4 ms·
Every time I see an article which talks about "Watson" as an individual solution to a problem I find myself wondering what the people paying for it all think th
by hvindin 11y ago
Every time I see an article which talks about "Watson" as an individual solution to a problem I find myself wondering what the people paying for it all think that "Watson" is?
Realistically its like 10+ racks filled with POWER systems loaded up with terabytes of RAM running SUSE Linux using Apache Hadoop for distributed computing, then some IBM Proprietary software solutions + a few other open bits and pieces (ie Apache UIMA) cobbled together to get something which behaves as though it understands natural language.
Honestly if you put most data-centric systems on a platform with dozens, if not hundreds, of POWER CPUs with 8-16 cores a piece and 4 threads/core and then put the entire data store into RAM so you could get at it quickly, I suspect that you could fool most people into thinking there was intelligence behind it, not just raw computing power.
- sp332 11y agoI don't think it's that odd. People talk about Siri and Cortana the same way.
- hvindin 11y agoBut Siri and Cortana are sold to consumers, I'm happy with my day to day life being made up of over-simplifications because, generally, if those things don't work it doesn't really matter. However, in large organizations, dealing with fairly critical things, it just seems like over-simplifying the solution is a recipe for disaster. That's not to say that Watson-type solutions have no place in the world. The recent stuff re: health and diagnostics - awesome. It's complexity you can't fix because people are living organisms so you just need to drag in the data and essentially evaluate your best guess. But where companies are using Watson to mask their needlessly complex processes or, as an example, banks are using Watson to decipher their needlessly complex loan schemes and agreement terms, those are problems that will only get harder to solve if you don't deal with the cause of the complexity (something within the scope of their control) as soon as possible. That being said, it's probably good business for IBM because if your multi-billion dollar investment in IBM hardware is insufficient to deal with the mess of policy that has grown unchecked over the years, masked by a more friendly interface, then you can always buy more servers and more hosting...
- ProAm 11y agoNLP is perfect for understanding purchasing requirements, rules, laws, etc.... You right, watson is just a task specific tuned supercomputer, but I see this as a good fit.
- rspeer 11y agoNLP pretty much has to be task-specific, but the task-specific NLP that will solve the Pentagon's particular problem almost certainly doesn't exist until someone writes it. When you describe "Watson" as a task-specific supercomputer, if you mean the Watson that played Jeopardy, that's certainly not what they're going to be using. "Watson" here means hiring IBM consultants to write them some software that runs on IBM hardware.
- apaprocki 11y agoIBM people would be the first to tell you that the mega-rack "Watson" is not the IBM Watson * suite of various products. They said it just needs a few U of rack space.
- karmacondon 11y agoIt seems like IBM is using the Watson name to re-brand itself in an attempt to turn their business around. So that "using Watson" is now synonymous with "hiring IBM consultants". Ten years ago this headline would have read, "Pentagon hires IBM to help optimize procurement". But the Watson brand has become popular with the media, in large part due to the exposure from the Jeopardy exhibition (which is seeming like a better and better investment in terms of marketing returns). I'm not sure if the re-branding will be enough to really change IBM's fortunes, but does seem to be helping them in the short term. As companies get better at using open source tools and expert knowledge to leverage their own data, it's going to get harder for tech consulting companies of all types.
- yo-code-sucks 11y agohttps://en.m.wikipedia.org/wiki/UIMA https://en.m.wikipedia.org/wiki/UIMA