here,…
True DX in the Pharma R&D Lab Defined by Enthought
Enthought’s team in Japan exhibited at the Pharma IT & Digital Health Expo 2022 life sciences conference in Tokyo, to meet with pharmaceutical industry leaders…
Exploring Python Objects
Introduction When we teach our foundational Python class, one of the things we do is make sure that our students know how to explore Python…