Mathematica man brews AI Google Killer
theregister.co.uk
theregister.co.uk
"AI Google Killer": The kiss of death and a box of poisoned chocolates.
I get the impression that tech journalists purposely overhype new ideas just so 6 months later they can trash the very thing they overhyped for not living up to its expectations. Job security for tech journalists.
i went to a lecture wolfram gave at my university when he was touring for NKS. he would answer a question from a CS professor until he went over the CS professor's head, then a biologist until he went over his head, then a physicist until he went over her head. flippantly passing off that level of genius by calling him "the lovable George Costanza of the mathematics community" says something. i understand wolfram has a huge ego, but that does not belittle his accomplishments.
just because your work "ruffled some feathers in the scientific community" currently, does not mean that it won't be considered ground breaking in 100 years. to be clear, i'm referring to how scientists become attached to their life's work and thus biased, as opposed to anything specific to wolfram or NKS and cellular automata.
finally, Alpha will be judged when it comes out. if it's worthwhile great, if not fine, but there's not much point in trying to persuade people of something before it is out.
As brilliant as he is, it seems Wolfram is rather guilty of disregarding a lot of the work already done in the fields he works in - coming up with his own jargon for parallel (published) ideas, and hyping up his unreleased work instead of publishing and just letting it speak for itself (like alpha).
Research is more about advancing the understanding of science in the community than it is about doing something brilliant on your own (because as brilliant as it is, if one can't communicate it, it'll die with the individual), and so Wolfram deserves quite a lot more criticism than he gets.
Is this guy serious ?
Yes, machine learning+semantic web "sounds like" like oil. But neither of those technologies are very relevant to alpha.
Yes, prolog had an algorithm for making deductions. But the point of alpha is not to make logical deductions, but to run computational models and analysis based on a vast dataset and human input.
The idea that alpha "answers your question" is something that wolfram implicitly put out there in his blog post, but often its more like a computational version of wikipedia.. you enter an entity and visualization-heavy page is generated to explain and explore it.
Hand-curating data and algorithms may sound silly to someone who hasn't built something as big as Mathematica, but the world of important/relevant data and algorithms is ultimately finite in size (if very large and complicated).
All joking aside, I'm really excited to begin playing with this and exploring the possibilities of a live web inference engine. I'm anxious to see what kind of API/developer tools will be provided, and (since it's Mathematica) whether there will be any cost involved. Has anyone seen this discussed anywhere?
Seriously, in the 1950s a computer was an "electronic brain" when it could add up numbers and get the right total. In the 1980s it was when it could play chess as well as a novice (basic minimax trees). It's always a moving target.
This was tried before, in the "expert systems" of the 1980s. Prolog and various 'production system' languages (e.g. OPS5) would decouple facts/rules from the execution engine. Then "just add" domain knowledge ...
As far as I can tell, what killed that was the unfeasibility of debugging. One bad item (or the lack of a critical true one) could poison the well, and good luck tracking it down.
The care and feeding of proper 'fact databases' remains expensive, and Google doesn't seem disposed to go there, apart from unit and currency conversions. (Add a 'hmm' for automated translation.)
So, this leaves two ideas for 'Alpha' as a product: a) they believe the database grooming expense can be profitable on a global web scale; b) they automated enough of the database grooming to make it profitable.