ദ്ദി(˃ ᵕ ˂ ദ്ദി) Bhooshan's Blog

Designing the Human-AI Relationship

Artificial Intelligence (AI) has permeated across major industries and has redefined the essence of technology.At a fundamental level, we are familiar with AI’s potential for correcting grammar or tone in your email or even drafting a response (e.g. Microsoft Copilot for Outlook), and summarizing your search results (e.g. Google Gemini and DuckDuckGo’s Search Assist), however, AI is shaking up industries and changing the way humans interact with data in performing their routine tasks. Yet a number of professionals and individuals are shunning the use of AI, and a few organizations are pretending as if they’d be insulated from the onslaught of AI by limiting the use of the technology. Many organizations continue to underestimate the pace and depth of AI-driven transformation.

It is becoming increasingly evident that a host of industries are undergoing a drastic change with the adoption of AI such as healthcare where AI has been utilized to analyze patient data and medical data to prescribe treatment options for cancer care (e.g. IBM Watson for Oncology). In the financial and banking industry, AI is constantly detecting fraud through real-time transactions and raising awareness and compliance levels (e.g. Mastercard Decision Intelligence). Who isn’t familiar with the courier delivery services, that sector itself is facing a churn with AI being used to optimize routes and provide delivery predictions to reduce fuel and delivery times (e.g. UPS ORION system). From these instances it’s proven that in the near future AI would not be limited in isolated places merely as a tool or an assistant, but it’d be operating as an invisible layer of our daily lives and residing within enterprise systems becoming imperceptible but important like electricity or the internet. Yet there’s a certain hesitation toward AI adoption for its intended purposes, and those who are convinced of its true nature have begun showing a lack of uncertainty and trust in human capabilities. Through this article I want to introduce key features of adopting a Human-AI relationship and addressing the issue of approaching AI from a design strategy viewpoint. To understand this transformation, we must examine the evolving relationship between humans, AI systems, and enterprise decision-making.

The Human-AI Trust Problem

In an episode on the TED Talks Daily podcast featuring deep tech entrepreneur D. Scott Phoenix, he mentions that AI today lives on the other side of the screen (PC or mobile), when you ask a question, it answers. But when you close your laptop it doesn’t go away. It learns from your query and keeps getting better at your job. If we stay separate from AI and continue to treat it as a tool without understanding the implications, it’ll replace us in the future, because it’s getting “smarter and faster and cheaper every week”.

It’s time we acknowledge the inevitable and move towards allowing AI into certain fields to make analysis and predictions more efficient, credible and faster, while restricting it in a few areas (such as schools and colleges) to improve and sustain fundamental human skills such as critical thinking, reading, etc. To achieve the goal of a veritable implementation of AI professionals and organizations would have to relinquish tasks and responsibilities to AI, such as reviewing and analyzing data while ensuring a gated approach.

Designing AI Adoption Strategically

Traditionally, organizations evaluated emerging technologies through theories, assumptions, and discovery-phase analysis before implementation. But what’s important is to involve the parties including the business stakeholders in understanding the operational context and the associated environment from an external perspective. Secondly, given the context and the organizational culture, how far does the organization feel comfortable in implementing the AI stack — does it look at the AI model to augment, assist, automate, or take independent decisions on the behalf of the individual? How far does it want the human to stay in control of the tasks and responsibilities? An essential question is the level of comfort the employees would have in transforming their work culture, are they willing to forego a certain proportion of freedom for the sake of organizational productivity. It’s equally important to address the concern of whether the subsequent increase in staff productivity would compel the organization to replace human resource with a structured AI model resulting in loss of jobs or consider another approach of distributing critical tasks around AI while allowing human oversight over the rest? These questions are important to foster trust and reliability with the AI deployment in building a long term Human-AI relationship.

The Four Dimensions of AI Integration

Let’s examine the context of an AI model through the prism of the four dimensions of implementation — augment, assist, automate, and autonomy. Does the organization or the individual align with one of these four dimensions, or are they looking to gradually increase dependency on AI in relation to a specific task or context?

  1. Augment — The individual remains in complete control of the task, except for a few elements delegated to the AI model for resolution. The aim is to enhance human capabilities.

  2. Assist — The human remains in control of the critical aspects of the task while seeking support from AI to improve the accuracy, speed, and efficiency of execution.

  3. Automate — For instance, an agentive AI system executes tasks independently based on a predefined algorithm while the individual focuses on other priorities.

  4. Autonomy — The AI model has acquired sufficient contextual knowledge to execute tasks independently by gathering inputs from multiple data points, with little to no reliance on human intervention. The decision-making process is handled autonomously by the AI model.

These dimensions are not sequential stages of AI maturity. Rather, they are strategic deployment models determined by organizational context, governance requirements, and the level of human oversight.

The research & analysis phases are indicative of the dimension(s) required to operate in the given context and environment to tackle the Human-AI trust problem. The questions we need to know the answers for before drawing a strategy are as follows:

  1. What’s the intended purpose of implementing the AI model, and what was the organizational approach earlier? (Understanding user journeys and why is implementing an AI model the only choice.)
  2. How far is the organization willing to relinquish control over its decision-making process? (Which of the 4 dimensions are more likely to bring impact)
  3. Is the organization looking at AI to — augment, assist, automate, or become autonomous, independent of any human intervention.
  4. How will the organization deal with loss of human control to AI in a given context/user journey? (Does it want an autonomous enterprise system)
  5. What’s the level of comfort with the staff to replace routine tasks with AI? (Which dimension would more likely bring acceptance & satisfaction)
  6. Crucially, would the implementation of AI models result in the discontinuation of specific roles from the organization? (Preparing the organization to brace for job losses, retooling, training & awareness)

After giving due consideration to the context and the environment, and having honed in on the AI model with one or all 4 dimensions, the next phase would be to visualize the structure of the AI implementation. For instance, consider having AI as a framework for automating tasks using AI agents would provide clarity on the nature of the tasks — the challenges in terms of complexity, replication, and time constraints that need to be overcome using automation. The conclusions from this phase would enable to visualize a journey map of the tasks pinpointing the exact instances to deploy the AI agent. The governance structure comprising of key decision authority would oversee the safe implementation and the security in case the agentive AI deviates from its assigned responsibility.

This phased framework of research → analysis → conceptualize provides clues at a basic level, but as more knowledge is gathered through the first two phases of research and analysis and by distilling the information you receive based on the context and the environment, you may find that more questions are emerging from the process.

Conclusion

In conclusion, instead of wavering with utilizing AI from any or all the 4 dimensions within the enterprise system, organizations can integrate AI to meet their desired objectives starting from a narrow to a broader prospect with a clear model for human oversight. Instead of looking at artificial intelligence as a tool that replaces human efforts to boost output, look at it from the lens of human-centered design in uplifting the capacity of the workforce. In situations where AI has overtaken individual tasks in capabilities look at reorganization rather than replacement. The plan should be to continue filtering the tasks through the lens of the 4 dimensions corresponding with the context and the environment until a saturation point is reached where human-control is neither compromised or feels discouraged in performing their duties.