* A good quality ontology costs a lot of money, it cannot be done automatically. That is the difference for example between the KEGG PATHWAY database and the GO ontology.
The KEGG PATHWAY database is a collection of manually drawn KEGG pathway maps representing experimental knowledge on metabolism and various other functions of the cell and the organism while the GO ontology is the product of computer scripts. When students have no interesting results to show in an experiment, they do a "GO ontology enrichment" [0] and there is often funny stuff that shows up.
* There is a contradiction in having a large ontology to reflect some knowledge field and having to find relevant information in this large dataset. Many relations are weak signals, which one is relevant?
* An ontology reflects the limited knowledge and bias of its designers.
* You cannot express complicated relations because ontologies must have acyclic relations and even multiple inheritance is usually not possible, so they end using hierarchical relations which cannot be very useful. In addition there is no high level logic relations such as something being a function of one or several other items (high order logic). [1]
[0] https://en.wikipedia.org/wiki/Gene_Ontology_Term_Enrichment
[1] https://en.wikipedia.org/wiki/Web_Ontology_Language#OWL_Full
It seems you are better off asking whether all ontologies are bad first.
And no, I don't think ontologies are intrinsically bad. There's still hope NLP can make their creation and maintenance cost-efficient.