This quote (see copy-pasted below) articulates why using machine learning is more robust... because "if you torture the data, for long enough, you can get it to confess to almost anything." The speaker justifies increased use of cognitive computing by stating that machines are better at the "discovery" portion because they can find the stories "that [are] trying to get out" of the data. So using machines for analysis are better than humans because they lack "any preconceptions or prejudices. That's discovery." 6 years later, the conversation in AI and machine learning has significantly shifted to acknowledge the inbuilt bias and prejudices that can be hard-coded into the technologies and softwares themselves (by the human beings that build these machines!).
---
Quote from Transcript:
"And finally, what about discovery? So I'd like to take you back to your days, we've all had courses in statistics. And I, I don't know if you had the same experience, but I had pretty cranky instructors in statistics. And they told me that we should always form a hypothesis and then test the hypothesis. Don't jump the gun, don't start looking at the data before you've formed a hypothesis. That was nice. But then [inaudible name] came along. Professor [inaudible name] came along and said, well wait a minute. You know, if you torture the data, for long enough, you can get it to confess to almost anything. What about the story in that data that's trying to get out so what about exploring the data and looking at it before you have any preconceptions or prejudices. That's discovery. And in fact, cognitive computing is helping us to find out stories that are trying to escape from that data are harmonious. And in fact, we're finding in our research in cognitive computing, that amazing things are beginning to reveal themself to us by taking a very new and cognitive approach to information."
This quote (copy-pasted below) argues that a technology solution like "Lucy" will help to understand cross-domain relationships. I think this is one of the most common benefits touted of Artificial Intelligence is that the machine can find new correlations that humans don't even know to start looking for. The "unexpected" findings are touted as a real benefit of doing this kind of big data analysis.
---
Quote from Transcript:
"But most importantly, we will be [inaudible] combine information between domains. So we will understand, we'll begin to understand and discover the relationship between water in the cities or the relationship between water and agriculture. Perhaps we'll better understand the relationships between energy and traffic and congestion and loss of efficiency. Because of the congestion that we have on our roads. Or we'll understand the relationship between financial inclusion and does it actually create jobs? So we'll begin to understand these things. So we'll create and discover cross domain knowledge which will then be available to these clouds, and these server systems and remember, it [inaudible] full connectivity."
AO: This interlocutor raised the following perceived benefits of sharing:
- Internal knowledge management (being able to make sure those in the organization can access each others' work/data)
- Inspiration: Being able to see the types of questions asked previously and how effective they were.
- Literature review: knowing what kind of work has been done previously.
- Reduced need for large sample size (= reduced cost to run the research); ability to build on existing work that has been done rather than redo the whole thing again
- Ability to test if a claim holds; increase the validity/credibility of a qualitative claim
- (With PECE), ability to get a sense of the perspectives and background context that goes into the creation of qualitative data (e.g. by sharing instrument; notes; stripping sheets; etc.)