Designing for AI

Intro

In this chapter we are going to focus on the user and the current state of AI development. User context is decisive to design for AI. How do users feel about the technology and what implications they assume from its usage? We will discuss the necessity of widening the traditional user research to include AI interactions, AI perception and AI impact on products. We will suggest new methods to test the human interaction with AI systems and questionnaires to deduct attitudes towards AI. We will signal what main reactions need to be avoided for them to not affect the product or service value perception. We will talk about the implications of using AI: How this affects people and the world we are living in.

01. Extended personas

Goal

Build and visualize personas for individuals who will interact with AI by researching their attitudes towards it, including trust, acceptance, and willingness to use AI.

Why

There is a gap when it comes to typical persona creation methods, as AI was not part of the initial User Centered Design methodology (UCD) when it was created. These comprehensive personas will help to account for not only the early adopters but also the full spectrum of attitudes towards an acceptance of an AI product, ultimately leading to a larger user base.

How

•• Identification of user groups

Focus on who will use the AI, including those who need to understand and interpret AI outcomes. These last may not be the ones directly interacting with it.

•• In-depth user data collection

When building your personas make sure you add researched info about how users interact with AI:

1. Attitude about AI, specifically trust, acceptance and willingness.

2. If not trusting AI, under which conditions users would trust it, if at all.

3. In your surveys, try to detect “edge users” to manage biases.

E.g. in a 5 users research you detect 3 of them are AI enthusiasts. This will bias your results. Ask them to compare their attitude with “typical users”. It could turn out that the AI enthusiasts are not representative for the majority of the users.

•• Data consolidation, analysis and persona visualization

Make sure that differences in the users’ attitudes about AI end up building different personas. Provide a specific section for this information in the final layout of your persona profile.

This chapter is mainly based on: A. Holzinger, M. Kargl, B. Kipperer, P. Regitnig, M. Plass and H. Müller, “Personas for Artificial Intelligence (AI) an Open Source Toolbox,” in IEEE Access, vol. 10, pp. 23732–23747, 2022, doi: 10.1109/ACCESS.2022.3154776.

02. Attitudes towards AI

Goal

Measure positive and negative emotions regarding AI.

Why

When designing for AI we need to measure the attitude of our users towards this technology and how much educated our users are. Typically, less educated users will show lower AI acceptance. Reluctancy towards AI needs to be neutralized designing proper interactions that inform the user and soften friction due to mistrust or suspicion. You need to design your AI products and services (p/s) taking both into account, in order to manage pre-conceptions, expectations and interpretation.

How

Here is one suggested survey for measuring AI positivity or negativity.

Rate from 1 (strongly disagree) to 7 (strongly agree)

• “Artificial Intelligence can provide new economic opportunities for this country.”

• “There are many beneficial applications of Artificial Intelligence.”

• “Much of society will benefit from a future full of Artificial Intelligence.”

• “Artificial intelligence can have positive impacts on people’s wellbeing.”

• “I find Artificial Intelligence sinister.”

• “I shiver with discomfort when I think about future uses of Artificial Intelligence.”

• “Artificial Intelligence might take control of people.”

• “I think Artificial intelligence is dangerous.”

Survey by J. Bergdahl, R. Latikka, M. Celuch, I. Savolainen, E. S. Mantere, N. Savela and A. Oksanen, “Self-determination and attitudes toward artificial intelligence: Cross-national and longitudinal perspectives”, Telematics and Informatics, Volume 82, 2023, 102013, ISSN 0736–5853, https://doi.org/10.1016/j.tele.2023.102013.

03. Improve user’s perception of AI

Goal

Minimize frustration in human interactions with user interfaces that incorporate AI systems so that they are perceived as valuable.

Why

With the new AI systems, the user no longer tells the computer what to do. Rather, the user tells the computer what outcome they want (2). The further the AI responses deviate from the desired outcome, the worst the p/s perception will be.

How

How 01: TRACK USER’S HABITS AND INTERACTIONS

• Consider the use of the ‘Wizard of Oz’ method in user research. It simulates AI functionalities using human operators behind the scenes to gather authentic user feedback without the initial investment in full AI development.

• Longitudinal studies and diary studies are valuable for understanding how user interactions with AI evolve over time, capturing shifts in behavior, emotional engagement, and user expectations.

• Continuous user feedback is crucial for refining AI models. User feedback, embedded in the product (e.g. thumbs-up / thumbs-down ratings), can directly train and refine machine learning algorithms, with each interaction serving as a new data point that enhances the AI’s accuracy and responsiveness to user preferences.

• Use AI tools to enhance user research methods, particularly in data analysis and behavior prediction.

How 02: ASPECTS THAT CONTRIBUTE TO THE USER’S PERCEPTION OF AI POWERED USER INTERFACES

Collaboration between UX teams, product specialists, tech experts, and other stakeholders is key to integrate user research findings in the product development cycle effectively and to discuss the following:

•• Inclusiveness, privacy and safety We know that LLMs have been trained with biased data. The internet is crowded with information from western countries and perspectives. It is easy that AI outcomes feel non-inclusive for minorities. Inclusion includes several important aspects from ethics to genre. AI systems need to be built with sufficient safeguards in sensitive areas.

Introduce consent forms and data management policies for responsible data collection.

Try user tests to ensure continuous ethical assessment during product development. For example, with the so called “Moral Machine experiment”, the moral decision-making in autonomous vehicles was investigated by collecting global human perspectives.

•• Effectiveness (3)

E.g. Healthcare. Poorly designed systems can misdiagnose. AI designed to heal in the context of the profitability of a business, might increase — rather than cut — costs, and programs that learn as they go can produce a raft of unintended consequences once they start interacting with unpredictable humans.

•• Sense of control

Define the scope of automation and ensure a balance between machine operation and user control. Keep the user informed about this balance.

This chapter is mainly based on: C. Jin, “How to build better AI products with user research”, Feb 9, 2024, URL: https://uxdesign.cc/how-to-build-better-ai-products-with-user-research-8c5a863bcfc3, read on Aug 28, 2024

(2) J. Nielsen, “AI: First New UI Paradigm in 60 Years”, Jun 18, 2023, URL: https://www.nngroup.com/articles/ai-paradigm/#:~:text=With%20this%20new%20UI%20paradigm,reverses%20the%20locus%20of%20control, read on Aug 28, 2024

(3) A. Powell, “AI revolution in medicine”, Nov 11, 2020, URL: https://news.harvard.edu/gazette/story/2020/11/risks-and-benefits-of-an-ai-revolution-in-medicine/, read on Aug 28, 2024

04. Socio-Technical Perspective of AI

Goal

Designers play a role in deciding if AI is really needed by analyzing the solution’s influence on people, economy and environment in advance. In the end, the benefits for humans and businesses must overweigh all the negative impacts of an AI solution. If that is not the case, you should consider not using AI at all.

Why

Costs and risks implied in the use of AI justify a thoughtful attitude towards it.

How

• The use of all AI systems may lead to diverse forms of harm, especially for people in vulnerable situations. As designers, it is imperative that we adopt a socio-technical perspective toward designing responsibly: we should question how will it improve the user’s experience, provide them with new capabilities, or address their pain points. (1)

• Define who (from user groups to whole society) or what (economy or environment) is directly or indirectly influenced by the AI

• The rise of artificial intelligence is now turbocharging

  Demand for bigger data centers

  (more land), transforming the landscape even more and taxing the region’s energy grids. The almost overnight surge in

  Electricity demand from data centers

  is now outstripping the available power supply in many parts of the world. Increase in power demands from Silicon Valley’s growth-at-all-costs approach to AI also

  Threatens to upend the energy transition plans of entire nations

  and the clean energy goals of trillion-dollar tech companies. Designers need to consider this context when advising in favor or against AI solutions. (2)

(1) J. He, M. Muller, G. Hoefer, R. Miles, and W. Geyer, “Design Principles for Generative AI Applications”, Feb 20, 2024, URL: https://medium.com/design-ibm/design-principles-for-generative-ai-applications-791d00529d6f, read on Aug 28, 2024

(2) The Big Take, “AI is already wreaking havoc on global power systems”, Jun 21, 2024, URL: https://www.bloomberg.com/graphics/2024-ai-data-centers-power-grids/?leadSource=uverify%20wall), read on Aug 28, 2024