With the development of artificial intelligence, it has gradually been applied to various fields, but when it comes to artificial intelligence solutions, designers may face new challenges. This article explores this and proposes a six-step approach that integrates the strengths of both disciplines, namely "AI design".
Write on the front
With the development of AI technology, it is becoming increasingly clear that neither technology nor design alone is enough to build effective AI solutions, solve real user problems, and have a positive impact on society. AI engineers working without designers may jump too quickly to solutions based on unproven assumptions, leading teams to solve "wrong problems." Conversely, a lack of technical knowledge can lead designers to develop unrealistic or vague ideas about the capabilities of AI, with situations where they either overestimate or underestimate its capabilities.
When it comes to AI solutions, designers may face new challenges, such as how to design AI systems for transparency, "explainability," or trustworthiness. Or how to evaluate the impact and outcomes of AI solutions on users and society. That's why we believe that artificial intelligence (AI) engineers and service designers should collaborate to create human-centered, ethical, and positively impactful solutions. In this paper, we propose a six-step approach that integrates the strengths of both disciplines, namely "AI design".
Do we need artificial intelligence (AI)?
We see AI design as a human-centered, iterative, and collaborative approach to AI innovation. It utilizes the designer's approach to problem-solving while considering the dynamic components of AI, applying AI's innovative engineering approach. The methodology is based on the Double Diamond1, CRISP-DM2 data management approach proposed by the UK Design Council, and our experience in AI innovation at OLX Group, one of the world's fastest-growing networks of markets and classification platforms, with over 300 million online visitors.
Both AI and design teams can benefit from collaboration. We believe this will lead to more efficient ways of working and more customer-centric solutions. Here's why:
1. AI innovation takes a human-centered approach
AI-by-Design takes a human-centric approach to AI innovation to gain a deep understanding of customer needs before developing solutions. Sometimes AI isn't the right solution. Sometimes, simpler solutions, such as spreadsheets, may do well and save resources. Therefore, before development begins, it is necessary to determine whether artificial intelligence is indeed the right tool to solve the problem. Teams should build anything that solves a problem and empowers humans, rather than prescribe specific tools or technologies.
2. Take advantage of interdisciplinarity
AI-by-Design encourages working across silos, rather than coldly handing over insights from design research teams to AI engineers. When designers and AI engineers collaborate, they can fill each other's blind spots and reduce the room for communication errors. The process will be more efficient and effective, and the solution will be more customer-centric and technically feasible. This will ensure they tackle the right challenges, saving the team time. In fact, in McKinsey's 2021 State of Artificial Intelligence, "using design thinking when developing AI tools" is considered the most important differentiator for AI high performers.
3.AI-by-Design solutions are designed for our dynamic world
AI models are often trained in sandbox environments, but end up being used in our chaotic, complex world. Therefore, we believe that AI models need continuous retraining.
In the real world, solutions are influenced by how users interact with the final product, while AI has many dynamic components. Designing a way to collect user feedback and actual behavioral data is critical. This input data is needed to improve the model and ensure that the AI solution works as expected and in an ethical manner.
There's a gap, and while that sounds ideal, we can't just put a few AI engineers and designers on a team and expect them to innovate together effortlessly. We often observe that the two disciplines do not have a common language, hold false assumptions about each other, and work in different ways.
For example, AI engineers work using Visual Studio code, while designers often use tools like Miro. Machine learning models need to work as meticulously and accurately as possible, and designing prototypes can be very conceptual and speculative. Machine learning metrics evolve around numbers and predictions, while design metrics evolve around human needs and customer experiences. The same differences exist when comparing methods, as shown in Figure 2. In the figure, the design committee's double diamond overlaps the CRISP-DM data management methodology. Three gaps emerge:
Gap 1: Missing "why"
If AI engineers are left out of the initial stages of a project, the solution they build risks deviating from the initial customer problem, and understanding customer needs is key.
Gap 2: Lack of technical understanding
On the other hand, designers often have unrealistic expectations about the possibilities of AI and are not always aware of the latest technological developments. Since engineers cannot easily understand engineering components, they often need help to grasp the feasibility of the proposed solution to avoid leaning towards difficult to implement.
Gap 3: Lack of feedback loops
At the end of the process, when building a solution, there needs to be a way to check that the solution works as expected, that the correct data is collected, and that the model is ethical. This can be solved by a feedback loop. Feedback loops can provide a lot of visibility and transparency into your solution. This is very important because, in most cases, the real world is very different from the training environment in which AI is developed. In addition, new data will be available due to the continuous interaction between users and the developed AI solutions. When the right data is collected, the model can be continuously improved by eliminating biases and outliers.
The six steps of AI-by Design
To fill in the gaps and find a way to work, we created a six-step methodology.
1. Discovery: The first step is to understand the goals of the project, customer needs as well as their problems, and business opportunities. It usually involves customer research.
2. Definition: In the second step, the team defines the scope of the challenge: they choose the problem they choose to solve or the opportunity to pursue. This step includes research background and possibilities of artificial intelligence.
3. AI design decisions: At this stage, encourage the team to ask themselves if this is a problem that can and should be solved. It can be solved with AI. If so, they assess what data is needed and study whether the solution may have unethical consequences. This is also a good result if AI is not the right solution. AI is expensive and time-consuming. If other alternatives solve the problem, you should choose them.
4. Development: The fourth step is to understand how best to solve the problem. Now is the time to explore different solutions and study the required data and modeling. AI engineers can do exploratory data scoring (EDA), which means digging deep into the data to better understand the data to see if there are outliers, missing values, and whether a baseline model can be built.
5. Testing: Before committing to building and deploying a solution, the team should identify what risk assumptions are in place and try to validate them, such as using prototype testing. This is the fastest way to check if the solution should be built or if a "pivot" is required.
6. Delivery and evaluation: Finally, the team should iterate, refine, pitch, and ultimately deliver the solution to the end customer and key stakeholders. However, the process does not end there. Instead, teams need to continually iterate on solutions. Over time, more data will emerge that can bring new insights into the solution. A monitoring system (feedback loop) needs to be designed to ensure that real-life deviations and data drift are checked and corrected as quickly as possible.
The first six steps are designed to provide guidance for those looking to improve the way their teams implement AI innovations.
conclusion
Companies need technology and design to build AI that: (A) is effective, (B) solves the problems of real users, and (C) has a positive business impact. There is still a long way to go, but we believe that our proposed approach will enable organizations to innovate in an efficient and ethical manner.
Serena Westra, Ioannis Zempekakis (the article has been authorized by the parties and the original author)
原文名称：AI-by-Design: Human-Centred AI Innovation: A six-step approach for building AI solutions
Translator: Chen YuzhiYeutz Chen, WeChat public account: YeutzDesign (ID: Yeutzsheji), focusing on the field of service design, committed to the research of service design innovation and transformation.
This article was translated by @陈昱志Yeutz Chen and published on Everyone is a Product Manager. Reproduction without permission is prohibited.
The title image is from Unsplash and is based on the CC0 protocol.
The views of this article only represent the author himself, everyone is a product manager, the platform only provides information storage space services.