NEAL ROMANEK: With that ability for AI to do so much, how do you build best practices for its use? How do you avoid, for example, indexing so overenthusiastically that your database becomes useless?
BILL ADMANS: When we use the term AI, most often we’re actually referring to machine learning. We’ve taught the system a very precise set of information. A good example is object recognition. A machine doesn’t have the intelligence to look at something random and say, ‘Oh, this is a mountain’, without us first teaching it what a mountain looks like. Face recognition is another area where we teach the system what a person looks like and what an individual person’s biometrics look like. Then, it’s a series of analysis across frames, compared to a database of pre-learned faces. As manufacturers and suppliers, we have a responsibility to teach our customers. That’s a key part of the process when we go to market with products. We educate folks on how our products work, what the features are and how they can use them. All the resources we have for getting information out are really important. RAOUL COSPEN: People are often thinking, ‘with AI, now I’m going to be able to do lots of things. I can NEAL ROMANEK: It sounds like there needs to be some education about AI. Do your customers really have a sense of what AI is and isn’t? “PEOPLE CAN COMETO IT NAIVELY AND THINK THIS MACHINE IS GOING TO FIX EVERYTHING”
ANDREW BROADSTONE: Our focus is on operations teams, and one of the real challenges is that you don’t want to keep adding more alerts. The problem some of these teams have is that they’ve got too much information. Too often, they’re shutting it off and saying that it’s just noise and they don’t want to hear it. So, the challenge is to use AI to say ‘these are the truly important things, and these are not’. By facilitating a human process of engaging with the data, where people can say this incident was significant and this one was just someone spilling coffee, we can extract tagging that helps us train our models and reduce alert noise. We can really start to understand what humans actually care about. MICHAEL PFITZNER: The goal of using AI is to be able to do more – better and effectively. Hopefully, it helps you concentrate on what’s important. If you examine a video recommendation system, for example, you don’t want to
spend time looking around, you just enjoy good suggestions. This is what AI should be. It should be subtle, giving support almost unnoticed. The most value we see in it are tasks that are repetitive or annoying, that are not the core work of journalists. They rightfully see their work as a kind of artistic action – as wordsmiths. It has to be careful and precise. The principle here is to let people concentrate on their core tasks and work on what they enjoy. Then, you have better results. RAOUL COSPEN: You want to stay in control. We need to rethink the way we want to expose AI to users and how to make it user friendly. One of the key aspects of AI right now is that it’s not necessarily super accurate. You want it as accurate as possible. But you want to be able to QC that AI-generated content yourself. It’s a matter of exposing AI results and giving users very simple interfaces to say, ‘I confirm/I don’t confirm’ or ‘let me make this suggestion’.
WATCH THIS VIDEO WITH QR CODE
WATCH THIS VIDEO WITH QR CODE
RAOUL COSPEN, DALET
“WE NEED TO RETHINK THE WAY WE WANT TO EXPOSE AI TO USERS AND HOWTO MAKE IT USER FRIENDLY”
ANDREW BROADSTONE, ZIXI
feedmagazine.tv
Powered by FlippingBook