Generative AI has recent genesis while predictive AI has been around for quite some time. Generative AI models create new content based on the patterns and data they have been trained on. Predictive AI, on the other hand, forecasts outcomes based on historical data.

Generative AI models hallucinate — they generate plausible but factually incorrect outputs. These models also show bias which they derive from the training data.

Predictive AI is used in finance, medicine, stock price prediction, patient diagnosis and customer behaviour analysis. There are challenges here too. They fail to mould to unseen data so far — leading to inaccurate predictions. The quality of historical data affects their performance. Prediction in the field of medicine could be problematic. GANs. There is confusion between correlation, and causation. In correlation, one number is in fact causing the other number to move in tandem. This causes errors.

Both the models require extensive manual training and continuous improvement. The nature of this training differs. Generative AI requires diverse and well-curated datasets to train. There are often human feed-back loops. Predictive AI needs clean, relevant and comprehensive historical dataset. It is necessary to identify, and create right features that the model will use to make predictions. It requires domain experience.

Google DeepMind used techniques such as GANs. They also exported self-supervised learning.