A negative externality arises when the private net marginal benefit (i.e. the marginal benefit minus the marginal cost) of an activity exceeds the net social benefit. In such cases, the self-interested private decision maker will increase their participation in the activity even though it is socially inefficient for them to do so.
A positive externality arises when the private net marginal benefit (i.e. the marginal benefit minus the marginal cost) of an activity are smaller than the net social benefit. In such cases, the self-interested private decision maker will not increase their participation in the activity even though it is socially efficient for them to do so.
This is a semantic distinction to the extend that if one thinks activity $A$ has a negative externality then one can define a new activity $B$, which is simply the "act of not doing $A$" so that $B$ has a positive externality. On this basis, one can argue that every externality is positive or that every externality is negative. For example, many people think that education has a positive externality because educated people make better citizens (e.g they make more informed voting decisions that benefit others). As a matter of semantics, one could argue that this is, in fact, not a positive externality and that what is really going on is that people who do not educate themselves are exerting a negative externality on those who do by virtue of their ignorance.
Whilst there is some merit to this reasoning, I do not find it helpful. Often, when we study the effects of behaviour, we are interested in comparing those effects to some baseline or benchmark in which the behaviour is absent. When communicating economics to others, it is usually the case that some benchmarks are more intuitive that others. We could, for example, rewrite all of consumer theory in terms of "the dis-utility people experience from not having goods" and look at the "problem of non-consumption dis-utility minimisation". Doing so would be formally equivalent to the more conventional approach of consumption utility maximisation (only the language changed), but would probably be less intuitive for people trying to understand the economics.(*) At least to me, it is more intuitive to think that people actively choose to undetake some level of education and exert a positive externality on others than to think that everyone receives infinite education by default and the choice of abstaining results in a negative externality.
Besides education, another example that I think fits most intuitively into the positive externalities box is network effects. If I buy a telephone then all of my phone-owning friends are made better off because now they can use their telephone to call one more person that they couldn't reach before. It seems weird to think of the negative externality of not owning a phone.
In terms of taxes versus subsidies: to get to the socially optimal intensity we need to ensure that the private net marginal benefit is zero precisely when the social net marginal benefit is zero. In the case of a negative externality this can be done either by increasing the private marginal cost (via a tax) for the activity or by increasing the private marginal benefit of not participating in the activity via a subsidy. For example, we could either subsidize low carbon firms or tax heavy polluters. As far as aligning incentives are concerned, the two are equivalent. In most practical cases, the more important consideration is likely to be that of budget constraints and politics:
- In the case of a tax: can the person you are taxing afford to pay the tax and can the tax be levied without seeming vindictive (for example, I think that taxing people without a university degree would be a no-no on these grounds).
- In the case of a subsidy: can the government raise enough popular support and funding for a subsidy without leaving people with the impression the some parties are receiving unfair government hand-outs?
In most cases, thinking about these political and financial constraints makes it clear whether a subsidy should be used. Sometimes a combination of both is used. For example, in the UK the government both taxes petroleum consumption and subsidises electric car ownership.
(*) Nevertheless, economists do often find it useful to convert the utility maximisation problems into their dual expenditure minimisation problems, which is somehow similar. This technique, though, is usually reserved for more advanced students who already have a well-developed intuition for the economics.