Comments on: Watson goes to college: How the world’s smartest PC will revolutionize AI

By: jahendler

jahendler — Tue, 05 Mar 2013 21:39:14 +0000

thanks Steve, high compliment coming from you

By: Steve Minton

Steve Minton — Tue, 05 Mar 2013 18:25:58 +0000

OK, I can’t write. I need to go back to school too. That should have been “Nice article”.

By: Steve Minton

Steve Minton — Tue, 05 Mar 2013 18:02:57 +0000

Nice written article. Wonderful that RPI students will have this opportunity.

By: jahendler

jahendler — Mon, 04 Mar 2013 19:13:55 +0000

I agree – the author doesn’t get to title his column. I wanted it to be:

Answer: It’s why the World’s Smartest Computer is going back to school.

By: therealmattsimpson

therealmattsimpson — Mon, 04 Mar 2013 00:05:09 +0000

I don’t think you can call Watson a PC (Personal Computer).

By: Steve Ardire

Steve Ardire — Sun, 03 Mar 2013 20:16:34 +0000

Hi Jim – agree the key is how NLP, hypothesis generation, evidence-based learning is used in combination w/the memory store. Also not saying there’s many competitive solutions ( because this is heady stuff ) just a select few that hold some promise ;)

By: Dan McCreary

Dan McCreary — Sun, 03 Mar 2013 19:46:25 +0000

Makes me wish I was back in grad school working for Jim! So many opportunities for innovation! Jim’s work at RPI converting the information at data.gov to RDF seems like a great starting point. I hope that some day the Apache UIMA frameworks benefit from these tools.

By: jahendler

jahendler — Sun, 03 Mar 2013 19:31:48 +0000

Steve -
While Watson does use those capabilities, saying it is just a collection of these is like saying that Deep Blue was just a search program. It’s the complexity of the task and level of performance that Watson achieved that showed something significant had happened. After examining the behaviors, it seems to me that the novel aspect was, as I wrote here, how the large memory, accessed appropriately, made the system work. It used learning primarily for adjusting weights on the answers of components, but those components worked to a large degree off the memory, So it is not that the three things you point out aren’t important, but rather it’s how they’re used in combination w/the store. It would be interesting to see a competitive solution that could function without the memory store – I certainly haven’t seen anything competitive to date

By: Steve Ardire

Steve Ardire — Sat, 02 Mar 2013 21:49:11 +0000

Hi Jim,

This is very exciting stuff so good for you and your team at RPI.

You say most competitive solutions “understand” text using some kind of deep “knowledge representation” or use complex statistical models……Yes agree

Then you say Watson uses a lot of memory and ‘clever ways’ of pulling texts from that memory and provide elaborate descriptions and examples especially for healthcare.

However, I’m curious why couldn’t you simply say Watson focuses on three key capabilities
1) NLP
2) Hypothesis generation
3) Evidence-based learning

This is interesting but these 3 capabilities are not unique i.e. there’s competitive solutions that can do pretty much do equivalent with same and different techniques.