1. Requires large amount of data in order to train the models to achieve good results
  2. Requires a lot of GPU and TPU power which makes it much more expensive to develop
  3. Training models don’t often have clear ways to show how it resolves the problem, which is troublesome in sectors like justice and healthcare.
  4. These models are trained to choose based on correlation of results instead of causation
  5. Tendency to remember training data instead of new real life sets of data
  6. Although AI models may be adaptive to new sets of data, symbolic AI is still more efficient in completing narrowly focused tasks
  7. AI may encounter certain dilemmas

References: