Especially given that the world's deadliest infectious disease (TB) already has a treatment. TB deaths today are caused by wealth inequality, not our insufficient medical knowledge.
It would be nice to see discussion of this in the article, but it focuses instead on the limitations of AI.
One of the main barriers to developing new medicines is that there millions and millions of potential compounds to test. You can go out to your back garden and take a soil sample, it'll have a ton of never before tested compounds, the multiple that by an entire planet worth of samples to test, and it overwhelms human capacity.
Testing these happens long before it ever goes to serious trials. Humans are very limited in how many results they can review, but if you can throw them to an AI and have it flag anything promising, then you've overcome a huge bottleneck.
This isn't magic or snake oil, and as the data often looks like microscope slides or tables of figures, it's well within the capabilities of current AI models.
Disease etiology spans both infectious and non-infectious. Infectious diseases are generally cured with antibacterial, antivirals, vaccines (what you call compounds).
Non-infectious (genetic, degenerative, metabolic, autoimmune) diseases typically involve therapies beyond “applying a compound”. We’ve seen breakthroughs like surgery (eg removing an appendix or a tumor). Gene therapies such as onasemnogene abeparvovec cures spinal muscular atrophy in infants. Certain leukemias or autoimmune diseases are cured via stem-cell transplants.
> (1) the huge majority of what people are calling AI these days is in fact machine learning. [...]
> (2) machine learning does not create any new knowledge. It rearranges information you have already obtained [...]
I do wish more folks who hype AI/ML as a means to revolutionize science would acknowledge point #2 and address it before their next clueless claim. Until ML can propose novel scientific hypotheses AND validate them, AI will continue to advance ONLY technology, as a tool does, but not science, as a theory does.
What is the difference between creating new knowledge and rearraigning existing knowledge? Everything is built on everything else after all, knowledge doesn't spontaneously generate. A lot of famous proofs, for example, are mostly existing information applied in new and unique ways. But i do agree, validation of claims is currently out of reach.
Pardon the newbie question, but isn't any "new" book/research a rearrangement of information you already obtained? Nobody invents anything starting from quarks.
Like with infinite monkeys, something will eventually turn up.
As it relates to biomedicine, the AI community is increasingly toeing the "snake oil" line.
Yes, machine learning/AI is powerful, but pie in the sky predictions such as the "end of disease" probably hurt more than help in the long run.
Especially given that the world's deadliest infectious disease (TB) already has a treatment. TB deaths today are caused by wealth inequality, not our insufficient medical knowledge.
It would be nice to see discussion of this in the article, but it focuses instead on the limitations of AI.
One of the main barriers to developing new medicines is that there millions and millions of potential compounds to test. You can go out to your back garden and take a soil sample, it'll have a ton of never before tested compounds, the multiple that by an entire planet worth of samples to test, and it overwhelms human capacity.
Testing these happens long before it ever goes to serious trials. Humans are very limited in how many results they can review, but if you can throw them to an AI and have it flag anything promising, then you've overcome a huge bottleneck.
This isn't magic or snake oil, and as the data often looks like microscope slides or tables of figures, it's well within the capabilities of current AI models.
Not all diseases are cured via compounds.
Disease etiology spans both infectious and non-infectious. Infectious diseases are generally cured with antibacterial, antivirals, vaccines (what you call compounds).
Non-infectious (genetic, degenerative, metabolic, autoimmune) diseases typically involve therapies beyond “applying a compound”. We’ve seen breakthroughs like surgery (eg removing an appendix or a tumor). Gene therapies such as onasemnogene abeparvovec cures spinal muscular atrophy in infants. Certain leukemias or autoimmune diseases are cured via stem-cell transplants.
Works well enough to attract capital, though.
And to dupe the public
> (1) the huge majority of what people are calling AI these days is in fact machine learning. [...]
> (2) machine learning does not create any new knowledge. It rearranges information you have already obtained [...]
I do wish more folks who hype AI/ML as a means to revolutionize science would acknowledge point #2 and address it before their next clueless claim. Until ML can propose novel scientific hypotheses AND validate them, AI will continue to advance ONLY technology, as a tool does, but not science, as a theory does.
What is the difference between creating new knowledge and rearraigning existing knowledge? Everything is built on everything else after all, knowledge doesn't spontaneously generate. A lot of famous proofs, for example, are mostly existing information applied in new and unique ways. But i do agree, validation of claims is currently out of reach.
Pardon the newbie question, but isn't any "new" book/research a rearrangement of information you already obtained? Nobody invents anything starting from quarks.
Like with infinite monkeys, something will eventually turn up.
[dead]