I don't know why people overlook the progress IBM seems to have made in the quantum computing world. Perhaps not as impressive as Google's progress but impressive none the less.
For me personally, it was because even from the inside at IBM, you wouldn't hear much about what the quantum folks are doing if you were doing general tech. They really are in their own little world. I don't think they were even in the whole-IBM Slack team, though they might just have been security-compartmentalized with ACLs.
Was Watson just smoke and mirrors? Have there been other systems in the NLP space that eclipsed it? Genuinely asking because I don't know much background behind the story of Watson's rise and fall.
Watson didn't rise or fall, it was the centerpiece in IBM's marketing campaign, which worked for a while before they moved on. It never got beyond or matched the state of the art in anything, that wasn't the point, it got the name "Watson" on TV.
IBM wanted to prove that they employed smart people, so they hired some smart people and had them write a computer program to play Jeopardy. Did it matter that it had nothing to do with anything they were selling? Perhaps only IBM has those figures.
OK fine but there was _some_ non trivial technology behind the system that won the Jeopardy game. What I'm asking is whether what seemed like the state of the art NLP system at the time got eclipsed by newer and better systems or whether the whole thing was never really a state of the art NLP system to begin with.
The problem with Watson is that they don't have a business case. IBM sent salespeople to big customers with big problems and tried to find things to fix. In some cases, they did. But most of the time, the insurance companies, government agencies, etc they do business with scratched their heads and didn't do anything. Learning new things about their data might be seen as a threat to whatever enterprise bullshit they do!
The problem is you have companies like Google, Microsoft, Amazon, Apple, Facebook, etc have problems that this tech solves. It's easier to come up with a product with a problem that you understand. I can ask Google Photos or Siri to show me pictures of my dog in the snow in 2015, and they do. So I give Google & Apple money to store my crap. Google and Facebook use AI with all of the data they hoover up to peddle products to me. My grandparents get ads for depends, I get ads for drones, Google and Facebook make $.
Now, companies like Amazon, Microsoft and Google can go to companies that were prospected by IBM with solutions. Microsoft is minting money with ATP, because enterprise security teams suck. Amazon is selling creepy facial recognition to people, because people see it on TV, have a Ring doorbell, and want the capability. Google is selling GIS solutions, etc based on work done on maps.
> Learning new things about their data might be seen as a threat to whatever enterprise bullshit they do!
No, the problem was that Watson would be unable to help you learn anything new about enterprise data. IBM didn't even have a plausible, non-trivial proof of concept to trot out six years ago.
State of the art NLP system isn't a very meaningful term. You can't really say that Watson was "better" than Google's NLP systems at the time because they were solving different problems.
In general, what they did was not trivial, but it also wasn't revolutionary. Some of it was novel, but novel in the sense that it applied specifically to the problem they were trying to solve. It had little impact on the technology behind what IBM eventually tried to sell as Watson.
Anybody with a bucket of money could have built the same thing at the time. The impressive part was IBM figuring out that it would be worth spending a bucket of money on.
I think it was fantastic, modern NLP could do much more (if put together with the cleverness of the Watson pipeline) and it's a pot of gold if IBM ever stop trying to apply it in healthcare and apply their brains to how it could work in customer service (for example).
Note : I am aware that there are "Watson" products that claim to be this - but they aren't because they are MBA's ideas of what the best route to selling crap to the unwary. If IBM had appointed someone with a clue and given them 10x the R&D budget for the gameshow to deliver a decent product I reckon they'd have got (at least) 100x. But... oh no.. promise 30x for 1x and get f-all.x^2
Cliches aside, I think you are right about the poor decisionmaking: IBM decided to chase the biggest, most bureaucratic market, healthcare, rather than pick a market that would actually be somewhat receptive to black-box services.
It was an amazing demo, as far as marketing was concerned. Now whether the technology behind it was as revolutionary as they made it sound, that's probably more controversial.
It was certainly a stye in the eye for Google (“organize the world's information and make it universally accessible and useful”), whether facbricated by marketing/advertising or not.
But it wasn't really putting itself out there as a Google competitor. IBM wasn't saying, "buy this to make a better search engine", so I never once drew the Watson vs Google comparison when it all went down in 2011. Watson seemed to want to market itself as tech to drive expert systems instead (though it never really panned out) and now 9 years later, Google Assistant can answer questions of astou ding complexity, so whatever research Google was embarking on in 2011 has since born much more fruit anyway.
Watson (the Jeopardy playing machine) really was an amazing demo for it's time (2010) and before the rise of neural networks in NLP it was one of the best demonstrations of a semi-practical system.
The "This is Watson" special edition journal[1] taught me more about traditional NLP pipelines than just about anything I've read before or since.
reply