Yahoo Finance published an article last week on a new capability of IBM Watson, the cognitive computing platform. The article, by Rick Newman, addresses Watson’s newly unveiled skill of autonomously constructing logical arguments for and against any given position. Watson adds these new debating skills onto its already game-show-tested question / answering prowess.
In a demo that you can watch for yourself by going to this Popsci article, Watson is asked to construct arguments for and against a law banning the sale of violent video games to minors. In the demo, Watson takes less than ten seconds to return three well-constructed arguments on both the pro and con side of the proposed law. Watson describes the process steps along the way to determining its statements. Here are the steps:
- Scan 10K Wikipedia articles
- Return the 10 most relevant articles
- Scan all the sentences of the 10 selected articles
- Detect sentences with candidate claims
- Identify orders of candidate claims
- Assess pro and con polarity of candidate claims
- Construct demo speech with top claim predictions
Watson does not fabricate the pro and con arguments completely out of thin air. It identifies arguments that humans have previously made and documented. Still, Watson’s debating ability is an impressive new talent for the cognitive computer. Watson has to have a basic understanding of the position that it is being asked to both defend and argue against. Saying that Watson comprehends the meaning is probably going too far, but it has to decipher the key concepts so that it can locate documentation sources that contain supporting and dissenting opinions.
It seems that the same abilities could be used to fuel a more intelligent conversational virtual agent. What if you were conversing with a virtual assistant and it said “I don’t like hotdogs” and you asked it “Why don’t you like hotdogs?” Couldn’t an assistant enabled with Watson’s debater technology quickly scan sources to come up with a response like “Because the average hotdog contains a large amount of sodium nitrate, which is considered a carcinogen.” Then is could add “And besides, they give me heartburn.”
Having a conversation like that with a virtual assistant would be pretty compelling.
One other thought about the IBM Watson debating function. Watson’s arguments can only be as good as the human beings that authored them in the first place. To take the example in the demo about outlawing the sale of violent video games to minors, Watson can only articulate a compelling pro or con case if they exist in the documentation sources. This may be problematic. If nobody has figured out a good argument to oppose something that really needs to be opposed, Watson isn’t going to be the one to come up with it.
The other issue is that the smart people who carefully crafted these convincing arguments aren’t going to get any credit for them. I doubt that Watson is going to credit its sources. That’s one of the big downsides of all the free content available on the Internet. Most of the individuals who created the content don’t earn either kudos or monetary compensation. But, then again, having your views picked up and spread broadly across society can be a reward in itself. Who knows, someday something really smart you said may end up coming out of the mouth of a virtual assistant.