Human beings are great at solving problems quickly when appropriately incentivized. Everything else will fall into place (data quality, AI transparency etc.), once economic incentives are aligned. If economic incentives are not aligned, no amount of policy or tech sledgehammer is going to solve this.
Art_Vandelay - 3 years ago
Agree with Lynn and I would add better curation of underlying data to avoid potentially inherent biases and more transparency in the data and algorithms to improve trust, understanding, and feedback from the user community.
The major problem with “AI” based on EMR data is poor data quality. This is a pervasive problem hampering clinical understanding and communication and is the major reason why clinical decision support hasn’t delivered what automation promised. Before worrying with what’s on your pick list there needs to be a major push towards better data- the right data if you will. FHIR broadly implemented would help. Radiology image data sets offer a model for other clinical descriptive data. Computers are not artificially intelligent - they need more and better human help.
Human beings are great at solving problems quickly when appropriately incentivized. Everything else will fall into place (data quality, AI transparency etc.), once economic incentives are aligned. If economic incentives are not aligned, no amount of policy or tech sledgehammer is going to solve this.
Agree with Lynn and I would add better curation of underlying data to avoid potentially inherent biases and more transparency in the data and algorithms to improve trust, understanding, and feedback from the user community.
What Lynn Witherspoon said
The major problem with “AI” based on EMR data is poor data quality. This is a pervasive problem hampering clinical understanding and communication and is the major reason why clinical decision support hasn’t delivered what automation promised. Before worrying with what’s on your pick list there needs to be a major push towards better data- the right data if you will. FHIR broadly implemented would help. Radiology image data sets offer a model for other clinical descriptive data. Computers are not artificially intelligent - they need more and better human help.