In the concentric circles of artificial intelligence, machine learning (ML) is a subset that learns from data and improves over time. Within ML, deep learning takes this further by extracting insights from complex patterns in large datasets. This field is evolving rapidly: When tasked with working on problems that human minds struggle to fully grasp in their complexity, ML is capable of expanding the boundaries of what we thought was possible.

That’s exactly what happened when Google subsidiary DeepMind cracked a protein-folding problem that had stumped researchers for 50 years. The results are now being used to create novel drugs to treat diseases such as malaria. This is also the science behind the predictive maintenance of smart manufacturing, autonomous vehicles, and even more accurate weather forecasting.

A lot of investment is being poured into ML and the business use cases are multiplying as its algorithms grow in maturity. I have seen firsthand how one of my customers has developed systems that serve as the “eyes” for ML applications across manufacturing, medical and life sciences, aerospace, and intelligent transportation systems. This machine vision technology collects and processes large volumes of visual data to generate insights that inform better decision-making.

But much of this happens at lightning speed, so let’s break down exactly how ML is processing more data across more dimensions, gaining deeper insights, and expanding human knowledge.

MORE DATA COLLECTION

ML doesn’t just enable the collection of much larger datasets; it also allows that data to be processed continually. For instance, researchers were confounded by the problem of predicting protein structures because there are tens of thousands in the human body and they can fold up into a vast number of three-dimensional shapes in seconds. 

These structures can result in disease when altered or misfolded, and before ML, figuring out the shape of just one would take years. Today, ML has predicted the structures of almost every known protein—about 200 million—based on their amino-acid sequences and by constantly refining its datasets to improve accuracy. 

1. Purpose-driven approach: Clearly define your objectives and align your ML strategy with specific business goals.

2. Model selection: Choose ML models that best suit your data type and business objectives, considering factors like interpretability and scalability.

3. Data comprehension: Gain a deep understanding of your data’s characteristics, quality, and potential biases before feeding it into ML models.

Smart manufacturing is one great example of the continual collection and processing of large datasets through sensors along the production line. ML algorithms are able to analyze this data in real time, anticipating maintenance needs before breakdowns can occur.

MORE DIMENSIONS

Across industries, it’s not just about data quantity but complexity. 

ML can analyze multiple dimensions of data simultaneously, including factors that might have been previously overlooked or difficult to measure. Predicting heart disease is a great case study, with recent research revealing a new ML model had achieved an unprecedented rate of 95% accuracy through analyzing four international databases across an original 76 different attributes. That dataset was further narrowed down to 13 input features in the final model.

To simplify how this works at the business level, think of analyzing data like your phone’s photo memories feature. It looks at your photos from different angles—dates, places, events—to find meaningful patterns (such as recurring celebrations). Similarly, examine your business data from various perspectives. Tools like Tableau Public and Power BI can combine data from multiple sources to identify patterns that are meaningful to your business—much like photo memories help people rediscover important moments.

CORRELATION DISCOVERY

The novelty of being able to process larger amounts of data across more dimensions naturally gives rise to more nuanced and impactful insights. Beyond that, however, cross-referencing data to this level of granularity gives rise to new knowledge that can reshape how we see the world.  

One key is ML’s capacity to detect patterns and anomalies in “noisy” data streams. NASA reports that scientists recently added 301 new exoplanets—planets on other solar systems—to their exoplanet tally thanks to a new deep neural network called ExoMiner. By constantly refining its discoveries, ExoMiner can distinguish real exoplanets from “false positives”—something it shares with the cardiovascular disease prediction model. 

The business use cases of correlation discovery range from using insights into hidden patterns of consumer behavior for better personalization to adapting farming practices based on how microclimate variations impact crop yields. As exciting as these potentials are, however, the ML implementation journey does not need to be rushed.

CHALLENGES AND CONSIDERATIONS

One of the primary concerns when implementing machine learning is change management. As ML systems are introduced, employees may fear losing their jobs. To mitigate this, focus on education and transparency. Also, involve team members in every step of the implementation journey as you:

⊗ Look for areas where data volume exceeds human capacity to handle efficiently.

⊗ Ensure you have sufficient, high-quality data and a clear potential for real benefits.

⊗ Start with small, incremental improvements rather than drastic overhauls. 

When it comes to ethical considerations, particularly around data privacy and potential biases in ML algorithms, I recommend establishing a clear understanding of legal and ethical boundaries in your industry and region. Where there are regulatory gray areas, prioritize openness and disclosure about your data practices.

THE POSSIBILITIES ARE LIMITLESS

We will never fully know how far machine learning will expand human knowledge because it is always an unfolding process. As ML learns more, so will we.

The caveat for businesses is to weigh cost versus realizable benefits, rather than rely on potential. The beauty of ML is it never stops working. It is logical to assume solutions that were not possible before will become possible, perhaps faster than we anticipate.

LeackStat 2024