An article on 'Machine Learning Causing Science Crisis' and my response...

A friend shared an article titled ' 'AAAS: Machine learning 'causing science crisis'. The article is available here

Screen Shot of the Article

Here is my response

Read through the article. it kind of touches upon an interesting area, but quite vague in its coverage.  Need the original paper/talk for more details. I googled, but couldn't get it other than the coverage on other papers like FT.

Here are my comments

1. If the hypothesis that you want to check doesn't align with the results, it is always advisable to change the hypothesis and redo the experiment . Modifying the hypothesis to suit the result is a practice people are not advised to get into. This is a general rule to be followed in data science

2. On the reproducibility crisis, i am sure i will understand it better with an example. right now, looks only a n hypothesis. This statement "“Often these studies are not found out to be inaccurate until there's another real big dataset that someone applies these techniques to and says ‘oh my goodness, the results of these two studies don't overlap‘," she said." simply doesn't make any sense to me  :-) 

3 the word 'reproduction' is bit dicey here. if ML suggests a pattern, what will be needed is to verify it using an independent experimental set up. In the absence of which for whatever reason - like a clear articulation of the problem to be verified, the cost of building the whole experimental study and many such factors - the pattern may just be that - generating such a pattern from another data set may not help much in this regard - other than strengthening the presence of a pattern, subject to terms and conditions

3.1 The Pattern has to be still verified independently in science and in business we need to see if the pattern makes any sense and is of use to us. Such experiments cost money and we have to be careful about where we spend our money in both science as well as business 

4. I recall - from a talk on gravity waves at Anna Centenary Library, Chennai- how setting up an observational set up in outer space takes time and needs lot of money - so experiment will always be behind hypothesis 

5. I personally feel that AI can help identify newer theories in areas like Gravitational waves or astronomy - and scientists should learn to look at it as an enabler 

6. We will see more / hear more on this area in the coming months/ years