How Humans and AI Can Work Together

Companies pushing to automate and introduce AI into the workplace should pay attention to the challenge of enabling people and technology to work together. AI’s ability to augment human capabilities is critical to its acceptance and realizing its potential. However, to unlock this potential, companies need to figure out how to employ algorithms and people alongside each other.?

Options for co-working

Frameworks for developing human/machine working arrangements should include the following models.

Data collection and analysis. In these systems, computers and technologies such as large language models collect numbers, text, pictures, videos, and signals from sensors. The data is analyzed and sent to decision-makers for action. For example, chipmaker Intel uses AI to help it procure materials, labor, equipment, and services from some 16,000 suppliers in 200 countries. The system crunches mammoth amounts of data, such as performance reports and news items, to rank suppliers along multiple criteria. This information is presented to procurement managers to help them make sourcing decisions. The AI system also flags pending issues like merger discussions that might impact the company’s supply lines.

Monitoring. Here, automation technology executes most of the work, and humans intervene only when the machine fails or cannot recognize that the context has changed. A looming recession, a data hack, or the onset of a pandemic are examples of such a context shift. Adapting quickly to a new environment requires the intervention of human experts. An example is automatic ordering systems based on demand forecasts. When a structural environmental change occurs, existing demand forecasts are irrelevant, and humans must intervene to retune the forecasts or place orders manually.

Escalation. Escalatory systems perform relatively simple tasks and automatically – or on request – hand over control to a person when the need arises. Consider, for example, a service center chatbot that can solve routine and standard queries. When a consumer poses a unique or specialized question that the bot is not programmed to address, it will “kick up” the interaction to a human operator.

In the loop. Control is handed back and forth between the system and the people in these systems until the job is complete. As an example, consider the workflow at an Amazon warehouse. Aisles of items are moved on “platform robots” to workstations where workers pick from the aisles, creating packages for deliveries. As packages may include multiple items, the robots keep moving, handing controls to pickers, then moving on while other robots arrive with different items. Thus, the fulfillment includes a workflow requiring both humans and machines.

Augmentation. Many tech systems support humans in the workplace. Wearable exoskeletons, also known as exosuits, that combine people and machines, are one of the best-known examples. A person wears a robotic framework to help support or amplify the individual’s strength and movements.

Other considerations

In systems where people only intervene occasionally, thought must also be given to how humans can remain engaged. This is not a new challenge. For instance, passenger aircraft are essentially drones in that most of the time, automated systems fly the planes. One way to help human pilots stay alert during periods of inactivity is to give them responsibility for communicating with ground control.

Also, as powerful as the new generation of deep learning AI systems is, their wide adoption will require more transparency into how they work. This is particularly the case with systems that provide prediction and analysis and initiate an action, like the ordering system mentioned above. Many AI systems are inscrutable black boxes; they provide answers without explaining how they arrived at those answers. Such explanations are needed to convince human stakeholders that the answers are correct. Moreover, humans need to be able to cross-check and validate an algorithm’s output and provide a rationale for AI-inspired decision-making that people can learn from. One possible solution is a new class of machine-learning systems under development, known as Explainable AI, that provides more transparency into the workings of algorithms.

New job descriptions needed

A future challenge that will become increasingly visible as AI workplace applications grow in both number and scope is designing job descriptions that accommodate the roles of algorithms. Future job specs must account for the advantages of machines (e.g., accuracy, ability to perform work continuously) in combination with the benefits of humans (e.g., recognizing context, the ability to spot structural changes, flexibility). Winning companies will be those that develop job specs and algorithms that integrate humans and machines successfully.

?

My latest book “The Magic Conveyor Belt: Supply Chains, A.I., and the Future of Work,” explores the role of AI in the workplace.

?