It amazes me that Machine Learning is all many people think of when they hear AI. Machine Learning (ML) is a subset of AI. You can also hear about Deep Learning (DL), a subset of ML where you have multiple layers/multiple models. There's a decent Quora thread on Deep Learning vs. Machine Learning [1].
Both ML and DL require a fair bit of data to do their training. There are some techniques to reduce training data, such as a Generative Adversarial Network (GAN) pitting an AI that generates data (some supposed to be fake and some legitimate) against an AI that tries to determine which of that data is legitimate.
However, ML and DL are not the only way to apply AI. There are a great many solutions you can solve with one or a combo of:
* Genetic Algorithms [2]
* Bayesian Inferencing [3]
* Semantic Reasoning using rules, and possibly using ontologies [4]
In some cases the approach might not generate a solution as good as from an ML-approach, but it also won't require the same level of data either.
Btw, semantic reasoning has a similar problem. Just like ML requires data, semantic reasoning requires an rules and/or an ontology, and creating a good ontology while avoiding the pitfalls is an art form.
For example: If I asked you to create classes for an ontology of vehicles, you might be fine with sea_vehicle, air_vehicle, and land_vehicle. But that's not going to work well. What about a submarine? What about an amphibious vehicle?
Instead, we can define the classes vehicle, movement_medium, as well as verbs moves_on, and moves_through. Now, a ship is a vehicle that moves_on the movement_medium water. A submarine though is a vehicle that moves_through and moves_on water. Both our sea_vehicles. Getting the ontology right is often one of the hardest aspects of applying semantic reasoning.
I recently had a great ML approach mapped out to solve a problem for a client, discovered the client hadn't kept the data they had gathered in a machine readable form (it was in notebooks and scattered PowerPoint decks), and we pivoted to using a Genetic Algorithms approach.
Both ML and DL require a fair bit of data to do their training. There are some techniques to reduce training data, such as a Generative Adversarial Network (GAN) pitting an AI that generates data (some supposed to be fake and some legitimate) against an AI that tries to determine which of that data is legitimate.
However, ML and DL are not the only way to apply AI. There are a great many solutions you can solve with one or a combo of:
* Genetic Algorithms [2]
* Bayesian Inferencing [3]
* Semantic Reasoning using rules, and possibly using ontologies [4]
In some cases the approach might not generate a solution as good as from an ML-approach, but it also won't require the same level of data either.
Btw, semantic reasoning has a similar problem. Just like ML requires data, semantic reasoning requires an rules and/or an ontology, and creating a good ontology while avoiding the pitfalls is an art form.
For example: If I asked you to create classes for an ontology of vehicles, you might be fine with sea_vehicle, air_vehicle, and land_vehicle. But that's not going to work well. What about a submarine? What about an amphibious vehicle?
Instead, we can define the classes vehicle, movement_medium, as well as verbs moves_on, and moves_through. Now, a ship is a vehicle that moves_on the movement_medium water. A submarine though is a vehicle that moves_through and moves_on water. Both our sea_vehicles. Getting the ontology right is often one of the hardest aspects of applying semantic reasoning.
I recently had a great ML approach mapped out to solve a problem for a client, discovered the client hadn't kept the data they had gathered in a machine readable form (it was in notebooks and scattered PowerPoint decks), and we pivoted to using a Genetic Algorithms approach.
[1] https://www.quora.com/What-is-the-major-difference-between-d...
[2] https://www.geeksforgeeks.org/genetic-algorithms/
[3] https://brohrer.github.io/how_bayesian_inference_works.html
[4] https://rubenverborgh.github.io/Semantic-Web-Reasoning/#