4 ms·

Pharmacology is an interesting example where better technology and scientific understanding have made things worse than earlier, low-tech methods ("inject plant

Pharmacology is an interesting example where better technology and scientific understanding have made things worse than earlier, low-tech methods ("inject plant extracts into animals and see what happens"). The number of new drugs discovered per dollar of research has been dropping since 1960, and obvious explanations (like "the easy ones have already been found") turn out not to explain the phenomenon. [http://www.nature.com/nrd/journal/v11/n3/fig_tab/nrd3681_F1.html http://www.nature.com/nrd/journal/v11/n3/fig_tab/nrd3681_F1....] This is something we should try to understand better, since it goes against the intuition that technology is an unalloyed good in scientific research. I applaud the money they're spending, but the level of technophilia in the announcement gives me pause.

notahacker 10y ago
I can't speak for the past, but I think the big deal for the testing of the future will be shifting from "forty percent of the test subjects responded positively to the drug; none of the side effects were that bad" to "80% of test subjects carrying thought-to-be-marginally-relevant gene Y showed positive responses to the drug but none carrying gene Yvariant; those also carrying not-suspected-of-be-relevant gene X exhibited much stronger side effects" That should make initial testing a lot more expensive, but in the long run launches of new drugs less risky.
- jonknee 10y ago
  And even farther in the long run make it possible to virtually test therapies against arbitrary genetic information.
- theptip 10y ago
  I've been thinking along these lines recently; we currently design treatment plans using statistical methods, i.e. averaging effects across 'representative populations'. But imagine if you took your car to the garage and they said "90% of cars showing this symptom respond well to this treatment". When we truly understand how these biological systems work (which could be more than 100 years from now) we'll be able to measure what is different from the individual's baseline, rather than extrapolating based population averages.
tangue 10y ago
If this initiative funds studies on non-profitable treatments (fasting, nutrition) or counter balance industry backed lobbying, they will earn my respect. But if its to dump money on AI, yes its just sad. Google and Facebook AI can't fight the most basic spam. Seeing the same tools curing cancer even in the long run is a little bit far fetched ...
maxerickson 10y ago
Apparently that paper coined Eroom's Law: https://en.wikipedia.org/wiki/Eroom%27s_Law https://en.wikipedia.org/wiki/Eroom%27s_Law the 'better than the Beatles' problem is at least superficially compelling. When Benadryl was discovered, the bar for allergy drugs was pretty low. Today you have to improve on drugs that have several advantages over Benadryl (and cost ~$0.50 a day OTC in the US).
dredmorbius 10y ago
Returns on R&D have been falling across a wide range of disciplines for a long time. Joseph Tainter references work by several other authors, looking at patents and Nobel Prize awards. J. Doyne Farmer and the late John C. Holland both work in the are of technological innovation as well. I'm not convinced technology and understanding are the causal factors here.

notahacker 10y ago

I can't speak for the past, but I think the big deal for the testing of the future will be shifting from "forty percent of the test subjects responded positively to the drug; none of the side effects were that bad" to "80% of test subjects carrying thought-to-be-marginally-relevant gene Y showed positive responses to the drug but none carrying gene Yvariant; those also carrying not-suspected-of-be-relevant gene X exhibited much stronger side effects" That should make initial testing a lot more expensive, but in the long run launches of new drugs less risky.

jonknee 10y ago
And even farther in the long run make it possible to virtually test therapies against arbitrary genetic information.
theptip 10y ago
I've been thinking along these lines recently; we currently design treatment plans using statistical methods, i.e. averaging effects across 'representative populations'. But imagine if you took your car to the garage and they said "90% of cars showing this symptom respond well to this treatment". When we truly understand how these biological systems work (which could be more than 100 years from now) we'll be able to measure what is different from the individual's baseline, rather than extrapolating based population averages.

tangue 10y ago

If this initiative funds studies on non-profitable treatments (fasting, nutrition) or counter balance industry backed lobbying, they will earn my respect. But if its to dump money on AI, yes its just sad. Google and Facebook AI can't fight the most basic spam. Seeing the same tools curing cancer even in the long run is a little bit far fetched ...

maxerickson 10y ago

Apparently that paper coined Eroom's Law: https://en.wikipedia.org/wiki/Eroom%27s_Law https://en.wikipedia.org/wiki/Eroom%27s_Law the 'better than the Beatles' problem is at least superficially compelling. When Benadryl was discovered, the bar for allergy drugs was pretty low. Today you have to improve on drugs that have several advantages over Benadryl (and cost ~$0.50 a day OTC in the US).

dredmorbius 10y ago

Returns on R&D have been falling across a wide range of disciplines for a long time. Joseph Tainter references work by several other authors, looking at patents and Nobel Prize awards. J. Doyne Farmer and the late John C. Holland both work in the are of technological innovation as well. I'm not convinced technology and understanding are the causal factors here.