Re: Where are the Linked Data Driven Smart Agents (Bots) ?

<Hugh rubs grey beard />

Ah, ’twas always thus, in every field I know.
To put it bluntly, that is because much research is about getting papers published, not about moving the field on.

<Old man’s meanderings - best avoided>
Functional programming (I used to be functional):

David Turner (he with a brain the size of a planet) said that his application of combinators to functional language implementation had “cost 10 years of wasted PhD students”.
This was because they had all striven to improve in tiny ways on the original implementation.
Of course most failed, as it had sprung perfectly-formed from his brain, and more importantly been perfectly-engineered by him into the system†.
But even when they succeeded, the increment didn't amount to a hill of beans.

And again, watching paper after paper purporting to improve some theoretical upper-bound on execution of some variant of the λ-calculus, which in fact could never be reached without immense execution overheads, was pretty depressing, in terms of wasted research time.

Of course, small increments are not always useless. In Operational Research they are bread and butter. But that is because it is a mature field with clear applications, and if you can improve a Search/Optimisation technique by a fraction of a percent, you might save millions on the billions cost of something.

In the first decades of a field, it is highly unlikely that incremental change will be significant in the long run, not least because entirely new methods and techniques will be discovered, making the base ones redundant, and rendering the increment moot.

†One of my favourite comments was in the garbage collector of David's C implementation of his SK-reduction machine: "Now follow everything on the C stack that looks like a pointer". :-)
</Old man’s meanderings - best avoided>

> On 8 Jul 2016, at 05:01, Ruben Verborgh <ruben.verborgh@ugent.be> wrote:
> 
> HI Krzysztof,
> 
>> this is all about finding the right balance
> 
> Definitely—but I have the feeling the balance
> is currently tipped very much to one side
> (and perhaps not the side that delivers
> the most urgent components for the SemWeb).
> 
>> as we also do not want to have tons of 'ideas' 
>> papers without any substantial content or proof of concept
> 
> Mere ideas would indeed not be sufficient;
> but even papers with substantial content
> and/or a proof of concept will have a difficult time
> getting accepted if there is no evaluation
> that satisfies the reviewers.
> (And, lacking a framework to evaluate evaluations,
> I see people typically choosing for things they know,
> hence why incremental research gets accepted easily.)
> 
> Best,
> 
> Ruben

Received on Friday, 8 July 2016 13:31:22 UTC