We'll email you when new articles are published on this topic. At Emirates Team New Zealand, after the AI agents recommended the top designs from the thousands they tested, the sailors then took the helm of the digital simulator once again to test the best hydrofoils and prioritize the final selections. In 2007 Fei-Fei Li, the head of Stanfords Artificial Intelligence Lab, gave up trying to program computers to recognize objects and began labeling the millions of raw images that a child might encounter by age three and feeding them to computers. Applying ML in a basic transactional processas in many back-office functions in bankingis a good way to make initial progress on automation, but it will likely not produce a sustainable competitive advantage. In all cases, whether building or rebuilding digital simulators, organizations should think beyond their existing use cases and make design choices that provide flexibility in supporting more advanced use cases that might not yet be on their radar. If distributed autonomous corporations act intelligently, perform intelligently, and respond intelligently, we will cease to debate whether high-level intelligence other than the human variety exists. Probabilistic: An automation solution that uses statistical functions to predict output based on trained behavior (If A, then most probably B). Unlike other types of machine learning, reinforcement learning uses algorithms (which often train AI agents or bots) that typically do not rely only on historical data sets, either labeled or unlabeled, to learn to make a prediction or perform a task. The unmanageable volume and complexity of the big data that the world is now swimming in have increased the potential of machine learningand the need for it. We have seen how quickly the technological environment can shift. The healthcare company built an ML model to screen up to 400,000 candidates each year. MLOps: The application of DevOps concepts to operationalize machine learning. Teams can use first principles to ballpark potential costs, and leaders should understand and discuss the potential cost drivers with their teams up front to help ensure a smoother process and free teams to focus on the work ahead. Emirates Team New Zealand, for instance, was able to test multiple designs simultaneously (something the sailors could never do), test tenfold more designs under more conditions than had previously been possible, and gain insight from the AI agent into new ways their sailors could execute on these boat designs on the water. The experts also had to think through real-world constraints that humans often take for granted. Because machine learnings emergence as a mainstream management tool is relatively recent, it often raises questions. As a result, all customers tagged by the algorithm as members of that microsegment were automatically given a new limit on their credit cards and offered financial advice. At the same time, models wont function properly if theyre trained on incorrect or artificial data. Some DACs will certainly become self-programming. Machine learning is based on a number of earlier building blocks, starting with classical statistics. A common refrain is that the three most important elements required for success are data, data, and more data. Manufacturers that have already used machine learning to minimize product defects can now expand their insights with reinforcement learning to prevent the rare remaining defects that pop up intermittently with seemingly no common root cause. Use an alternative data set with similar features: Rather than creating a data set from scratch, the team can find an alternative with similar features and behavior of the production data set. Human in the loop: In situations where the data set is available only in the production environment (often for legal reasons) or data quality is sparse, the delivery team will want to gradually create the outputs via manual processing and use those to train and iteratively improve the ML model. Thus, we begin by recounting Emirates Team New Zealands journey, after which we offer ideas for where and how businesses should consider applying reinforcement learning. It also can help teams manage complex manufacturing processes. These wing-like structures attach to the hull and lift the boat above the water, enabling the vessel to reach speeds of over 50 knots (60 miles or 100 kilometers per hour). Please try again later. Translators can bridge the disciplines of data, machine learning, and decision making by reframing the quants complex results as actionable insights that generalist managers can execute. Using reinforcement learning, experts from Emirates Team New Zealand, McKinsey, and QuantumBlack (a McKinsey company) successfully trained an AI agent to sail the boat in the simulator (see sidebar Teaching an AI agent to sail for details on how they did it). At Emirates Team New Zealand, the development team drew from such frameworks where possible and then focused on the value-added tasks that hadnt yet been commoditized. The latest iterations in reinforcement learning algorithms, such as soft actor-critic, are dramatically improving training efficiency, substantially driving down compute costs. But by the time they fully evolve, machine learning will have become culturally invisible in the same way technological inventions of the 20th century disappeared into the background. With this, adoption is increasing, and in a few years, we anticipate that reinforcement learning will become more common in many industries, such as telecom, pharmaceuticals, and advanced industries. Leaders looking for new ways artificial intelligence (AI) can provide a competitive edge may have found the 2021 Americas Cup Match as exciting for one teams groundbreaking use of reinforcement learning as for its radical boat designs and close races. This can often be a question of data management and qualityfor example, when companies have multiple legacy systems and data are not rigorously cleaned and maintained across the organization. Set up an artificial production environment: If a data set is available for the production environment, companies can create a simulated, preproduction environment that uses the data for training purposes without live systems used by end users. Even though ML models can be trained in any of these environments, the production environment is generally optimal because it uses real-world data (Exhibit 3). The value at stake is significant. The same foundational practices and organizational and cultural changes in which enterprises are already investing for other AI also apply to reinforcement learning. Frontline managers, armed with insights from increasingly powerful computers, must learn to make more decisions on their own, with top management setting the overall direction and zeroing in only when exceptions surface. That pattern was accompanied by a steep decrease in their savings rate. For example, if a company wanted to train an ML algorithm to distinguish cats from dogs, it would show two collections of images and clearly delineate which are cats and which are dogs. Subject-matter experts and data scientists gave the agent examples to learn from and established rewards for the agent to guide its choices, including the sacrifice of short-term benefits for long-term benefits. See Bruce Fecheyr-Lippens, Bill Schaninger, and Karen Tanner, . Subject-matter experts and data scientists need to constantly refine incentives, commonly known as reward hacking, to figure out how to properly calibrate rewards to enable an agent to make complex decisions optimally. Asking managers of siloed functions to develop individual use cases can leave value on the table. Think end to end. A recent McKinsey Global Survey, for example, found that only about 15 percent of respondents have successfully scaled automation across multiple parts of the business. Never miss an insight. A frequent concern for the C-suite when it embarks on the prediction stage is the quality of the data. Outlining the reward function to enable an AI agent to learn effectively requires as much art as science, often making it the costliest part of the development process. Research is currently under way into whats called offline reinforcement learning, where the learning is done exclusively on existing empirical data, rather than through simulation. Something went wrong. Please try again later. Understanding the teams experience can help leaders gauge where and when to use the technology because many organizations will travel a similar path: implementing more traditional technologies first to solve a problem and then applying reinforcement learning to ascend to a previously unattainable tier of performance. At the same time, the cost of compute itself has declined significantly. Manufacturers and consumer-packaged-goods companies are under pressure to build dynamic supply chains that account for climate, political, and societal shifts anywhere in the world at a moments notice. C-level executives will best exploit machine learning if they see it as a tool to craft and implement a strategic vision. But as they define the problem and the desired outcome of the strategy, they will need guidance from C-level colleagues overseeing other crucial strategic initiatives. Machine learning: Advanced algorithms that can learn from data without relying on rules-based programming. Also, new tools and strategies enable teams to manage the compute they use. This article was edited by Christian Johnson, a senior editor in the Hong Kong office. Too often, departments hoard information and politicize access to itone reason some companies have created the new role of chief data officer to pull together whats required. IBMs Watson machine relied on a similar self-generated scoring system among hundreds of potential answers to crush the worlds best Jeopardy! Operationalizing ML is data-centricthe main challenge isnt identifying a sequence of steps to automate but finding quality data that the underlying algorithms can analyze and learn from. Think about archetypical use cases, development methods, and understand which capabilities are needed and how to scale them. DevOps: A set of practices that combine software development and IT operations. In addition, as the Emirates Team New Zealand agents knowledge of sailing increased over time, the sailors began learning maneuvers from the agents that they had not considered, enabling them to improve their performance for a given design. In this way, the agents quickly reached a level of mastery to outperform world-champion sailors in the simulator and begin testing design concepts for the team. We find the parallels with M&A instructive. But that means putting strategy first. They have also built microtargeted models that more accurately forecast who will cancel service or default on their loans, and how best to intervene. It is, after all, not enough just to predict what customers are going to do; only by understanding why they are going to do it can companies encourage or deter that behavior in the future. As ever more of the analog world gets digitized, our ability to learn from data by developing and testing algorithms will only become more important for what are now seen as traditional businesses. In Europe, more than a dozen banks have replaced older statistical-modeling approaches with machine-learning techniques and, in some cases, experienced 10 percent increases in sales of new products, 20 percent savings in capital expenditures, 20 percent increases in cash collections, and 20 percent declines in churn. This way of learning is just one aspect of reinforcement learning that makes it different from other AI techniques (see Exhibit 1 and An executives guide to AI for more on the different types of machine learning). Reinforcement learning therefore might not be well suited to situations where regulators or operators require transparency. ML has become an essential tool for companies to automate processes, and many companies are seeking to adopt algorithms widely. This leaves leaders with little guidance on how to steer teams through the adoption of ML algorithms. The role of humans will be to direct and guide the algorithms as they attempt to achieve the objectives that they are given. Todays cutting-edge technology already allows businesses not only to look at their historical data but also to predict behavior or outcomes in the futurefor example, by helping credit-risk officers at banks to assess which customers are most likely to default or by enabling telcos to anticipate which customers are especially prone to churn in the near term (exhibit). Statistical inference does form an important foundation for the current implementations of artificial intelligence. Please try again later. Something went wrong. We'll email you when new articles are published on this topic. Please email us at: Coca-Cola: The people-first story of a digital transformation, Americans are embracing flexible workand they want more of it, The potential value of AIand how governments could look to capture it. By building ML into processes, leading organizations are increasing process efficiency by 30 percent or more while also increasing revenues by 5 to 10 percent. While design rules for the Americas Cup specify most components of the boat, they leave enough freedom for designers to make radical choices on some key elements such as hydrofoils. To sail as well as the worlds best sailors, the AI agent needed to learn to execute different maneuvers in varying conditions, choosing the best course to set under a wide variety of winds and seas, adjusting 14 different boat controls accordingly, assessing the results of its decisions, and continually improving decisions over long time horizons. A transportation company, for example, can optimize travel routes in real time based on changing traffic, weather, and safety conditions.