Human-in-the-Loop (HITL) has become a necessary and practical method to achieve success in all mission-critical applications of artificial intelligence. Whether through liability concerns, regulatory pressure, or good old-fashioned error checking, engineers have arrived at this conclusion through bitter experience with systems that produce mistakes.
Underneath the important ethical and practical considerations for an HITL or Augmented Productivity approach lies an essential fact that cannot be dismissed.
HITL is not a temporary scaffold to be removed once AI reaches a certain maturity level. Increasingly, AI proponents have steered away from this vision for ethical reasons. It also distracts from the real benefits of AI. The importance of HITL lies in the practical acknowledgment of something that the German philosopher Hans-Georg Gadamer understood long before the first neural networks emerged.
Gadamer’s fundamental idea is that interpretation requires a “fusion of horizons,” something AI alone can never replicate. A “horizon” represents what can be understood from a particular vantage point – similar to a “point of view” – it includes historical, cultural, and linguistic factors.
Interpretation requires bringing a text’s horizon into interaction with the reader’s horizon.
If you think of interpretation as a circular function where the reader is projecting meaning onto an input (such as the results of an AI prompt), you start to see a cycle of interpretation and value creation, where the worker is ‘thrown forward’ by the AI input and ‘projects forward’ the anticipated meaning, value, and conclusions to be drawn from the text. Human interpretation adds critical context, nuance, and ethical judgment that AI alone cannot provide, making the feedback loop more meaningful and applicable to real-world situations. This feedback loop is also the essential innovation of the chat-based UI that we see with AI today.
However, the circle of interpretation requires at least 2 points of view, and 1 of those points of view must be grounded in a reality that is appropriate to the situation and applicable to it for it to be accurate and meaningful.
Put simply, a large language model can provide millions of examples of grief, and yet has never felt grief. Choosing the right words for the right situation requires deep knowledge gained only through lived experience, resulting in a point of view. Human interactions require empathy, and empathy must be derived from the myriad of inputs that only lived experience can generate.
My clients are in the professional services space, and I am witnessing a fascinating realization of Gadamer’s hermeneutic philosophy in consulting, in particular. Without a full understanding of a client’s goals, motivations, creative biases, and industry knowledge, AI’s conclusions will never fully succeed, emphasizing the importance of human understanding and engagement.
By employing AI to remove rote, inefficient tasks over time, the feedback loop between the consultant and the AI Insights engine will sharpen, resulting in faster conclusions with highly helpful, innovative input. However, the consultant-in-the-loop will always remain an essential aspect of delivering expertise to a client’s situation. The consultant will be more efficient, and perhaps even “smarter” because of AI, but the consultant remains the essential and central player in the process. There can be no real value delivered to a client without HITL, in the form of a consultant’s or creative’s point of view.
AI without a strong “you” is of no real value to anyone.
