The naive picture has scientists observing neutrally, generalising from data, and testing predictions, with objectivity guaranteed by keeping personal views out. Nearly every part of this has been challenged — and the honest response is neither to defend it nor to conclude that science is just another opinion.
Start with observation. What you notice, and what counts as a relevant fact at all, depends on the framework you bring. A radiologist and a layperson looking at the same scan do not see the same thing. This is theory-ladenness, and it means observation cannot serve as a wholly neutral court of appeal between theories, since the theories partly shape what registers as evidence.
Next, underdetermination. For any body of evidence, more than one theory can in principle accommodate it — especially given that auxiliary assumptions can always be adjusted, which is the Duhem-Quine point. If evidence alone does not uniquely select a theory, something else contributes: simplicity, explanatory scope, fruitfulness, coherence with existing knowledge. These are reasonable criteria, but applying them is judgement rather than calculation.
Then there are values, and this is where discussion usually goes wrong. Some value influence is straightforwardly bad — funding pressure distorting results, motivated reasoning. But other value involvement is unavoidable and legitimate. Deciding how strong the evidence must be before accepting a hypothesis depends on the consequences of error: the standard appropriate for a drug's safety differs from one appropriate for a cosmological claim, and that difference is a value judgement, not a contaminant. Philosophers call this the argument from inductive risk, and it implies that a value-free science was never the right target.
Where does that leave objectivity? The most influential answer relocates it. Individual scientists are biased; the process can nonetheless be objective, because it subjects claims to organised criticism from people with different commitments and incentives to find errors. Peer review, replication, adversarial scrutiny and open data are the machinery. On this view objectivity is not the absence of perspective but the presence of effective criticism — which explains why diversity of viewpoint within a scientific community improves reliability rather than compromising it.
It matters to state the upshot clearly, because these arguments are routinely misused. That science is not objective in the naive sense does not make it arbitrary, nor put it on a level with unmethodical opinion. It means the reliability comes from something more interesting than individual neutrality: a structured social process that catches errors better than any individual could. Recognising that is what makes it possible to improve — through replication standards, pre-registration, and transparency — rather than merely to trust.