The phrase “humans in the loop” is becoming a watchword in today’s companies in the adoption of artificial intelligence. AI is primarily Enhance The technology is believed to be best deployed alongside a human worker as a co-pilot.
Understanding AI as a technology and its relationship to humans is a stark departure from the traditional vision of full automation that has led to successful adoption of new technologies in business. The automation of financial market making introduced in the 1990s, which is now familiar to us, has been aptly described as a “game-changing event.” Automation in this field has human It removes the need for market makers and enables entirely new ways of trading on a global scale.
But which of these two visions of the future – augmented or fully automated – best suits an AI-powered economy?
Answering this question is important because each approach to technology adoption can lead to dramatically different economic conditions that impact value creation and competitive advantage now and in the near future. For example, when an organization commits to a vision of expansion, the technology itself changes as it is designed. The surrounding area Human workers. As a result, productivity and performance gains are inherently limited by what humans (even “augmented humans”) can accomplish.
Herbert Simon’s Artificial Science This book illustrates these limitations well. Simon, an expert on organizational decision-making, recounts the U.S. State Department’s switch from teletypes to line printers in the 1960s, designed to improve message processing during crises, but which failed because humans were still needed to process large amounts of information. The scaling paradigm of technology adoption can sacrifice much of the economic value of automation: increased standardization, security, speed, and accuracy.
Because humans are slow, serial information processors with such limitations, the gap between computationally powerful machines and “augmented humans” will only widen in the AI era. So will the gap between the economic viability of workflow-level automation and its constituent task-level augmentation. That’s why understanding where full automation may be feasible tomorrow is a powerful guide to investing today, especially in emerging technologies like generative AI. By assessing the obstacles that are well known to prevent full automation, we propose a set of investment criteria to help leaders navigate the mixed feelings of uncertainty and expectation that characterize the dawn of GenAI.
Cognitive Factory Floor
Imagine a bank working on a “human-in-the-loop” model for AI-powered lending. Such a bank needs to design risk assessment algorithms that work, but are limited to what a human worker can reasonably interpret. The volume and speed of credit approvals are also constrained by worker throughput. By comparison, Alibaba’s MyBank, founded in 2015, has no loan officers or human risk analysts. MyBank’s AI risk model is based on over 100,000 variables and can approve loan applications in minutes with a competitive default rate (1.94%), at less than 1% of its peers’ processing costs. The MyBank model is possible because it removes humans and their cognitive limitations from the process, allowing technology to fully automate the complex lending decision process.
In the real, physical world of operations, like the MyBank model, automation efforts tend to: full Automating complex processes. For example, Japan’s FANUC uses robots in unmanned areas to manufacture new robots. Human intervention is only required for routine maintenance and problem solving. Now, three years after installation, the system is so efficient that it can operate unmanned for 30 days at a time and produce 11,000 robots a month. Similarly, the port of Tianjin in China, the world’s seventh largest port,Number– Largest scale — In 2021, the port partnered with Huawei to launch an automated terminal. The port has an AI “brain” that automates and coordinates schedules, remotely operates 76 autonomous vehicles, and manages container movements from more than 200 countries with essentially no human intervention (less than 0.1% versus the industry standard of 4%-5%).
These automation success stories are no longer the exception, but a common pattern runs through each one. Machines work best when they interact with other machines. Humans are too idiosyncratic, even mercurial, to bewildering partners. Karl Marx rightly noted that “the organized system of machines is the most advanced form of mechanical production.” That’s why it’s easier to design a fully automated warehouse (as Amazon does) than it is to design robots that work effectively and safely alongside human workers whose behavior may be unpredictable. The port of Tianjin would not achieve the same performance if half its fleet was operated by humans.
AI essentially extends this logic to a broader range of human behavior, putting many of the most cognitive activities in humans (so-called knowledge work) on par with the factory floor of old. For example, the performance achieved by LLM proves that natural language, despite its subtleties, is sufficiently systematic and pattern-based to be repeatable and, more importantly, amenable to automation. Of course, AI can augment the human workforce, but working independently, it can also lead businesses into bolder automation efforts.
The limits of automation as an investment guide
Let’s return to the conundrum facing many business leaders today: modern AI technologies are great, but where should you invest after the now ubiquitous GenAI-powered chatbots and copilots? In our view, leaders can make better decisions if they can distinguish between technology applications that are a step toward full automation and those that primarily augment human workers. The best way to do this is to focus on known obstacles to full automation; the existence of obstacles indicates the relatively modest returns associated with merely augmenting the use of technology.
Integration constraints: Interfaces. The more a process relies on interfacing with different systems, the harder it is to automate. Consider the infamous case of the London Stock Exchange’s Taurus project, launched in 1983 and abandoned in 1993 after suffering losses of around £75 million. Taurus was intended to automate London’s paper-based stock trading system, but it couldn’t withstand the weight of redesign requirements, and human registrars continued to play the role of “middleman” in the trading process, using their own systems that then had to interface with the stock exchange’s systems.
The interface issue may be alleviated somewhat once autonomous agents gain the ability to directly access, control, and implement changes to external systems (e.g., a GenAI-powered chatbot could directly issue refunds, rebook flights, etc.). That technology is not here yet, and until it is, interfaces remain a constraint on automating end-to-end workflows that rely on legacy systems.
Engineering Constraints: Systematic. The less structured a particular process is, the more difficult it is to automate, because the lack of structure makes exception management more complex and harder to design.
It’s important not to mistake complexity for systematization. Complex processes involving natural language, such as real-world interactions with human customers, are effectively systematized by LLMs. In contrast, global supply chain operations remain difficult to grasp because they are subject to unpredictable shocks, such as violent conflict, regulatory changes, and dramatic climate events. This lack of predictability makes systematization difficult and will require human judgment for the time being.
Economic constraints: Uniqueness. Even if an activity is sufficiently stable and systematic from an engineering and design point of view, Repeatable To make automation economically feasible. This is typically the case for “one-off” activities with their own significant specifications, such as construction. Constructing a building always requires adapting a general blueprint to the particularities of a particular terrain. As such, automating such adaptive designs is often a more cumbersome task than a human could perform. For example, electronics manufacturer Foxconn found that using robots to manufacture many consumer electronics products was often ultimately unprofitable. With short production cycles and rapidly changing specifications, by the time they could automate the production of a particular item, the production cycle has moved on to a new product.
For business owners, these three types of constraints are particularly important to address because they tend to persist in some form even as technology advances. Therefore, human labor remains essential to every business, and agility strategies will remain important despite efforts to achieve full automation. The key for companies is to understand which workflows are actually adaptable. full Technology strategies need to be developed that promote automation and clearly distinguish between “scalable” and “automatable.” Even when automation is indeed feasible, technology adoption should not be designed “around” human workers if companies are to maximize the value of AI as a source of competitive advantage.
***
Business leaders would do well to be reminded that the economic value of technology is greatest when it enables the full automation of workflows. Those who understand this reality can evaluate any AI “use case” presented to them by asking a series of questions: Is the use case part of a broader process that lends itself to full automation? Are other components in the process codified enough that a machine could replicate it in the near future? Or are they too unique for an automated alternative? Are they entangled in various systems, each with their own logic? The answers to these questions can help guide leaders to the most effective use of precious technology dollars.
Read other Fortune columns by François Candelon:
FranSowa Candelon is a partner at private equity firm Seven2 and a former global director at the BCG Henderson Institute.
Henri Salha is a former partner and managing director at Boston Consulting Group and former senior vice president of operations at Essilor.
Namrata Rajagopal is a consultant at Boston Consulting Group and an ambassador for the BCG Henderson Institute.
David Zuluaga Martinez is a partner at the Boston Consulting Group and an ambassador for the BCG Henderson Institute.
Some of the companies mentioned in this column are past or present clients of the author’s employer.
