I had a very educational conversation with Stephen Scarr, the CEO and Co-founder of Info.com and eContext.ai – which is the world’s largest semantic text classification engine. eContext provides the taxonomy for bringing real-time data structure to any data source.
- 00:51 Stephen Scarr CEO and Founder eContext.ai and Info.com
- 01:38 What is Artificial Intelligence (AI)
- 02:06 What is Machine Learning (ML)
- 03:16 Artificial Narrow Intelligence (ANI)
- 03:38 Artificial General Intelligence (AGI)
- 04:38 Curated Taxonomy is needed for Business Artificial Intelligence (AI)
- 04:52 What does Artificial Intelligence mean for Marketers
- 05:50 How Programmatic Advertising uses Taxonomy to Create Context
- 07:00 The Importance of Context in Marketing
- 09:49 Challenges with Machine Learning
- 11:01 Chatbots and Virtual Agents need Taxonomy to Understand Context
- 14:29 Which Businesses Use eContext (eContext.ai)
- 15:32 Artificial Intelligence (AI) Trends – Chatbots
- 16:48 Advise for Businesses Implementing Artificial Intelligence (AI)
- The key if you’re an agency or brand – if you want to connect with people you need to understand how people interact with computers.
- If brands are unable to communicate through Chatbots and understand the context of conversations, then they are going to have as much trouble in three years as the retailers are having now trying to compete with Walmart and Amazon.
- If your business is new to Artificial Intelligence. The first thing you should do is to collect as much data as possible on how people are interacting with your brand, whether that be through search, social or reviews. Take that unstructured text (data) and use technology like eContext to structure and segment the data into usable, digestible bits of information to identify anomalies and understand what makes your data unique and valuable.
Justin Jones: What are your general thoughts on artificial intelligence (AI)
Stephen Scarr: The first thing I would say about Artificial Intelligence (AI) is that it’s a buzzword within the digital community at large. A lot of people believe they should be running before they can walk, keep it simple and begin with the AI fundamentals.
Let’s look at machine learning. Machine learning is very good at identifying anomalies in data. A huge challenge for the marketing industry is to establish a way to take those anomalies and turn them into real marketing insights, real business intelligence which in turn can be used to provide greater value to customers.
Marketers are underestimating the importance of the data sets being used to train machine algorithms to identify these anomalies. Where I believe eContext differentiates itself is in its capacity (in real-time) to take third party unstructured data sets and create a structured knowledge base that companies can use for their own machine or deep learning.
There’s also a big distinction within machine learning, whether you are able to do Artificial Intelligence based on AGI (Artificial General Intelligence) or ANI (Artificial Narrow Intelligence). We hear a lot about Artificial Narrow Intelligence versus Artificial General Intelligence.
ANI is where you are asking your learning algorithm to focus on narrow, specific tasks. Some of these tasks they’re doing very well. For example, machine translation work is using deep learning very effectively as well as image recognition. But when it comes to AGI, this is an area that the marketing world needs to embrace because we’re taking a holistic view or your audience, their preferences and their interests to clearly understand interests across all verticals and categories. But that’s very hard for AI to do because we’re many miracles away from understanding (in the correct context) human conversations from an AI perspective. There are many things they are doing to try and understand patterns and recognize anomalies – but to really be effective you must have an understanding across all verticals.
We believe that the approach of blending curated taxonomy with Artificial Intelligence is the way forward and probably will be for many years to come, despite what people tell you in Silicon Valley or in the newspapers. What does that mean from a marketing perspective? If you’re able to take unstructured text and organize that across all verticals then that general understanding is what’s known as a general index. Meaning, if you can take a corpus of data; in our case it may be social, search, or web content and then categorize that to our category tree. We have the unique advantage of taking this knowledge set and being able to understand what people talk about, search, or share across all verticals. When using that standard index, a client’s data can be categorized – help us compare interests against a general index to see where the key nuggets or business insights are.
Justin Jones: How do marketers leverage eContext?
Stephen Scarr: Let’s talk about programmatic advertising as an example. You’ve got the demise of the browser cookie. You’ve got millennials known as the swipe generation, getting more frustrated by irrelevant messaging from advertisers.
I believe firmly that we’re moving back into an era of context, delivering the right message at the right time to the right audience. To understand the context of your user you need to understand the text that the user is reading and engaging with. Context is very important.
For example, in the context of search when people are talking about Roller Coasters, typically they may be talking about amusement parks. If someone is having a social conversation about a Roller Coaster, it could be about Wall Street or about their love life. I believe what’s important from a programmatic perspective is the ability to understand the context of each webpage at a more granular level than is currently available in the industry. In our case, eContext has over 450,000 topics that in itself is hugely important and helpful when looking to understand unstructured data. By bringing unstructured data into a category tree it provides a common language for marketers to optimize against.
Let me use an analogy. If you’re driving visitor traffic for Nike Air Jordan and that performs well from a KPI perspective, you can then go into Nike basketball shoes, and if that performs, you can go into athletic footwear. What artificial intelligence machine learning is very helpful with is looking at these ontological relationships. People who talk about or search Nike Air Jordan; what else are they talking about or searching for. In the machine learning world, you’ve already got things like vectors often referred to as edges or word clouds exploring the relationships of words. But ultimately that’s all it is, it’s a bag of words or a collage.
Unless you bring that data into a structured hierarchical tree, it’s very difficult to understand what those relationships are about. If you’re uncle Ben’s Rice and you want to explore all the vectors concerning rice, you’ll get some lovely information nuggets like Basmati and steam rice cookers. But there are other things like Rice University, Tim Rice, and Condoleezza Rice which has no bearing to what marketers want to do from a content writing perspective or from advertising, creative perspective.
Justin Jones: What is the eContext Category Tree?
Stephen Scarr: The eContext category tree is specific to digital behavior. One of the challenges with AI machine learning is that they’re training (if supervised) on documents that are in the public domain. They are not proprietary, and typically these public documents are newspaper archives or freebase (bought by Google). You’ve got Wikipedia for example. Wikipedia doesn’t necessarily speak to digital insights. It may be very good when it comes to history topics and knowledge, but perhaps not very good in terms of brands or other nuances that are trending today. Things happen in real-time in the digital world. Therefore, it’s important that you have the ability to adapt and create new category nodes when they’re important, rather than relying on a public corpus. Our digital knowledge base is very important when it comes to building out these categories. Because we’re a search engine through Info.com we have a lot of interesting insights into people’s intent going back to 2003.
Ultimately the key if you’re an agency or brand – if you want to connect with people you need to understand how people interact with computers. It’s of no interest how people would read a history article on Wikipedia. You want to know if your audience are on a smartphone, how they engage with the outside world, or how they search in the real-time in the digital world. How do they interact through social media? How do they post content, is it articles, videos or tweets? It’s hugely important to understand these human and computer interactions.
Silicon Valley companies are investing a lot of money in finding people who understand computer-human interaction, because ultimately that is what it’s about. It’s about how companies and brands can become a part of a trusted consumer conversation.
Having a knowledge base built around computer-human interactions and then being able to structure that into 450,000 categories provides an incredible advantage when you’re doing deep learning to understand the relationships between various topics and text.
For example, we may have over a billion computer-human interactions, however in machine learning terms that isn’t enough data. To effectively map a billion interactions to topics; eContext can collapse the node level from 450,000 topics to tier 5 nodes where we have 30,000 categories. This gives you a lot of rich, accurate data for the machine algorithm to start learning and identifying the relationships of the words, create automatic text classifiers. If we see new data that hasn’t been mapped, the machine will likely be able to categorize this new information.
One of the challenges with supervised machine learning on public documents and broad data is this. Let’s say you have an article about soil erosion and you ask the machine to learn about that article. The article talks about other things beyond just soil erosion. It talks about topics that may be related but not necessarily to the same topic. It may talk about fencing, irrigation, organic farming, poverty and rainfall – all these things are topics within themselves. If you train a machine about soil erosion, then often the machine gets confused and maps things incorrectly which is why it’s important to have a knowledge base of text that is amped and extremely accurate when giving your machine algorithm a topic to train against.
eContext has API’s companies can leverage. If somebody wants to take their client’s unstructured data and make sense of it for applications, business insights, machine learning, deep learning they can use eContext’s technology. We’re able to provide real-time mapping to topics which quickly delivers a knowledge base from which to draw valuable insights and opportunities.
Justin Jones: From your perspective, what would you say is the most promising AI trends that marketers need to be aware of?
Stephen Scarr: I think the biggest movement and threat will be within the virtual agent community (Chatbots). You may be familiar with Apple’s Siri, but recently you’ve also got Google Home, Amazon Alexa and Echo. Apple are due to bring out an updated version of their Artificial Intelligence for home use. I think people are hugely underestimating the importance of these virtual agents. If brands are unable to communicate through Chatbots and understand the context of conversations and how to deliver messaging services and products against that, then they are going to have as much trouble in three years as the retailers are having now, trying to compete with Walmart and Amazon.
Justin Jones: If a company is looking to leverage Artificial Intelligence for the first time, what advice would you give them?
Stephen Scarr: The first thing you need to do is define the problem you’re looking to solve with AI.
What businesses need to understand is what resonates with their audience. To understand your brand, you need to understand what people think or are saying about your brand – what resonates with your audience. The first thing you should do is to collect as much data as possible on how people are interacting with your brand, whether that be through search, social or reviews. Take that unstructured text (data) and use technology like eContext to structure and segment the data into usable, digestible bits of information to identify anomalies and understand what makes your data unique and valuable.
Doing this, you will give you a better understanding of your audience – identifying the passions, desires and interests that resonate with your brand so you can design targeted marketing and content writing against these incredible insights.
eContext – Discover the world’s largest text classification system and find out what it can do for your business.