Learning About Machine Learning: Part II
Learn how AI is being used to empower business strategies and what is needed for organizations to succeed in pursuit of AI initiatives.
In Part I of this series, I discussed my decision to go back to school to learn more about the business strategy and implications of artificial intelligence (AI), the difference between machine learning and deep learning, and the ethical considerations related to the expansion of AI’s influence on our lives.
In this installment, I will share some reflections on how AI is being used to empower business strategies and what is needed for organizations to succeed in pursuit of AI initiatives.
The success of Facebook, Amazon, Netflix, Google and other high growth companies has been partly attributed to their successful application of machine learning upon the vast flow of data that is collected by these FANG companies. Inspired by the expectation of similar success, many organizations are starting to adopt AI for their business.
Strategically, choices have to be made to match the appropriate AI data sources, resources, formulation, infrastructure and techniques to the overarching business strategy. The following are some observations from my coursework and experience.
A new emerging strategy
AI can be applied to organizational competitive strategy to enable Michael Porter’s classic trio of cost leadership, differentiation and focus.
Based on experience, I would add a fourth emerging strategy empowered by AI: cultural transformation to develop organizational collective intelligence, particularly for knowledge workers, information analysts, customer enablement and supply chains.
Knowledge graphs are the next wave in organizational initiatives that are transforming legacy business culture into data-driven decision cultures.
Do you want your business to become collectively smarter or continue to do business with pockets of excellence and high friction, time-consuming access to expertise? Why not make subject matter expertise instantaneously available, in context, to the right employee, to customers, or to your key supply chain stakeholders?
Effective ML solutions lower the costs of prediction. From the C-suite to senior managers and contact center teams, organizations are looking for answers, whether months over the horizon or in the very next moment, to determine the optimal action, offer, investment or dedication of resources. To do so, workers make countless predictions with cloudy or limited information upon which to base their prediction. If you could augment your organization’s predictions at a lower cost, would you?
Enhance your organization with AI
The best early stage AI applications involve ML or DL trained machines that augment human behavior or enhance decision-making. Early stage AI initiatives are most successful when they are tied to:
- Improving existing use cases that can be executed more effectively by machines or which can augment human decisions to help organizations become smarter about their customers, capabilities and competitive environment.
- Human decision-making augmentation provided by algorithmic models incorporating a learning loop, a continually refined cycle of human and machine as "peer."
- Use cases that are routine, involve aspects of perception (see, hear, read, find), involve aspects of learning or contextual linkage, or which automate manual tasks.
In one study, a ML-trained model recognized 92.5% of early stage cancer cell images, meaning a 7.5% error rate – unacceptable and below trained doctor recognition that had a 3.5% error rate.
However, working together the model and doctors collaboratively reduced their error rate to 0.5%, a tremendous improvement that only gets better – and saves more lives – the more that machine and human work together.
It's a data science
The effectiveness of an AI-derived solution is tied to the quality of the data and the skill applied to formulating its contributory data sources. This is why talented data scientists and data engineers working under strong data governance can develop effective AI models. It’s not just about data volume and variety; creating value from data requires data hygiene.
Increasingly, algorithms will become less proprietary while data becomes more valuable. Data science will become less artisanal and more directed at pointing ML and DL muscle - or whatever technique comes next - at the data that an organization can access.
The last several years, there has been an explosion of software, tools, algorithmic frameworks, and high-performance compute and storage infrastructure dedicated to making AI solution development less difficult and more agile.
Getting started with machine learning is affordable; getting to scale at significant production levels requires massive data flow, which means higher CapEx or OpEx for infrastructure and policy issues related to data strategy and data governance.
It cannot be overstated that developing effective AI solutions is difficult work. Getting a machine learning model to a 90%-93% confidence level can be done rapidly. Getting that same model to the 99% or greater confidence levels frequently needed for successful AI implementation is exponentially more challenging.
The growing effect of AI
The last few years have brought us an explosion of OEM architectures, software, tools, and containers to accommodate the growing marketplace demand for greater agility, performance, and data to feed AI solutions.
Machine learning is growing, as demonstrated by its surging popularity on college campuses and in online learning. AI can provide significant advantages if applied to the appropriate business situations.
I encourage C-level execs, technologists, and digital transformation leaders to take some time to learn about AI and understand the uses, advantages, infrastructure needs and constraints of machine learning. You will gain more than enough understanding to point your organization in the right direction and invest your resources to support your business strategy.
You can take the same courses I took. Courses on machine learning and deep learning are available on Coursera and via other online sources.
MIT Sloan and the MIT Computer Science AI Lab (CSAIL) offer an excellent course on the business implications of AI and organizational strategy. If you have a computer science or developer background, you will be surprised by how quickly you can use deep learning to build and train your own computer vision model that will recognize humans, objects and creatures, using an open source image library and existing frameworks to train your model.
Machine learning is not only in your future, it’s already in your life.