14 The Myth of Hard & Soft Sciences
Take-home Message — Realization 2
Hard & Soft are also used as inadequate, hierarchical adjectives for Scientific Disciplines
14.1 Hard & Soft Disciplines
So far we’ve considered how the words hard and soft describe skills. But they also describe scientific disciplines. Returning to this use of hard & soft will help us to appreciate what really distinguishes these skills and disciplines.
The value perception of a field has serious consequences for research. The lower perceived value of soft skills translates to the same lower perceived value of soft sciences leading to poorly-supported research areas, both financially, institutionally and publicly. Ethics, privacy, transparency, explainability, reproducibility and human rights have potentially all suffered as a result. Aside from under-supporting valuable research that may only indirectly relate to work in STEM fields, these areas are directly relevant to how we do our work and the impact it has on society.
14.2 Revisiting the Hard-Soft Value Hierarchy
Similar to hierarchy in skills, we can once again observe a value hierarchy of scientific disciplines.
- Level 1: All Scientific Enquiry
- STEM (hard sciences) above,
- Non-STEM (Everything else) below
- Level 2: within STEM, subtler hierarchies exists
- Pure maths, theoretical physics above,
- Applied maths, chemistry, physics below
- Maths of uncertainty, biology, ecology, etc., further below,
- Interdisciplinary and systems approaches, further below still
- Level 3: within disciplines, subtler hierarchies exists
This is less an appreciation of distinguishing features and more a tallying of “pureness”, reinforcing masculine traits of dominance and power (explainability) and elitist attitudes of purity and exclusivity, even among STEM fields. Indeed, when considering disciplines, as I’ve argued with skills, hard & soft reinforce a misogynistic STEM culture that diminishes the value of everything soft. The long long history of systematic sexism in STEM fields, when women were explicitly excluded from the scientific endeavor, is casually, consistently and persistently reinforced. It is the basis which supports communication as a female pursuit, not worthy of the masculine disciplines. It helps explain the late integration of biology as a “real science”, although it was acceptably practiced by women outside of academia (e.g. Beatrice Potter). It explains how Rosalind Franklin’s essential contribution to the discovery of the double-helix structure of DNA was unrecognized for so long.
The hard/soft value hierarchy does us all a disservice.
14.3 Metrics used to define Hard & Soft sciences are outdated oversimplifications
- The metrics which distinguish hard and soft scientific disciplines are provided in the following table and are generally taken as binary.
- This is the same false dichotomy that happens in the examination of skills.
| Metric | Hard Science | Soft Science |
|---|---|---|
| Falsifiable hypothesis | ✓ | ✗ |
| Controlled experiments | ✓ | ✗ |
| Quantifable data | ✓ | ✗ |
| Mathematical models | ✓ | ✗ |
| Objectivity | ✓ | ✗ |
| High Accuracy | ✓ | ✗ |
| High level of consensus | ✓ | ✗ |
| Applied/practical | ✓ | ✗ |
| Explanatory success | ✓ | ✗ |
| Cumulativeness | ✓ | ✗ |
| Reproducible and replicable | ✓ | ✗ |
| Scientific method | ✓ | ✗ |
14.4 Outdated Terms
As with skills, we can see how hard and soft scientific disciplines rarely still exist. It’s possible that these metrics were clear, diagnostic binary metrics at some point in the past. However, in the current research environment, supported by modern technology and the scale it facilitates, it seems unlikely that many disciplines can still be characterized so simply as hard or soft. Quantitative methods, typically associated with STEM are routinely used in non-STEM fields, e.g. digital humanities, Natural Language Processing, experimentation, the scientific method, etc. Practices which would have once been frowned upon in STEM are now routine:
- Gathering data and then developing a testable hypothesis,
- Machine Learning & Deep Learning models which are so complex that they need further study just to understand
- The introduction or both randomness (dropout deep learning) and stochastic processes
Nonetheless, there is a dichotomy that can be applied to both skills & scientific disciplines, but it’s not the hard/soft. So what it is?