Gary Smith

There is no doubt science is currently suffering from a credibility crisis. This thought-provoking book argues that, ironically, science's credibility is being undermined by tools created by scientists themselves. Scientific disinformation and damaging conspiracy theories are rife because of the internet that science created, the scientific demand for empirical evidence and statistical significance leads to data torturing and confirmation bias, and data mining is fuelled by the technological advances in Big Data and the development of ever-increasingly powerful computers. Using a wide range of entertaining examples, this fascinating book examines the impacts of society's growing distrust of science, and ultimately provides constructive suggestions for restoring the credibility of the scientific community.

Big Data, Data-Torturing, and the Assault on Science

Distrust

里森

We knew many researchersinduding ourselves who readily admitted to dropping dependent variables, conditions, or participants so as to achieve significance. Everyone knew it was wrong, but they thought it was wrong the way it's wrong to jaywalk. We [now know that it] was wrong the way it's wrong to rob a bank.

Responses are from Inflection AI's Pi model; accessed 2028-08-12.

AI is like a parrot that can say that it will meet you in an hour at the nearest Costco, but it won't be there.

The book is from this year, but this passage didn't age too well.

[Computers have “memory without reckoning”] […] [That is] why they have trouble answering simple questions like these correctly: Is it easier to field a ground ball if I close my eyes? Can I run faster if I hold my feet?

Fame, fortune, and funding await those who publish memorable research. Unfortunately, the lure of these rewards leads many to p-hack their way to semi-plausible, provocative results that will be reported widely. The achievements of scientists created the reward system; the scientific demand for rigor created the statistical significance requirement; and the scientific journals that disseminate research created the p-value launchpad for announcing eye-catching findings. The unintended consequence is that, even the best journals now publish rubbish that makes some question the sanity of scientists and the value of scientific research.

-hacker might begin by considering the data as a whole. Then look at males and females separately. Then look at different races. Then differentiate between children and adults; then between children, teenagers, and adults; then between children, teenagers, adults, and seniors. Then try different age cutoffs. Let the senior category be 65+; if that doesn't wok, try 55+, or 60+, or 70+, or 75+. Eventually, something clicks and a low p-value is discovered. As Ronald Coase, an Economics Nobel Laureate, cynically observed: If you torture data long enough, they will confess." Stephen M. Stigler, a renowned statistics pro- fessor added the punchline, "but confessions obtained under duress may not be admissible in the court of scientific opinion.”

1S used Or targeted das nat matcn companies with the people they want to lat matCh companies with the people they want to IS USed đas reach and influence. As they say, when you use a search engine, you are not the customer, you are the product.

The stories are outlandish and entertaining, yet they evidently resonate with many-perhaps because people struggling with personal difficulties find it eas- ier to blame their troubles on others than to take responsibility. For conspiracy believers, bad things happen because bad people make them happen. If I buy a stock and the price goes down, it is because insiders are manipulating the market. If I lose my job, it is because the Fed is manipulating the economy. If my life isn't working out the way I wanted, it is because the elite control the government for their own benefit.

 The Internet is the problem, not the solution. A former chief historian of NASA noted that The reality is, the internet has made it possible for people to say whatever the hell they like to a broader number of people than ever before.”

No other country in the world has the concentration of scientific skill, technical rowl. edge and productive capacity possessed by the US. It is the world's scientific superpower bar none. And yet this colossus of science utterly failed to bring its expertise succesfulv to bear on the policy and politics of the nation's response.

The irony here is that scientists created statistical tools that were intended to ensure the credibility of scientific research but have had the perverse effect of encouraging researchers to torture data—which makes their research untrustworthy and undermines the credibility of all scientific research.

The book expounds insightful yet somewhat repetitive ideas around sound empiricism and rationality; from crypto, crop circles, and conspiracy theories to p-hacking, HARKing, the replicability crisis and the fact that academic publishing is thoroughly broken. While it feels more like a loose collection of ideas than a coherent whole, it's still worth reading.

The author's views on machine intelligence have aged badly indeed—and this book is from 2023! He repeatedly asserts that machines lack true understanding without spending any time on how one might define it, seemingly appealing to a sort of human essentialism, which might be okay if he were making an argument about the moral value of human and non-human agents, but its unconvincing in a discussion of which jobs will be performed by machines in future.

His demonstrations of failure modes in LLMs are unconvincing even in the context in which the book was written and today cannot be replicated anymore (talk about the replicability crisis!).

All in all, worth reading for someone not already familiar with most of the concepts.

The book expounds insightful yet somewhat repetitive ideas around sound empiricism and rationality; from crypto, crop circles, and conspiracy theories to p-hacking, HARKing, the replicability crisis and the fact that academic publishing is thoroughly broken. While it feels more like a loose collection of ideas than a coherent whole, it's still worth reading.

The author's views on machine intelligence have aged badly indeed—and this book is from 2023! He repeatedly asserts that machines lack true understanding without spending any time on how one might define it, seemingly appealing to a sort of human essentialism, which might be okay if he were making an argument about the moral value of human and non-human agents, but its unconvincing in a discussion of which jobs will be performed by machines in future.

His demonstrations of failure modes in LLMs are unconvincing even in the context in which the book was written and today cannot be replicated anymore (talk about the replicability crisis!).

All in all, worth reading for someone not already familiar with most of the concepts.

Distrust Big Data, Data-Torturing, and the Assault on Science

Reviews

Highlights