Accelerating Workforce Training: Introducing the CTA in E/Affect Initiative

An NDM Perspective on Recent Advances in Artificial Intelligence

Recent advances in artificial intelligence (AI) systems that demonstrate natural language understanding are very impressive. People are astonished by the progress and excited by the promise these emerging technologies offer, and rightly so. AI tools are already revolutionizing some workplaces and even everyday lives.

At the same time, there is concern about over-promising their potential, worry about ethical issues, and debate about the appropriateness of implementation.

NDM brings valuable contributions to this field. NDM seeks to inform the creation of tools to amplify and extend the human capacities to know, to perceive, and to collaborate.  NDM provides the empirical methods necessary to study expertise. For decades, the NDM community has studied how people make good decisions, tough decisions, high-stakes decisions—and can do so under pressure.

“NDM seeks to inform the creation of tools to amplify and extend the human capacities to know, to perceive, and to collaborate.”

As AI techniques become ever more powerful and ubiquitous, it is imperative that we continue to recognize the sizable gap between what experts do and what AI does – and how it does it.


Simply put:

Expert performance is more than data processing.


AI Systems

Engage in the complex and messy world.

Work only within a highly structured digital world.

Pursue goals even as they may change.

Pursues the completion of fixed tasks.

Have experience that includes tacit, local, historical, and organizational knowledge.

Use only knowledge that has been made explicit.

Can make mistakes, but their judgments are not “guesses”.

Attempt to fill in data gaps, but can do so by organizing data into patently untrue presentations.

Attempt to solve challenges that are ill-structured and wicked.

Succeed with problems that may be complex or difficult, but are “tame.”

Respond to ongoing problems while anticipating and mitigating emergent ones.

Require specific, well-defined prompts to initiate and guide responses.

Can work fast at a broad array of cognitive and perceptual tasks.

Are remarkably fast but only at a narrow range of tasks.

Are slower when they must deliberate, but deeply understand problems and typically reach decisions that are a workable fit for the situation.

Select responses based on optimization strategies without consideration for whether responses are context-appropriate.

Develop mental models of the world that inform causal reasoning and common sense.

Rely on probabilistic reasoning, irrespective of correctness.

Rely on experience to select and refine routines, enabling them to perform in reliable, consistent ways, and improvise creatively.

Generate novel responses to every situation, even when relying on historic performance.

Take social interactions into account, including with teammates and adversaries.

Execute by itself, even when connected to other computational systems.

Have selves that evolve through interaction with the world and imbue the world with meaning.

Process digital and logical information via computation.

The NDM community is eager to engage with these new technologies, and their designers, in order to develop new ways of working and anticipate their potential consequences, even as we continue to seek an even richer understanding of expert performance.