Slashdot Mirror
Misleading Results From Widely-Used Machine-Learning Data Analysis Techniques (bbc.com)
Long-time Slashdot reader kbahey writes: The increased reliance on machine-learning techniques used by thousands of scientists to analyze data, is producing results that are misleading and often completely wrong, according to the BBC.
Dr. Genevera Allen from Rice University in Houston said that the increased use of such systems was contributing to a "crisis in science".
She warned scientists that if they didn't improve their techniques they would be wasting both time and money. Her research was presented at the American Association for the Advancement of Science in Washington.
This is the oft-discussed 'reproducibility problem' in modern science.
The BBC writes that this irreproducibility happens when experiments "aren't designed well enough to ensure that the scientists don't fool themselves and see what they want to see in the results." But machine learning now has apparently become part of the problem.
Dr. Allen asks "If we had an additional dataset would we see the same scientific discovery or principle...? Unfortunately the answer is often probably not.â
Dr. Genevera Allen from Rice University in Houston said that the increased use of such systems was contributing to a "crisis in science".
She warned scientists that if they didn't improve their techniques they would be wasting both time and money. Her research was presented at the American Association for the Advancement of Science in Washington.
This is the oft-discussed 'reproducibility problem' in modern science.
The BBC writes that this irreproducibility happens when experiments "aren't designed well enough to ensure that the scientists don't fool themselves and see what they want to see in the results." But machine learning now has apparently become part of the problem.
Dr. Allen asks "If we had an additional dataset would we see the same scientific discovery or principle...? Unfortunately the answer is often probably not.â
The nutter Amerikuks are sure Zombie Jeebus is going to rapture them all away from the consequences of their selfish luves so fuck it! The Earth will go on, humanity.. answer not clear, try again.
Seems like a little run on growing distrust with AI. Predictive policing fails, image recognition Tom Foolery, and now basic science falling victim. Please, somebody think of the children!
Something I never liked about machine learning, and 'new fangled AI' in general is how opaque it is, you get fast interesting results but you can not explain how you got them or defend them directly. But GOFAI techniques are out of style right now, and it is getting worse as GOFAI systems are so much slower and resource intensive not to mention require so much more domain knowledge to set up and just can not compete with the sexy instant gratification that machine learning can give you.. or give your customers/sponsors.
I worked as a ML researcher in a science lab. Was often asked for results they wanted rather than good methodology, which I pushed back hard on, but the lab frequently contracted out analysis and then chose which results they liked best for publication. They got a few publications in Nature. Don't trust the ML results of any science paper unless they fully present and you understand their data, methodology, and statistics, and even then take things with a grain of salt.
Yes, wel, that is what one does, look for patterns in the data. But the idea is that the data is a good representation of the real world, and that the patterns you find can be generalised to something useful. If you are just drawing conclusions from whatever your machine learning algorithm finds in the data, you need to look over your method, research questions and evaluation.
(Discalimer, the article doesn't give any details, and briefly mentions astronomy and biomedical research, areas I am not too familiar with, but I would think that what I mentioned is common practice in all scientific research.)
It was a common phenomenon to be observed ever since a complex methodology existed that researchers, especially the most successfully extraverted, did not understand what they were doing, analysis-wise. But it is a relief, I suppose, for machine-learning metholodists, that they for sure find the find-what-I-want switch easily. Amen ...
TaijiQuan (Huang, 5 loosenings)
This is exactly how a world without education can seemingly function. Idiocracy is not an impossible parody, but one possible future.
Q: "What do you see here?"
A: "Ummm... a meerkat!"
It reveals more about the science community's collective subconscious, I guess.