Machine learning 'causing science crisis'
Techniques used to analyse data are producing misleading and often wrong results, critics sayRead full story
A growing amount of scientific research involves using machine learning software to analyse data that has already been collected. This happens across many subject areas ranging from biomedical research to astronomy. This is causing a reproducibility crisis with results that can’t extend beyond specific data sets. What is real when your experiments are virtual?
Machine learning is not a silver bullet - nor is its close relative artificial intelligence. It’s another tool in the scientific toolbox, albeit a big and important one. Like most tools, it can rarely be used in isolation.
This is a parallel to what is happening in many areas where ML is viewed as a panacea, a “causation is dead” argument. We should give up on that line and respect that patterns and correlation are not sufficient to fuel sustainable discovery.
A very few in every industry knows exactly what’s going to happen when technology takes over. And no, it’s not about drones delivering stuff like Amazon make us believe.
Is about telling us what street is the more likely to be full of cars, block, closed, or completely free. And we already trust in those technologies that are actually profitable and have a lot of followers using several apps to get fom one place to another following previous paths. It has built also a great community, a visible one, but what about the hidden ones?
So much for using the fundamentals of scientific research and the scientific methods. Aren't students getting taught that any more? Don't executives follow and adhere any more? I suppose not.
Machine learning at this point is like asking a first grader to analyze the data and they only see patterns. No learning beyond that. Just give humans the jobs.
Now, let's take this a step further, just in case you weren't adequately frustrated: these deeply flawed results are used beyond academic publication, but rarely if ever dies anyone retract known bad results.
This is increasingly the kind of shoddy science upon which policy is based.
If models do not reflect the real world, then the results are inaccurate. A lesson not only for scientists but also policy makers.
A community of leaders, subject matter experts, and curious minds bringing nuance back to how we talk about the news.
No content overload: our editors will curate the most notable and discussion-worthy pieces for you every day.
Don’t just read the story, tell it: contribute your ideas and experience to the dialogue.