Why Human Centered Design is the Perfect Fit for AI Development.
?? Nicholas Marchand
Action + Outcomes > Words ????| Strategy Architect | Design Thinking | Aspiring R&D Data Labs
Why Human Centered Design is the Perfect Fit for AI Development.
By definition, Artificial Intelligence (AI) refers to the ability of machines to perform tasks that normally require human intelligence – for example, recognizing patterns, learning from experience, drawing conclusions, making predictions, or taking action – whether digitally or as the smart software behind autonomous physical systems. – Department of Defense AI Strategy, 2018
The race to prepare, design, develop, operationalize, and deploy Ai is already underway; infrastructure is being modernized to support faster movement of data, policies are being written, AI enablement organizations are being stood up, and AI ethics are being debated.
Each of those aspects are vital to deploying AI, but I would argue that we are missing a key ingredient: emphasizing a Human Centered Design (HCD) approach to every new AI design and development activity. Most AI projects don’t fail because of the technology, they fail because of their lack of understanding of the context of the data they are working with. Context is King.
IDEO defines HCD as a creative approach to problem-solving that starts with people and ends with innovative solutions that are tailor-made to suit those people’s needs. HCD empowers an individual or team to design products, services, systems, and experiences that address the core needs of those who experience a problem. Going a step further in the HCD process, the creators of products immerse themselves in the lives and perspectives of the users involved. Through methods of observing the human experience, understanding the challenges and envisioning the future possibilities, teams are able gain a holistic frame of reference to approach solutions.
If we truly want to deploy AI apps that are aligned to user needs and are able to be trusted by those users, a human centered approach must be part of the process.
What makes HCD different from design concepts currently in use today like MBSE or traditional SDLC methods? The difference resides in the layers of empathy from a human perspective that uncover insights not captured in other methods. Feelings, emotions, behaviors, and motivations are just a snapshot of the data and feedback derived from creative HCD sessions. It is these elements that we must uncover and understand to truly build successful AI.
“A commander on a future battlefield that can make a better decision, faster than his opponent based on the amount of information available - will have a significant advantage” Gen. John Murray, Commanding General of AFC.
Using HCD, design teams can help the data team break down a Commander’s objectives and intent to scope out the problem space, refine the requirement priorities, identify unintended bias, and ensure what we are building or solving for really is the right solution.
Through different ethnographic research and analysis techniques we can understand behaviors, motivations, and key decision drivers that may go unnoticed which can help inform the data teams as they build algorithms and iterate on their models.
Let’s take for example a scenario where a command has tasked the data team to develop an algorithm to identity certain structures. The context that is needed to develop and iterate on the algorithm can be pulled straight from HCD activities. Think about how that particular algorithm is constructed. The scientists need to understand elements such as:
·??????what data is involved and how is it involved
·??????who looks at the data and why
·??????what decisions do users make with this data and why
·??????what courses of actions are effected by the intended results and why
·??????how do they make the decisions, or
领英推荐
·??????or why do they make those decisions.
·??????what type of methods they should use (Deep Neural Networks, pattern of life, etc.),
All of these answers can be generated from HCD activities combined with how users feel about certain decisions or what motivates them to look at certain locations, for example, can improve an algorithms accuracy, explainability, and can increase the consumer confidence of the AI product. ?
The ultimate objective isn’t to just deploy a working model, it is to deploy a machine model that has generated a high enough confidence level through its development to gain a Commander’s trust to execute based on its results. A trust that is only built through understanding the human element.
References:
?
?
?