Sramana Mitra: The job titles become critical because all the search engines and the AI on that side operate on the basis of keywords. Using the right keyword vastly enhances the findability of certain things.
Mike Flannagan: Absolutely. Say for example I tell you, “I want you to go find me a data scientist.” You’d say, “Okay, I’ll go search for data scientist.” What I really meant was I want you to go search for somebody who has these five job skills. Our system is becoming more and more intelligent about understanding what skills underpin various job descriptions and those job descriptions may apply to different job titles.
If it’s the same five underlying skills, then recommending that if you’re considering this role and this title, you might also look at these three other roles because the skills are almost an exact match. If you said data scientist is what you wanted and these are the skills of the data scientist, here’s a confidence level of how well someone with a job title of data analyst might be able to meet the requirements that you have. It wouldn’t be 100% but it could be 60% to 80%. You might say, “If I could get that job skill in the market that I wanted at a much better price point, that might be fine for me.”
Sramana Mitra: Yes, you may be able to train that person to get the rest into the job. A question comes to my mind that’s a follow up question to what we just discussed in this particular use case.
I’ll tell you where I’m coming from. A few days ago, I interviewed a startup CEO for this Thought Leaders in Artificial Intelligence series who’s working in the domain of advertising personalization – personalized advertising using machine learning models, not just ad targeting in large platforms. With all the different categories that he’s catering to, his experience is a very clear mandate that data can’t be shared across models. It can only be developed within a client, not across clients.
The question that I’m going to ask you is, are there use cases or verticals where customers are less willing to share data across the portfolio to develop the learning models versus domains and verticals where customers are more willing to share data?
Mike Flannagan: I think what you’re asking about goes to the heart of data monetization. Some people believe that they have data that is so valuable within their company or within their ecosystem that providing that data to someone else to aggregate and anonymize would somehow cut into the value of the data that they own. I don’t see this as a technology issue for machine learning algorithms.
Sramana Mitra: No, not at all.
Mike Flannagan: It really goes to the heart of, “Do you understand the value of the data that you own?” If you understand the value of the data that you own, you’re able to accurately assess whether the best value is to share that data freely in exchange for getting some aggregated anonymized intelligence back, or is your data so valuable that you should look to use it for your own purposes. There will be varying answers to that depending on who you are as a company.
If I were to say, “My knowledge of movie preferences is so personal to me that I don’t want to share it with Netflix because I don’t believe that the value of the recommendations that their engines can provide offsets the giving up of this very personal information for me.” I’m making a conscious trade-off decision that what you would give me back using collective intelligence is not worth what I’m giving up in order to get access to that.
The biggest challenge that I see as I talk with large enterprise customers is the vast difference in their understanding of the value of their data. It causes them to be all over the spectrum from “I’m not thinking about monetizing my data and I don’t have any plan for my data.” all the way to “My data is so sacred that I would never share with anybody for whatever reason.” The truth for most companies is going to be somewhere in between.