Welcome to the ninth edition of the Ki-Navigator column of the Doag Ki community!
Advertisement
Dr. Konstantin Hopf Business Information Sciences, especially at Bamberg University, heads the data analytics research group in the chairmanship of especially energy efficient systems. In industrial research projects, he develops operating applications of machine learning processes. He also researches concepts for strategic management of AI initiative and data science teams. Their research results appear in major magazines and conference versions of business information science, but also in experts and daily presses.
If you buy in a chemu, then you have to know these 5 scams, in which the alarms are engraved among the customers.AI promises a huge increase in efficiency and automation through customized production processes, smart control, automatic texts and images or individual voice assistants, which are with us at every turn.
However, anyone who deal with the implementation of AI initiative in companies will fulfill different, sometimes the opposite approaches. There is already a view of appreciation of colleagues from users, management and successful show cases and various AI devices. It seems easy to apply a lot: AI can prepare email, can move conversations and generate attractive images.
However, the perception that you can convert your company into AI, as a recently suggested commercial, proves to be a decline. In practice, making AI models and applications is usually difficult handicrafts. These mostly handle data scientists, data engineers or machine learning experts with many years of training and interdisciplinary backgrounds. In cases rarely, you work with installed procedural models and standard software. Rather, applications are driven by exploration. Creating AI app is more like manual activity than mechanical processing. I have an international team of researchers Investigated 55 AI initiatives in companies around the worldIn the study, we identified and found challenges in behavior, causing them damage. These findings allow the recommendations to be obtained how companies can deal with these challenges.
Five major challenges for AI initiative in behavior
Excessive expectations about the development of AI applications are one of the five central challenges that we recognized in the study. Another problem is that many companies treat their AI projects such as classic IT projects. You need more recurrence and discovery approach and sometimes have more agile than software development. The very rigorous project management approaches or net following of the guidelines obstructs innovation and leads to disable use of resources.
While agile work methods help to adapt software to rapidly replace customers’ requirements and users to integrate quick integration, these strength of AI projects methods can be a clear cause of problems: training and testing Good AI models are usually difficult to plan.
A third challenge is the integration of AI in existing systems. The missing data access not only questions the AI ​​projects, but also waste resources. Data scientists are often waiting for data access for weeks because data security, technical or organizational obstacles are standing in the way. In addition, if the results of intelligent systems cannot be traced back into heritage systems, AI projects are facing end.
It is also difficult to explain the answers of the AI ​​model in a sensible way, especially in important or legally heavy regulated areas. AI models with good predictions are usually opaque. This can reduce the trust of users because they do not understand how predictions come and what are the causes of errors. This re -reveals the craftsmanship of AI development: many data scientists reported that they have led to a prediction to identify and explain the factors affecting them.
Finally, the dynamic nature of business and data landscape requires continuous care and adaptation of the AI ​​model to ensure their performance over the long term. Since data changes occur continuously, data -based products are subject to a continuous change.
Initially listed Identify 15 strategies that help support the management of AI projects in the contradictory fields of stress in behavior.
Excess management expectations:
- Introduction to reality checks: Develop equipment such as checklists to show AI’s possibilities and limitations and to avoid unrealistic expectations.
- Start starting, earning and growing small: introduce pilot projects that gradually expand to achieve quick successes and reduce risks.
- Integrating external experts: Experts can create knowledge and best practices in early project stages.
AI projects are considered incorrectly like classic IT projects:
- Practice -Training of managers: Data scientists should express special features of AI in a practical way during projects.
- AI courses for managers: Classic training managers on AI and machine learning help to create basic knowledge.
- Define the appropriate major performance indicators (KPI): Develop success indicators that not only measure technological progress, but also the use and addition value of AI.
Missing AI integration in existing systems:
- Construction of technical bridges: Use middleware or robotic process automation (RPA) to connect the AI ​​system with the existing IT system.
- Give action -oriented recommendations: Convert predictions into concrete measures that can be easily applied to users.
- Use “Learning Apprentice” approach: Take AI system through observation of human work to generate training data.
- Promoting data consciousness: Indicate the lack of data quality and motivate departments for cooperation for better data base.
Explanation for model decisions:
- Use a post-hoc explanatory model: Provide models that make the results of complex AI models understandable.
- Priority to transparent models: Develop simple, easy-to-differential AI models that can still be distributed to accurate predictions.
- Cause focus on relationships: Use ideological models to better understand the experiments.
Dynamic environment and changing data:
- Monitor data drift: Set the system to detect data changes to ensure the quality of the model over the long term.
- Introduce mlops approaches: Immine the development and operation of AI models to ensure quick adjustment and strong processes.
Different approach
All five challenges can be explained with various approaches at the work of data science. Manager and users often see AI with a mechanical perspective that is based on (new) possibilities of automation and division of labor. As a result, these people often consider AI projects as an estimated and structured projects that can be applied using standardized methods and clearly defined results.
Data scientists, as the manufacturer of AI applications, experience their work as a creative and recurrence process, which requires experiment, deep experience, inherent expert knowledge and a flexible approach. These separate approaches create a contrasting voltage that can lead to misunderstanding and wrong control.
AI projects require combination
Stress should not be seen as an obstacle, but as an opportunity. Companies with a conscious balance between manual and mechanistic approaches can develop innovative and efficient AI solutions and create a permanent value. Both approaches bring the required aspects: the manual perspective helps find out which solution eventually cures the problem – it increases effectiveness. On the other hand, the mechanical perspective, ensures efficiency: AI projects continue to remain qualified and will be completed. In addition, a manual perspective can place people at the center of AI development and help us play a new division of labor.
(RME)
There is a risk in the use of AI DEPEEK
