AI Navigator #4: Phoenix from the ashes — AI needs a long-term perspective

Welcome to the fourth edition of the DOAG KI Community’s AI Navigator column!

Advertisement

Dr. Benjamin Linnik, who has a doctorate in nuclear physics, specializes in data science, software engineering and IT consulting. As a Lead DevSecOps Engineer, he has gained extensive experience in cloud-based data-driven projects. His commitment is also reflected in his leadership of the Data Science Meetup in Nuremberg. In his private life, he likes to relax with his family surrounded by two cats and loves automating his smart home.

SUSE buys observability platform Stackstate

A look at AI reveals the difference between success stories of technological achievements and projects that fail in practice. On the one hand, there are reports about AI tools that perform difficult tasks, recognize complex relationships in data, enable new treatment methods in medicine and help better integrate renewable energy into the power grid. But operational practice shows that many AI projects failed: government chatbots recommend illegal activities, image generators produce racist images, inadequate facial recognition causes for false arrestNutrition Apps suggests toxic remedies and there are robots Barely able to assemble an IKEA chair.It seems as if the real world is too complex to be handed over to AI.

Long-term AI initiatives instead of temporary AI projects

To learn from the mistakes of the past, we must move away from a short-term focus on AI projects and instead promote sustainable AI initiatives. These initiatives cover the entire life cycle of AI systems and allow for a philosophy of learning systems rather than moving from one learned model to another.

A project view often stands in the way of such a project, because projects are handled according to the magic triangle, i.e. balancing project scope, cost and schedule. If there is not enough money or time, teams have to compromise on the project scope. For example, in software projects, functions may not be delivered or may be delivered later. This balancing of the triangle is much more difficult in AI applications because comprehensive models often only produce results at the end, for example recognizing cancer cells on an X-ray image. This severely limits flexibility and adaptability.

The fact that companies use AI in a variety of areas is another reason to look at applications in the long term. Although AI can automate individual tasks or entire business models such as ordering in an online shop, these areas of application will probably remain exceptions in the medium term. Mostly there will be assistance systems that contain AI components and support human decision makers or perform unpleasant tasks, for example by pointing out errors or suggesting options for action.

In addition – and this is often overlooked – machine learning methods can provide novel insights into data and thus generate knowledge. While the added value of automated, intelligent agents can be easily measured, the advantages of AI-based assistance systems or purely knowledge-generating systems are not immediately obvious. However, in the long term, a cleverly made strategic decision can be far more valuable than an impressive customer service robot.

A long-term look at AI applications shows that their value contribution to companies can change over time: some projects start too ambitiously and fail due to the demands, for example if the predictions with the given data are ultimately not good enough. Other projects start small and after some time you realise that the analytics can support routine tasks and later – after extensive testing – perhaps automate them completely. This rapid shift in operational value creation in AI projects has only recently occurred. An empirical scientific study has shown and worked out the circumstances under which such a change might occur.

What does this ability of AI applications to change mean for developers and decision makers? The dynamic nature of AI projects requires the inclusion of continuous adaptation and feedback loops to ensure that systems remain effective and meet the real needs of the business.

Supposedly failed projects, like a phoenix rising from the ashes, can create value in the company if the people involved share their experiences and try to gain knowledge from the AI ​​models. If planned automation is not possible, a support system that speeds up processes or increases employee satisfaction may be an option instead.

But even successful projects can come to an abrupt end if, for example, the legal framework changes, as is the case with the EU AI Act, or society changes, as is the case with the EU AI Act. Consumer Behaviour during the COVID-19 Pandemic,

In such scenarios, companies must not only respond quickly but also be prepared to rethink their approach and adapt their value creation mechanisms accordingly. This could mean recalibrating ML models or shifting from fully automated processes towards human decision-making as they develop and test new, optimized models.

The practice of readjustment and the integration of feedback loops not only helps companies recover from crises, but also helps them continuously improve their AI systems and design them so that they can make valuable contributions to the company’s success in the long term. AI initiatives benefit from companies’ agile mindset that enables adaptation to constantly changing requirements and environments. This mindset promotes rapid iterations and continuous improvements, which are essential for the best use and calibration of AI systems. This is especially important in fast-moving fields such as artificial intelligence.

Data Scientists: The Architects of AI Value Creation

When planning and implementing AI initiatives, the initial focus is on managers or team leaders. But data scientists and related professional groups such as data engineers or ML specialists also play a central role in managing AI initiatives. You act as a bridge builder between technical teams and management.

Your ability to interpret complex data patterns and translate them into business-actionable insights is critical to maximizing the value of AI. Their involvement in all phases of an AI initiative – from data acquisition to model implementation and optimization to monitoring – makes them indispensable players in effectively using AI in the company.

Long-term perspective and agile approach

Companies need to take a long-term approach to AI to get the full benefit from it. Individual projects may fail, technological developments may overwhelm them. A long-term approach combined with agile methods makes it possible to optimize the approach and not only achieve the set goals, but also benefit from all aspects of AI.

(RME)