soylent green was made of people, too

Call me a humanist luddite and beat me over the head with my liberal arts degree, but I rather enjoyed Will Davies' rebuttal Chris Anderson's The End of Theory, titled why Google can't replace theory. A lengthy snippet:

I would suggest that Anderson is extending the Chicago School project of selectively dismantling the bases for authoritative knowledge claims. Chicago economics renders social knowledge so fragile and polluted with self-interest, that it becomes impossible to produce a better model for society than that of the unimpeded market. Again, there is a sleight of hand at work here - man's epistemological condition leads not just contingently but logically to the technological solution of the market.

(snip)

For Anderson it is not the market that comes to our rescue, but the world wide web. What the market can do for material resources, the web can do for knowledge. In each case, we are relieved of the political and theoretical burden of trying to produce a good, coherent model for society, and put ourselves in the hands of an ignorant, amoral mechanism - price in the case of material resources, algorithm in the case of immaterial ones.

Davies may be overreacting...but when Anderson writes argues that "science can advance even without coherent models, unified theories, or really any mechanistic explanation at all," the logical retort is Sure, science may advance that way...but can society?

Comments

[Bill Seitz]

I'd say the first problem is confusing "social science" with science.

Since none of Anderson's examples involve "social science", he should have left it out of his piece entirely.

[Michael Sippey]

@bill -- that's a good point; yeah, he probably should have left out the social science bits in his piece.

The thing that resonated with me about Davies' piece was that even if we use MapReduce as a replacement for the scientific method, humans still have a need for a mental model about How Things Work. Also, call me a deconstructionist at heart (after all, it is Derrida's birthday today), but even algorithms are designed with a point of view. This may be overly simplistic of me, but relying on machines to find the correlations just pushes the ghost in the machine back by a few degrees.

[Bill Seitz]

There have been 3 awards made so far.

http://www.nsf.gov/awardsearch/progSearch.do?SearchType=progSearch&page=2&QueryText=&ProgOrganization=CSE&ProgOfficer=&ProgEleCode=7782&BooleanElement=true&ProgRefCode=&BooleanRef=true&ProgProgram=&ProgFoaCode=&RestrictActive=on&Search=Search#results

[Bill Seitz]

Theories are generally born from observation. So I see this tech as a way of providing new observation material.

But I tend to agree that just non-interpreted correlations are shaky things to depend on. But, again, Anderson's piece is such puffery that it's hard to find a coherent point of view to even argue with.

[Jamie Flournoy]

Davies should know better than to take anything printed in Wired seriously. It's the new Omni magazine, with some Fast Company and lad-gadget-mag seasoning sprinkled in.

Norvig says "increasingly you can succeed without them". Fair enough. Anderson says step right up, science is "becoming obsolete". Sounds like Norvig is being reasonable but Anderson is heralding an age when researchers simply do every possible experiment (!) and let the results stand up and identify themselves. No more will we have to think about where to look for data. Just look everywhere!

I wonder who will fund these research projects to just measure everything with no particular goal. I'd love to read that grant application...

[charlie]

I like the Wired = the new Omni.

There are times where scientists have done what we call 'brute force' experiments - experiments where you just do a lot of one type of experiment to create a ton of data, with little or no hypothesis to base it on. Such 'brain-dead' (in a good way) experiments are indeed the 'throw a ton of observations and see what sticks'.

A classic example I remember is the discovery of body patterning genes in drosophila. Such mindless search for mutants was also done in zebra fish, in deference to the work done in drosophila. It's more phenomenology than scientific method, we used to say, if you get what I mean.*

But, in the end, science must advance, guided by theories and testable hypothesis and human-understandable models. If all we do is let the computer create a (black-box) simulations (which is what Anderson is really implying), then science _won't_ advance. (but this might be just my deep indoctrination in the scientific method clouding my mind)

Oh, and sorry, I might be tired (it's really late at the moment), but Davies makes absolutely no sense to me. :-)

And, Sippey, this post has really put you up much higher on erudite I admire.

*And at the same time, there are very deeply thought out hypotheses that require a galaxy of data for a single concept to be shown, as exemplified by the Hadron Collider in Geneva. On a smaller scale, my experiments had thousands of data points fit to a model that reduced them to two numbers that revealed complex interactions. I think Anderson is just bedazzled by the fancy googleplex computers and visualizations of large data sets. Heh.