Teamit employees have had the opportunity to participate in the journeys of both large and small organizations, implementing artificial intelligence as an essential component of the work environment. We have condensed our observations into a workshop series, which we are now presenting in our blog.
Our methodology combines strong human-centricity, the research phase of service design, experimentation with and utilization of AI possibilities, and future-oriented thinking. The outcome consists not merely of isolated experiments but of both major and minor service innovations based on the genuine needs of people within the organization.
Our seven-part series introduces a human-centered process that enables organizations to transition from experimentation to profound utilization of artificial intelligence. In the coming weeks, we will examine the essential phases we have identified and how to derive maximum benefits for the future by employing strategic service design methodologies. We will also address questions related to AI ethics and security, and explore the application of future-oriented thinking.
We will not delve extensively into individual technologies or applications, as suitable options for each situation will emerge during the process. Our process provides answers to the questions that influence future technology choices. These writings illuminate the intellectual framework of our approach, and each article concludes with essential and concrete actions.
We aim to maintain human control and strive to understand, as much as possible, why a particular solution functions as it does.
The ultimate outcome of the process is validated concepts that can confidently serve as the foundation for UX design work. Mapping human motivations, perceptions, and thought processes will be extremely important in the future as system user experience becomes increasingly personalized.
Experimentation!
Since you are reading this article, you may have already used artificial intelligence: conversations with ChatGPT for work and leisure, entertaining images, music, text summarization, and language translations…
It is wonderful that experimentation is now so accessible. With our working method, lessons learned from experiments can be systematically incorporated to increase the overall knowledge capital. Presumably, most ideas will represent small continuous improvements, but when applied broadly enough, one can occasionally expect something radically new.
Collecting all kinds of ideas is really worth it.
Artificial intelligence has been emerging for a long time, and science fiction-like expectations for technological capabilities can be found across at least five generations. The latest wave originated with the release of ChatGPT, which brought large language models within everyone’s reach two years ago.
As of autumn 2024, the biggest Gen AI bubbles have likely already burst. Our approach maintains realism, beginning with the premise that AI can beautifully complement but rarely completely replace human work. We aim to maintain human control and strive to understand, as much as possible, why a particular solution functions as it does.
Start with People in the Present Moment
The entire process begins with people because we believe that technology exists for humans. There is immense potential for learning from the working methods and workflows of individual experts.
At the start of the process, it is important to gather numerous and diverse ideas, including those that may not immediately seem important or feasible.
When preparing for the process, it is advisable to identify boring and repetitive tasks in your own and organizational workflows that would be liberating to delegate to machines. Opportunities also arise in comprehending entities too large for human perception, identifying recurring patterns or groups—significant benefits can be achieved through an augmented intelligence approach.
This can lead to a transformation in which the machine helps the human worker achieve more, rather than humans merely verifying machine outputs.
Many Solutions That Do Not Work
When discussing AI implementation and examining workflows it is intended to complement, situations often arise where a more traditional system solution would be preferable. Typically, it is undesirable for a company’s customer service bot to respond randomly or for an image generator to create offensive images.
Identifying these use cases is excellent and even desirable, as the intention is not to replace everything with AI, but to find the most suitable applications.
There remains an abundance of interesting and useful targets. AI can be leveraged in numerous application areas based on images, audio, text, as well as market and measurement data. Insightful opportunities can be found in marketing, automation, optimization, process control, testing, monitoring, maintenance…
It is crucial to consider and understand what constitutes a sufficiently good outcome expected from artificial intelligence.
When working with language models, it is advantageous to proceed in small steps. It is not always entirely clear why a model provides a particular answer. Research by Sourav Banerjee, Ayushi Agarwal, and Saloni Singla examines generative AI hallucination as a feature rather than a flaw. It may never become completely reliable in the sense that a specific workflow would always produce exactly the same outcome. Progressing with smaller components helps to understand what happens and perhaps why.
It is therefore important to consider and understand what constitutes a sufficiently good outcome expected from artificial intelligence in your own use cases. Do we have different criteria for AI than for humans? And are the prerequisites for technical success present? Available data is of great significance. Human skills, such as aesthetic judgment, must still be employed in evaluating the outcomes of generative AI.
Could We…
It is important to begin by outlining the possibilities. A diverse working group is advantageous when collecting ideas. Everyone can consider how they would like to improve their work performance.
At this stage, it is important to emphasize positive thinking. Could we? Not “should we” or “would it be worthwhile,” but “could we.”
If you have attended work-related courses in recent decades, you have undoubtedly encountered the overly positive “mistakes are lessons” type of mantra. However, this cliché should not be avoided like the plague, as it contains genuine opportunities.
In situations of intense time pressure, perspectives may begin to narrow, making it difficult to perceive broadly when focusing on solving the immediate problem. Fear of failure creates anxiety and suppresses creative thinking—precisely what is needed more, the more challenging the situation becomes.
When gathering to outline the first steps of AI implementation, it is worthwhile to actively think together, “could we?” When considering “could we,” the fear of failure disappears and liberates thinking. The possible outcomes are either proving the idea worthwhile or recognizing that the idea was not in the right time or place on this occasion.
Culture of Experimental Co-Development
Everyone engages in ideation within their minds. Regular gatherings to share insights, however, are already less common.
Discussing findings is an important part of the entire process, and bringing together different experts is extremely fruitful. Facilitating discussions is necessary to ensure that voices beyond just the extroverts are heard. People’s ways of reflecting and sharing their reflections vary greatly.
Our methodology always includes three essential factors: people, their operational environment, and available data. Regular examination and listening to this trinity provides the development project with the information it needs.
Utilize artificial intelligence: brainstorming ideas with a language model (e.g., ChatGPT or Copilot) often provides new perspectives. Request ideas be saved in CSV format and analyze them later.
What is Needed?
- A confident atmosphere and a place for experiments: experimental results should not accidentally leak into the public domain or confidential information end up as training material for language models. When there is time and enthusiasm, experimentation must be possible.
- Continuous discussion about experiments: both advanced and beginners are excellent at advising each other. Often, pondering a problem can be more important than the actual solution.
- Continuity, so that conversation becomes natural.
- Documentation platform: notes on the wall or digital notes.
What Can You Do Yourself?
- Reflect on your own activities: which work phases would be meaningful to delegate to machines?
- Experiment with what ChatGPT can accomplish related to your ideas.
- Organize a gathering, compare thoughts with colleagues, and document them.
In the next article, we will discuss the processing and validation of ideas and practical arrangements for experiments.
Let’s Meet!
We would be happy to explain how we have initiated fruitful processes. One hour is sufficient. Let’s connect!
Aleksi Manninen
aleksi.manninen@teamit.fi
