At SlavConf, Jayaprakash Sundararaj, Lead Engineer at Google, approached artificial intelligence from a position that is often missing in public discussions. Instead of presenting AI as a disruptive force in isolation, he framed it as the result of long-term technological evolution, shaped by decades of research, failed expectations, and recent breakthroughs in data and computation.
Drawing from over twelve years of experience working on large-scale systems such as web search and recommendation engines, his talk moved between foundational concepts and practical entry points into the field, offering a grounded perspective on how AI actually works and how people can realistically begin working with it.
A central part of the session focused on clarifying what artificial intelligence and machine learning represent in practice. Rather than treating them as abstract or futuristic ideas, he described AI as a collection of systems designed to process information, identify patterns, and make predictions or decisions. Machine learning, as a subset of AI, operates by learning from data alone. Given sufficient scale, these systems are able to perform tasks that would otherwise require complex human reasoning, from predicting weather patterns to recommending content or detecting fraud.
What stood out in this explanation was not the definition itself, but the emphasis on scale. The effectiveness of modern AI systems is directly tied to the volume of data and computational power available today. This is what allows models to move beyond simple automation into more complex forms of prediction and interaction. The result is a layer of intelligence embedded in everyday systems, often invisible but consistently shaping user experience.
Instead of presenting a single path, Sundararaj outlined two distinct approaches. The first is the traditional technical route, which involves a strong foundation in mathematics, particularly linear algebra and probability, combined with programming skills in languages such as Python and familiarity with frameworks like TensorFlow and PyTorch. This path is essential for those aiming to build or optimize models at a deeper level.
At the same time, he highlighted an alternative entry point that reflects the current state of the ecosystem. Many modern tools provide layers of abstraction that allow users to build applications without needing to fully understand the underlying mathematical models. This opens the field to a broader range of participants, including developers, designers, and creators who are interested in applying AI rather than developing it from first principles. This distinction between building models and building with models is becoming increasingly important, especially as AI tools become more accessible.
When discussing practical experience, the advice was deliberately restrained. Rather than encouraging ambitious or large-scale projects, Sundararaj emphasized the importance of starting with small, well-defined problems. He referenced projects such as recommendation systems and machine translation as effective entry points, not because they are simple, but because they offer clear feedback loops and are easier to understand in terms of how inputs translate into outputs.