Ako využiť umelú inteligenciu s pomocou dát, inovácií a regulácií (10:55

AI projects don’t fail as often as people tend to claim, says Phil Winder, CEO of Winder AI. Based on his experience and surveys, over 70% of companies report that they use artificial intelligence. In his talk, he summarized why data is decisive, which use cases make sense, and what regulation brings.

Why Data Determines Outcomes

AI success rests on representative and clean data: free of errors, corruption, and nonsensical values. Developers spend most of their time sourcing, moving, and managing data, not training models. Investment in data management and quality management therefore pays off sooner than the chase for a “magical” algorithm. Moreover, if you can’t say where your data is and how it’s used, you’ll run into compliance issues.

Transparency increases trust and enables better decision-making. Winder showed a map of average earnings in England and Wales and then the same picture after taking housing costs into account – the north–south divide was rearranged and new regions emerged that would deserve support. With the Corruption Perceptions Index from Transparency International, he pointed out that looking at rankings and error bars changes the interpretation of trends. Open data thus makes it possible to create useful derivatives from which citizens, business, and politicians can benefit.

From Idea to Practice: Which Use Cases Work

Stable Audio’s music generation shows how genAI shortens producing backing tracks from weeks to minutes. An application for Temple University combines language models to find the right legal texts, segment them, and answer questions by jurisdiction. In both cases it’s not just a “wow effect,” but a scalable saving of time and costs. The biggest change in recent years is that we approach AI through natural language, so people outside IT can use it too.

The quirky “AI Jesus” reminds us of the technology’s creative potential, while the project The Ocean Cleanup addresses real risks by detecting plastics and marine animals. Key takeaways: choose the right problem the company actually has; focus on tasks with a high volume of decisions; “garbage in, garbage out” applies. And design systems to collaborate with people – recommendations are accepted more readily than full automation and speed up adoption.

Regulation, Privacy, and Custom Models

Alongside the GDPR comes the EU Artificial Intelligence Act, the first law aimed directly at AI developers and operators. The approach is risk-oriented: some applications are banned, for high risk the requirements for testing and documentation increase, and even low-risk systems must be transparent to users. It makes sense to prepare in good time processes that demonstrate what the system does, on what data, and with what limitations. These rules make internal traceability easier and increase trust.

Data sovereignty remains an open issue: copyright protects content, but disputes over “fair use” in training models are still unresolved. Hence the growing interest in private LLMs and solutions in the company’s data centers, so that sensitive or proprietary data do not leave the organization. Strong data governance, an audit trail, and the transparency principles from regulation also work for internal corporate data. And let’s not forget: AI is a tool – people are the ones who define the problem, choose the right tools (often practical ones like Python with PyTorch), and bear responsibility for the outcome.

Download presentation (353kB)

Harnessing AI With Data, Innovation, and Regulation

Why Data Determines Outcomes

From Idea to Practice: Which Use Cases Work

Regulation, Privacy, and Custom Models

Phil Winder

Páčil sa ti článok? Zdieľaj ho a povedz o ňom aj ostatným