ML-driven AI vs. generative AI: A strategic overview

- [Instructor] Now that we've covered the STAR method for strategic decision making, let's explore key choices in AI development. We'll be looking at the differences between ML-driven AI versus generative AI. Understanding their differences will help you to collect data, train models, and achieve your product goals effectively. The key difference comes down to control versus convenience. With ML-driven AI, you build and train your own models. That means choosing the right algorithms, optimizing performance, and working with frameworks like TensorFlow or PyTorch. You have full control, but it takes more time and expertise. With generative AI you use a pre-trained model like GPT-4o or Sonnet-3.5 to generate text, images, or code. Instead of training a model from scratch, you send input through an API and get results instantaneously. It's fast and easy, but you have less customization. The bottom line, if you need full control, go with AI-driven AI. If you want a quick, powerful solution, generative AI might be the better fit. Let's look at how these two approaches might play out in healthcare. With ML-driven AI, you might build a model to predict patient readmission risks. You'd work with structured data like patient history, treatment outcomes, and vitals, aiming for high accuracy within a specific prediction window, say 48 hours. But with generative AI, you could analyze unstructured medical notes to generate detailed patient risk assessments. Instead of just making the predictions, the model could summarize doctors' notes, extract key insights and improve documentation quality. Both are valuable, but they serve different needs. ML-driven AI focuses on structured predictions, while generative AI helps interpret and generate insights from complex data. IBM for Oncology is a somewhat notorious example of ML-driven AI in healthcare. It was designed to help doctors by analyzing medical literature, patient records, and clinical trial data to recommend cancer treatments. Unfortunately, this product did not work as expected and it ended up costing IBM a lot of money. But just in terms of how it worked, instead of generating new content like a generative AI model would, Watson used natural language processing and structured reasoning to retrieve relevant medical insights. Despite problems that this product ultimately had in the market, this example shows how ML-driven AI has been used in healthcare practice to analyze large data sets and make evidence-based recommendations rather than creating entirely new outputs. Now, let's compare these AI approaches in financial trading. With ML-driven AI, you might build a model to predict stock movements using structured data like historical prices, trading volumes, and market trends. Techniques like time series analysis help forecast specific stock performance. With generative AI, you could analyze market reports, news articles, and social media sentiment to generate insights or suggest trading strategies. Instead of making direct predictions, the model processes the unstructured data to identify emerging trends. Both have value. ML-driven AI is great for precise data-driven predictions, while generative AI excels at analyzing broader market context and generating insights. No matter which AI approach you choose, the STAR framework helps you make better decisions. First, survey the industry. Look at how other companies use AI for financial trading. Are they relying on ML-driven models for structured market data or generative AI for trend analysis and sentiment tracking? What's working and what challenges have they faced? Next, you'd take stock of your own resources. Do you have high quality structured data in a team capable of building and maintaining an ML-driven model? Or does generative AI make more sense because you need insights from unstructured data like news articles and social media trends? Then, assess your current infrastructure and capabilities. Can your system handle real-time stock predictions with ML-driven AI, or do you need a model that processes large volumes of unstructured text? Identify gaps that could limit your ability to execute. Finally, recommend a clear implementation plan. If you choose ML-driven AI, focus on high-accuracy predictive modeling. If you choose generative AI, ensure proper filtering and validation to avoid misleading insights. Consider scalability risks in how the AI system fits into your existing workflow. By following the STAR framework, you'll make a well-informed choice between ML-driven AI for structured predictions and generative AI for broader market analysis. This helps you to ensure that your strategy is practical, scalable, and aligned with your business goals.