2021/03/22
What Data Can't Do
"if more data isn’t always the answer, maybe we need instead to reassess our relationship with predictions—to accept that there are inevitable limits on what numbers can offer, and to stop expecting mathematical models on their own to carry us through times of uncertainty." Full article at the New Yorker.
Labels: #data, #Modeling, #Numbers, #Prediction
2019/05/01
AI Uses Images and Omics to Decode Cancer
Machine learning can analyze photographs of cancer, tumor pathology slides, and genomes. Now, scientists are poised to integrate that information into cancer uber-models. Full article @ The Scientist.
Labels: #AI, #Cancer, #data, #MachineIntelligence, #MachineLearning
2018/06/05
Social biases in AI
The Hippocratic oath for data scientists would be a good start, though I am sure greater regulation is needed. There needs to be government agencies (NIST?) who query commercial and government AI systems with blackbox system identification techniques. They would statistically test against non-biased response distributions; If public systems fail hypothesis testing (e.g. chi-square) against fair distributions, they should be further investigated, their algorithms subpoenaed, and prosecuted if need be---same for data scientists shown to fail any future oath. See book reviews @ The New York Review of Books. (Thank you Thiago for link)
Labels: #AI, #data, #Ethics, #Racsim
2017/12/04
On data dredging
From A Failure to Heal By SIDDHARTHA MUKHERJEE. Thank you to Thiago Carvalho for the link.
"Perhaps the most stinging reminder of these pitfalls comes from a timeless paper published by the statistician Richard Peto. In 1988, Peto and colleagues had finished an enormous randomized trial on 17,000 patients that proved the benefit of aspirin after a heart attack. The Lancet agreed to publish the data, but with a catch: The editors wanted to determine which patients had benefited the most. Older or younger subjects? Men or women?
Peto, a statistical rigorist, refused — such analyses would inevitably lead to artifactual conclusions — but the editors persisted, declining to advance the paper otherwise. Peto sent the paper back, but with a prank buried inside. The clinical subgroups were there, as requested — but he had inserted an additional one: “The patients were subdivided into 12 ... groups according to their medieval astrological birth signs.” When the tongue-in-cheek zodiac subgroups were analyzed, Geminis and Libras were found to have no benefit from aspirin, but the drug “produced halving of risk if you were born under Capricorn.” Peto now insisted that the “astrological subgroups” also be included in the paper — in part to serve as a moral lesson for posterity. I’ve often thought of Peto’s paper as required reading for every medical student".
"Perhaps the most stinging reminder of these pitfalls comes from a timeless paper published by the statistician Richard Peto. In 1988, Peto and colleagues had finished an enormous randomized trial on 17,000 patients that proved the benefit of aspirin after a heart attack. The Lancet agreed to publish the data, but with a catch: The editors wanted to determine which patients had benefited the most. Older or younger subjects? Men or women?
Peto, a statistical rigorist, refused — such analyses would inevitably lead to artifactual conclusions — but the editors persisted, declining to advance the paper otherwise. Peto sent the paper back, but with a prank buried inside. The clinical subgroups were there, as requested — but he had inserted an additional one: “The patients were subdivided into 12 ... groups according to their medieval astrological birth signs.” When the tongue-in-cheek zodiac subgroups were analyzed, Geminis and Libras were found to have no benefit from aspirin, but the drug “produced halving of risk if you were born under Capricorn.” Peto now insisted that the “astrological subgroups” also be included in the paper — in part to serve as a moral lesson for posterity. I’ve often thought of Peto’s paper as required reading for every medical student".
Labels: #data, #Statistics
