Let’s look at each of these elements in detail.

- Symbolic & Sub-symbolic AI. AI can be broadly broken into logical (or symbolic) AI and Statistical (or subsymbolic AI). Symbolic AI includes topics like Knowledge Representation, Logical Reasoning, and Planning. The key idea is that knowledge is represented and reasoned upon explicitly. This branch of AI was popular till the 90s with the rise of expert systems, rule formalisms, and logical programming languages like Prolog. Sub-symbolic AI includes topics like Machine Learning, including Deep Learning. The key idea is that knowledge can be learned with sufficient data. This branch of AI dramatically increased in popularity since the 2000s, and today is probably the dominant branch. Most AI and Data Science courses focus on subsymbolic AI, but I feel that a true AI system will need a combination of Symbolic and Sub-symbolic techniques. Hence, I believe that AI researchers must focus on Symbolic AI and its co-existence with Sub-symbolic AI.

- Algorithms & Ops. A lot of emphasis today is placed on different algorithms and techniques for processing data, building models and evaluating or tuning them, but not on the operationalization of these algorithms for production settings. I believe that an AI curriculum must cover different aspects of AI or ML Ops like robustness, scalability, and monitoring. Furthermore, it also needs to cover the data architecture – how is data sourced, transformed, stored, and made available for training and scoring. This is especially important for AI applications architects.

- Theory & Applications. I believe the AI curriculum must have a good mixture of theory and applications. For example, for various Machine Learning models, AI professionals must understand both the internals of the training and scoring algorithms, as well as various applications of the model (for solving different problems in different industries). A good understanding of the theory of the algorithms is essential to understand its behavior and pros and cons. And the use-cases are important so that a Data Scientist or a Researcher quickly knows what algorithms to apply for different kinds of problems they might encounter.

- Formal Rigor & Story-telling. Todays’ AI curriculum focuses on formal rigor – i.e. making sure that the algorithms and techniques perform well in a formal sense (e.g., they have high accuracy, precision, recall, or other metrics). Not as much emphasis is put on storytelling, i.e. creating a story about the data. The story about the data is critical to convey the results to business or other stakeholders in an easy-to-understand manner. The story might include different elements like an overview of the key takeaways, a description of the source and properties of the data, a diagnosis of key findings, predictions of what might happen next, and prescriptions of what actions might be taken to optimize some outcomes. The stories help weave different elements of the data and the models to communicate insights effectively to the listener. This is especially important for Data Scientists.

For a high-school student considering a career in the AI space, I would recommend looking for courses that cover a gamut of the topics I mentioned above. This will help them acquire skills that are immediately useful in the marketplace, as well as skills that will be useful even in the future as the state of AI evolves. Furthermore, I’d recommend them to take inter-disciplinary courses to get more ideas of interesting ways of applying AI in different disciplines.