Generative AI strategy, user experience, and consulting. San Francisco, California

Organizational Design

It's been noted, somewhat ironically, that AI threatens not blue-collar work so much as white-collar, knowledge work. Generative AI threatens jobs among knowledge workers, where expertise, obtained often at great expense and from years of dedicated educational effort, has long been richly remunerated. Large Language Models are capable of broad and deep knowledge and expertise, but unreliably so. This unreliability is being chipped away as model training and design improves, and as data used in model training, reinforcement learning, and other modifications allows generative models to perform with greater accuracy and relevance.

Whilst machine learning developers and researchers discover and innovate on technical and architectural solutions, designers should have a role to play in model design also. Integration of generative AI into large organizations requires management and organizational design if it is to proceed successfully and with minimal disruption. Whose roles, which workflows, tasks, and jobs, in which cases, under what conditions, need all be determined before AI can be effectively deployed. All jobs matter but not all jobs are equal. AI is capable of intelligence and increasingly of semi-autonomous actions as well, but not with evenly distributed gains and risks.

I see a role for researchers, some of whom may have been customer-facing UX researchers, in conducting internal organizational readiness and discovery for the purposes of AI integration. UX researchers are familiar with the stakeholder interviews, workshops, research and organizational design briefs, service design blueprints, and other internal communication and presentation vehicles used to socialize technology adoption internally. These same steps will be needed for implementation of generative AI across workforces.

Design can contribute to definitions of roles and responsibilities, jobs and tasks, workflows, dependencies, risks, reporting, and so much more. Working with developers, user experience designers can shape and define benchmarking needs and goals. What kinds of use cases will match the capabilities (and costs) of generative AI integration? What company data is ready for use, and where are workflows often bottlenecks in work processes? What roles should AI supplement, if it is deployed as an assistant? What conversations can it contribute to, if added to Slack channels and communication? What question-answer jobs can it handle without hallucinating badly; and which departments have the conversational transcripts and exchanges on which to fine-tune or specialize a model for domain-specific use?

Workforce integration of AI will present massive opportunities and risks as AI stretches the very definition of knowledge-based practice. Knowledge is not just information, but is the judgment required to act on observations framed by that information. AI can have information but not the right kind; it can have information but not the context; and it can certainly have the information but not the judgment. These are all limitations for which there will likely be incremental solutions, however. The sheer immediacy and speed of generative AI deployed in many organizational use cases will be too much to resist. Implementation of AI for knowledge work will present a wide range of opportunities for organization transformation.