0
$\begingroup$

From Wikipedia

Causal AI is a technique in artificial intelligence that builds a causal model and can thereby make inferences using causality rather than just correlation. One practical use for causal AI is for organisations to explain decision-making and the causes for a decision.

Are the AI like LR, SVM or ANN correlated? If yes, how come?

$\endgroup$
2
  • 2
    $\begingroup$ The statement does not say that "AI is correlated" but that "they make inferences using correlation". So, maybe this question should be reformulated. Also, can you please provide the link to where we can find the quoted text? $\endgroup$
    – nbro
    Commented May 24 at 12:42
  • $\begingroup$ I've tried to save this question by making some assumptions about what you really want to ask and about the source of the quote, but please next time do try to write a clearer question and provide the link to the article that you're quoting. $\endgroup$
    – nbro
    Commented May 24 at 15:13

2 Answers 2

0
$\begingroup$

Most predictive machine learning, that creates model with input and output, is completely agnostic to how or why two the two sets of variables (inputs and outputs) are linked.

As such the model learning process will use any observed correlation between inputs and outputs observed to inform the learning process and reduce loss.

This doesn't prevent the process creating an approximate causal model. However, unless you put the effort in to uncover causation and structure the learning process to model it, then you cannot tell.

This problem is not limited to machine learning. It's a common issue in scientific studies of all kinds, and you will often hear the warning "correlation doesn't imply causation", when e.g. studies show people who eat a particular diet have a certain incidence of an interesting health outcome. It is rarely strong proof that the diet causes the health issue, at least not on its own.

To answer your question more directly, then yes most uses of ML build models of correlation. However, this is not really a model class issue. You could structure experiments and data collection to isolate causation using many of the same model types.

$\endgroup$
0
$\begingroup$

Traditional AI Techniques and Correlation Linear Regression (LR): Think of linear regression as drawing a straight line that best fits a bunch of data points on a graph. This line helps predict future points based on past data. However, it only shows how things are related, not if one thing causes the other.

Support Vector Machines (SVM): Imagine trying to separate different groups of data points with a clear boundary. SVMs find the best way to do this. They work based on patterns in the data, showing how different points are grouped together but not why they are grouped that way.

Artificial Neural Networks (ANN): ANNs are like complex webs that learn to make predictions by looking at a lot of examples. They find patterns in the data and use these patterns to make guesses about new data. However, they also don't understand why these patterns exist, only that they do.

Why These Techniques Are About Correlation Looking for Patterns: All these methods are about finding patterns in data. They can tell you that two things often happen together but not that one thing makes the other happen. No Cause and Effect: They don’t figure out if changing one thing will change another. They just know that when one thing changes, another thing often changes too. Coincidences: Sometimes, they find connections that are just coincidences and not really meaningful. Causal AI: Understanding Cause and Effect Causal AI works differently:

Causal Diagrams: Think of these as maps that show how things are connected in a cause-and-effect way, not just that they are related. Testing Changes: Causal AI involves making changes to see what happens. For example, changing one variable to see if it affects another. What-If Scenarios: It can answer questions like, "What would happen if we did this differently?" This helps in understanding the real impact of changes. Why This Matters Traditional AI can tell you that two things are related, which is useful for making predictions based on patterns. But it can't tell you if one thing causes another. Causal AI, on the other hand, can help explain why things happen, which is crucial for making informed decisions and understanding the true impact of changes.

In essence, traditional AI methods are like seeing a pattern without knowing why it's there, while causal AI tries to explain why the pattern exists in the first place.

$\endgroup$
1
  • $\begingroup$ Don't use chatgpt/gemini $\endgroup$
    – quanity
    Commented May 25 at 7:36

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .