I think your provocation is based on an example that is not apples to apples.
So let me throw a provocation back.
An AI model for sport rules has a binary outcome: does the action fall into one case or another (in or out, for example). Training of the model is fairly simple because the variables are limited.
A medical model is most likely trained to highlight anomalies based on a much wider range of variables and requires ongoing interaction and validation by the doctor (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9955430/).
The first provides an answer, the second accelerates a process.
Which makes the provocation on AI as a contributor vs. a sostitution a bit of a moot point.
Complete replacement of the human element is something that has a data availability problem, a data cataloguing problem, a tokenization problem, a model prioritization problem (LLMs? Diffusion? They all have pros and cons) and a tech availability problem (GPUs).
More than anything an interoperability problem. Private systems (the larger part, since public organization don't have the funding to develop AI systems that are just as powerful) won't be interoperative.
We will see an increasing closure of data as IP: private companies won't share their data with others and so a complete data lake of medical pathologies will most likely not be available - unless some international organization emerges.
Until all these elements - which are ultimately economical and not technological - will be solved, humans will still have a role.
I think your provocation is based on an example that is not apples to apples.
So let me throw a provocation back.
An AI model for sport rules has a binary outcome: does the action fall into one case or another (in or out, for example). Training of the model is fairly simple because the variables are limited.
A medical model is most likely trained to highlight anomalies based on a much wider range of variables and requires ongoing interaction and validation by the doctor (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9955430/).
The first provides an answer, the second accelerates a process.
Which makes the provocation on AI as a contributor vs. a sostitution a bit of a moot point.
Complete replacement of the human element is something that has a data availability problem, a data cataloguing problem, a tokenization problem, a model prioritization problem (LLMs? Diffusion? They all have pros and cons) and a tech availability problem (GPUs).
More than anything an interoperability problem. Private systems (the larger part, since public organization don't have the funding to develop AI systems that are just as powerful) won't be interoperative.
We will see an increasing closure of data as IP: private companies won't share their data with others and so a complete data lake of medical pathologies will most likely not be available - unless some international organization emerges.
Until all these elements - which are ultimately economical and not technological - will be solved, humans will still have a role.