Machine Learning Community Has Toxicity Problem
First , the peer review process has been disrupted. A quarter of the work from the NeurIPS conference is laid out on arXiv. DeepMind has researchers who publicly harass reviewers criticizing their submission to ICLR. In addition, articles from reputable institutes with arXiv are accepted at leading conferences, even if reviewers decide to decline the work. Conversely, some articles with most positive reviews are rejected (I don’t want to name any names, just take a look at this year’s openreview ICRL page).
Secondly , there is a reproducibility crisis . Apparently, setting hyperparameters on a test suite has become standard practice at the present time. Articles that do not outperform the best method to date have zero chance of being accepted at a good conference. As a result, hyperparameters are tuned in with subtle tricks to get performance gains where they don't exist.
Thirdly , there is the problem of worship . Each article related to Stanford or DeepMind is hailed as a breakthrough. For example, BERT cites seven times more than ULMfit. Belonging to Google gives the article a lot of trust and fame. At every ICML conference, there is a crowd of people in front of each DeepMind poster, regardless of the content of the work. The same story with Zoom meetings at the ICLR 2020 virtual conference. Moreover, NeurIPS 2020 collected twice as many applications as ICML, although both are top-level conferences. Why? Why is the word neural so lauded? Further, Bengio, Hinton and LeCoon [2018 Turing Prize winners for AI research - approx. per] are indeed pioneers of deep learning, but to call them “the godfathers of AI” is insanity. This is already becoming a cult.
Fourth , Yan LeKun spoke rather softly about the topics of bias and justice. However, in response received a completely inadequate toxicity and negative reaction. Getting rid of LeCun and plugging a person’s mouth is not a solution.
Fifth , machine learning and computer science as a whole have a huge problem of inequality (diversity). At our CS faculty, only 30% of students and 15% of professors are women. Taking parental leave during graduate or doctoral studies usually means the end of an academic career. However, this inequality is often abused as an excuse to protect some people from any form of criticism. Reducing each negative comment in a scientific discussion to race and gender creates a toxic environment. People are afraid to participate in discussions so that they are not called racists or sexists, which in turn strengthens the problem of inequality.
Sixth , morality and ethics are established arbitrarily . In any discussion, US domestic policy dominates. At this very moment, thousands of Uighurs enter concentration camps based on computer vision algorithms invented by this community, and no one seems to care about that at all. Adding a “Wider Impact” section at the end of each work will not solve this problem. Heaps of crap rise when some researcher is not mentioned in the article. Meanwhile, the African continent with a population of 1 billion people is practically excluded from any meaningful discussion of ML (except for a few Indaba workshops).
Seventh , the mentality of the “publish or die” type is common. If you do not publish 5+ articles a year at NeurIPS/ICML conferences, you are a failure. The research teams have become so large that the supervisor does not even remember the names of all graduate students. Some serve on NeurIPS more than 50 jobs per year. The sole purpose of this article was to add another NeurIPS article to the resume. Quality is secondary; The main goal was to go through the preview stage.
Finally , the discussions became disrespectful . Helmholtz Prize winner from the International Society of Neural Networks Jürgen Schmidhuber calls the thief of the Royal Society of London, Jeffrey Hinton, Ethiopian-American and co-director of the Ethical Artificial Intelligence Team on Google, Timnite Gebra calls the white supreme statist of “the godfather of AI” Jan LeKun, Professor I Kalteha at Nvidia, Anima Anandkumar calls sexist executive director of Geometric Intelligence and book author Gary Marcus.Everyone is under attack, but nothing is improving.
Albert Einstein opposed quantum mechanics . Please, can we stop demonizing those who do not share our views exactly? Let's let people disagree without cutting their throats.
The moment we shut people up because of their opinions, scientific and social progress is simply dying.