Instrumental variables are those that are highly correlated with the causal variable understudy but unlikely or not correlated with the response under measure. These variables can then be used to approximate the effect the causal variable in the absence of correlated confounding variables. In other words, when you are trying to disambiguate a causal variable from confounding correlated variables, this can be a useful technique1.
References
1.
Ben Lambert. Instrumental Variables - an introduction. at https://www.youtube.com/watch?v=NLgB2WGGKUw (2013).