Is AI leading to a reproducibility crisis in science?
https://www.nature.com/articles/d41586-023-03817-6
EXCERPT: ...“These examples might be amusing”, Shamir says — but in biomedicine, misclassification could be a matter of life and death. “The problem is extremely common — a lot more common than most of my colleagues would want to believe.”
A separate review in 2021 examined 62 studies using machine learning to diagnose COVID-19 from chest X-rays or computed tomography scans; it concluded that none of the AI models was clinically useful, because of methodological flaws or biases in image data sets. The errors that Shamir and Dhar found are just some of the ways in which machine learning can give rise to misleading claims in research.
Computer scientists Sayash Kapoor and Arvind Narayanan at Princeton University in New Jersey reported earlier this year that the problem of data leakage (when there is insufficient separation between the data used to train an AI system and those used to test it) has caused reproducibility issues in 17 fields that they examined, affecting hundreds of papers. They argue that naive use of AI is leading to a reproducibility crisis... (MORE - missing details)
IBM releases first-ever 1,000-qubit quantum chip
https://www.nature.com/articles/d41586-023-03854-1
INTRO: IBM has unveiled the first quantum computer with more than 1,000 qubits — the equivalent of the digital bits in an ordinary computer. But the company says it will now shift gears and focus on making its machines more error-resistant rather than larger.
For years, IBM has been following a quantum-computing road map that roughly doubled the number of qubits every year. The chip unveiled on 4 December, called Condor, has 1,121 superconducting qubits arranged in a honeycomb pattern. It follows on from its other record-setting, bird-named machines, including a 127-qubit chip in 2021 and a 433-qubit one last year.
Quantum computers promise to perform certain computations that are beyond the reach of classical computers. They will do so by exploiting uniquely quantum phenomena such as entanglement and superposition, which allow multiple qubits to exist in multiple collective states at once.
But these quantum states are also notoriously fickle, and prone to error. Physicists have tried to get around this by coaxing several physical qubits — each encoded in a superconducting circuit, say, or an individual ion — to work together to represent one qubit of information, or ‘logical qubit’.
As part of its new tack, the company also unveiled a chip called Heron that has 133 qubits, but with a record-low error rate, three times lower than that of its previous quantum processor... (MORE - details)
https://www.nature.com/articles/d41586-023-03817-6
EXCERPT: ...“These examples might be amusing”, Shamir says — but in biomedicine, misclassification could be a matter of life and death. “The problem is extremely common — a lot more common than most of my colleagues would want to believe.”
A separate review in 2021 examined 62 studies using machine learning to diagnose COVID-19 from chest X-rays or computed tomography scans; it concluded that none of the AI models was clinically useful, because of methodological flaws or biases in image data sets. The errors that Shamir and Dhar found are just some of the ways in which machine learning can give rise to misleading claims in research.
Computer scientists Sayash Kapoor and Arvind Narayanan at Princeton University in New Jersey reported earlier this year that the problem of data leakage (when there is insufficient separation between the data used to train an AI system and those used to test it) has caused reproducibility issues in 17 fields that they examined, affecting hundreds of papers. They argue that naive use of AI is leading to a reproducibility crisis... (MORE - missing details)
IBM releases first-ever 1,000-qubit quantum chip
https://www.nature.com/articles/d41586-023-03854-1
INTRO: IBM has unveiled the first quantum computer with more than 1,000 qubits — the equivalent of the digital bits in an ordinary computer. But the company says it will now shift gears and focus on making its machines more error-resistant rather than larger.
For years, IBM has been following a quantum-computing road map that roughly doubled the number of qubits every year. The chip unveiled on 4 December, called Condor, has 1,121 superconducting qubits arranged in a honeycomb pattern. It follows on from its other record-setting, bird-named machines, including a 127-qubit chip in 2021 and a 433-qubit one last year.
Quantum computers promise to perform certain computations that are beyond the reach of classical computers. They will do so by exploiting uniquely quantum phenomena such as entanglement and superposition, which allow multiple qubits to exist in multiple collective states at once.
But these quantum states are also notoriously fickle, and prone to error. Physicists have tried to get around this by coaxing several physical qubits — each encoded in a superconducting circuit, say, or an individual ion — to work together to represent one qubit of information, or ‘logical qubit’.
As part of its new tack, the company also unveiled a chip called Heron that has 133 qubits, but with a record-low error rate, three times lower than that of its previous quantum processor... (MORE - details)