For most of our history, weâve thought that learningâthe ability to adjust our behavior based on collected informationâwas something only humans did. The past few decades have changed all that. We now know that animals of all kinds learn from experience, teaching, and even play. But it is not only animals that learn: thereâs increasing evidence that plants do, too. And if youâve ever unlocked a phone with facial recognition, or interacted with a virtual assistant, youâve experienced firsthand that machines, too, are capable of learning.
Michael Chui is a partner of the McKinsey Global Institute and is based in McKinsey’s San Francisco office. Tamim Saleh is a senior partner based in McKinsey’s London office. Alex Singla is a senior partner based in McKinsey’s Chicago office. Alex Sukharevsky is a senior partner in the London office.
Machine learning is a form of artificial intelligence (AI) that can adapt to a wide range of inputs, including large data sets and human instruction. (Some machine learning algorithms are specialized in training themselves to detect patterns; this is called deep learning, which we explore in detail in a separate Explainer.) The term âmachine learningâ was first coined in 1959 by computer scientist Arthur Samuel, who defined it as âa computerâs ability to learn without being explicitly programmed.â It follows, then, that machine learning algorithms are able to detect patterns and learn how to make predictions and recommendations by processing data and experiences, rather than by receiving explicit programming instruction. The algorithms also adapt in response to new data and experiences to improve over time.
Today, the needâand potentialâfor machine learning is greater than ever. The volume and complexity of data that is now being generated is far too vast for humans to reckon with. In the years since its widespread deployment, machine learning has had impact in a number of industries, including medical-imaging analysis and high-resolution weather forecasting.
Machine learning as a discipline was first introduced in 1959, building on formulas and hypotheses dating back to the 1930s. But it wasnât until the late 1990s that machine learning truly flowered, as steady advances in digitization, computing languages capable of greater nuance, and cheaper computing power and memory enabled data scientists to train machine learning models to independently learn from data sets rather than rely on rules written for them. The broad availability of inexpensive cloud services later accelerated advances in machine learning even further.
Deep learning is a more advanced version of machine learning that is particularly adept at processing a wider range of data resources (text as well as unstructured data including images), requires even less human intervention, and can often produce more accurate results than traditional machine learning. Deep learning uses neural networksâbased on the ways neurons interact in the human brainâto ingest and process data through multiple neuron layers that can recognize increasingly complex features of the data. For example, an early neuron layer might recognize something as being in a specific shape; building on this knowledge, a later layer might be able to identify the shape as a stop sign. Similar to machine learning, deep learning uses iteration to self-correct and to improve its prediction capabilities. Once it âlearnsâ what a stop sign looks like, it can recognize a stop sign in a new image.
This technological advancement was foundational to the AI tools emerging today. ChatGPT, released in late 2022, made AI visibleâand accessibleâto the general public for the first time. ChatGPT, and other language models like it, were trained on deep learning tools called transformer networks to generate content in response to prompts. Transformer networks allow generative AI (gen AI) tools to weigh different parts of the input sequence differently when making predictions. Transformer networks, comprising encoder and decoder layers, allow gen AI models to learn relationships and dependencies between words in a more flexible way compared with traditional machine and deep learning models. Thatâs because transformer networks are trained on huge swaths of the internet (for example, all traffic footage ever recorded and uploaded) instead of a specific subset of data (certain images of a stop sign, for instance). Foundation models trained on transformer network architectureâlike OpenAIâs ChatGPT or Googleâs BERTâare able to transfer what theyâve learned from a specific task to a more generalized set of tasks, including generating content. At this point, you could ask a model to create a video of a car going through a stop sign.
Foundation models can create content, but they donât know the difference between right and wrong, or even what is and isnât socially acceptable. When ChatGPT was first created, it required a great deal of human input to learn. OpenAI employed a large number of human workers all over the world to help hone the technology, cleaning and labeling data sets and reviewing and labeling toxic content, then flagging it for removal. This human input is a large part of what has made ChatGPT so revolutionary.
Convolutional neural network (CNN). CNNs are a type of feed-forward neural network whose connectivity connection is inspired by the organization of the brainâs visual cortex, the part of the brain that processes images. As such, CNNs are well suited to perceptual tasks, like being able to identify bird or plant species based on photographs. Business use cases include diagnosing diseases from medical scans or detecting a company logo in social media to manage a brandâs reputation or to identify potential joint marketing opportunities.
Recurrent neural network (RNN). RNNs are artificial neural networks whose connections include loops, meaning the model both moves data forward and loops it backward to run again through previous layers. RNNs are helpful for predicting a sentiment or an ending of a sequence, like a large sample of text, speech, or images. They can do this because each individual input is fed into the model by itself as well as in combination with the preceding input.
Continuing with the banking example, RNNs can help detect fraudulent financial transactions just as feed-forward neural networks can, but in a more complex way. Whereas feed-forward neural networks can help predict whether one individual transaction is likely to be fraudulent, recurrent neural networks can âlearnâ from the financial behavior of an individualâsuch as a sequence of transactions like a credit card historyâand measure each transaction against the personâs record as a whole. It can do this in addition to using the general learnings of the feed-forward neural network model.
For more on deep learning, and neural networks and their use cases, see our executiveâs guide to AI. Learn more about McKinsey Digital.
McKinsey collated more than 400 use cases of machine and deep learning across 19 industries and nine business functions. Based on our analysis, we believe that nearly any industry can benefit from machine and deep learning. Here are a few examples of use cases that cut across several sectors:
Canny leaders have been applying machine learning to business problems for years. Here are a few examples:
To help capture the full potential value of AI and machine learning technologies, mainstream adopters can consider the following actions:
Machine learning is here to stay. Gen AI has shone a light on machine learning, making traditional AI visibleâand accessibleâto the general public for the first time. The efflorescence of gen AI will only accelerate the adoption of broader machine learning and AI. Leaders who take action now can help ensure their organizations are on the machine learning train as it leaves the station.
Learn more about McKinsey Digital. And check out machine learning-related job opportunities if youâre interested in working with McKinsey.
