Artificial Intelligence is here and our world is getting better every day! Except, not yet, not really, and not without a lot of work. But, hey, that describes all deployments of automation from digital assistants to paint robots.
To make good on the promises of AI, we need to take a closer look at content. Content powers AI – today and will well into the future. There’s a very specific type of content that makes systems like Watson smarter: content that is structured, organised by taxonomy, infused with metadata, and easily managed, tracked, and updated. In other words: Smart Content.
What is Artificial Intelligence?
There is considerable hype about the deployment of AI. It's a catchy phrase and an intriguing concept: That a computer can "think," as in answer a question, in a way that was not the result of calculation by pre-defined algorithms (i.e. traditional software code). There are different types of AI – some say three types, others say 33 types. Let's start simply, though, with three types of AI:
- Applied AI (also Weak, Narrow, or Specific) – is the most common AI deployed today and is built to focus very narrowly on a related set of information to accomplish specific tasks. Applied AI requires significant "training" in the specific problem domain. Self-driving cars are an example of Applied AI, because the AI for a self-driving car would not be able to drive a boat.
- General AI – is what most people think of when they hear the term 'AI'. While there is a lot of research in General AI, it does not exist today as a commercial offering (if at all). General AI can learn without further programming, it can extrapolate potential results from past and current situations, even those that are not directly connected to the current problem state.
- Super AI – is AI that has intelligence that far surpasses the ability of humans. This is the most feared result in science fiction: Skynet, I, Robot, the Matrix, etc. Currently it's just a science fiction concept.
Today's Use Cases for AI as Applied to Content Applications
In this article I want to take a deeper look at the most realistic and common type of AI: Applied AI. Applied AI may or may not use "memory" in the generation of a result or answer to a question. For example, Deep Blue, IBM's chess computer, did not look at the history of the current game, it only calculated from the current board state to the future board state and picked the optimum move to achieve the optimum future board state, checkmate.
According to industry analysts, AI implementation specialists, and customers, the current, most common application of AI related to content (document-based) applications, is to improve customer support results through self-service on the web.
If you've purchased a product or service, and something is not working as expected, the first thing you want to do is find the answer via that company's website. The last thing most people want do is to place a call, sit on hold for a time, and then talk with a customer support representative (CSR). Self-service is preferred in most cases over human service. But, finding the answer to a problem can be very difficult, especially for complex technology. A full-text search is simply not good enough for most problems.
For example, if you have a problem with your internet connection – it’s slower than you expect – what words do you use to search for an answer? If you search for the phrase 'slow speed,' the results might return information about the movie, Speed; marketing documents that declare how their connection speeds are promoted; how to upgrade to a higher speed; and what cable modem to use for the highest speed connection. If the search returns 25 items on the first page of search results and none of them are related to your problem, it's time to call support.
But AI can improve the results by knowing:
- The search was started from the business' customer support website
- That you are a customer, what equipment you have, how long you have been a customer, and your level of service
- If there is a system performance issue already being investigated
- If other customers in your area are also logging into the support site
- If your equipment is working as expected by sending a short test signal to your internet modem to check its state and the connection performance
Then it can derive a better result for you, up to and including a message that could avoid any further work. You'll be a happier customer, the service provider will have spent very little resource and everyone saves time – the ultimate win-win.
Artificial Intelligence Implementation Challenges
When it comes to document-oriented content such as customer support-case processing, the success of an AI system depends on the quality of the content and its ability to parse, identify, and interpret the words written. However, a document rarely conveys all of the knowledge needed to identify the right context, make a decision or take an action. Metadata that describes a document can be very helpful, but only if that metadata is complete, accurate, and fits into a known knowledge ontology.
In some cases, the format of the document (HTML, PDF, scanned images, Microsoft Office documents, etc.) may make it impossible for an AI system to "know" much if anything about the document. For example, many companies publish their technical content in PDF. Technical support documents often use tables to format complex ideas including system requirements, feature comparisons, and decision trees such as: "if power on, then do X; if power off then do Y."
Even when a table, for example, holds real text it is currently very hard for a computer system to understand what the purpose of the table is, how it is structured, and what the relationship between the rows, columns, and cells means. This presents a problem when the system is required to make automated decisions. The solution? Fuel artificial intelligence with Smart Content.
What is Smart Content?
The ideal description of content that feeds an AI system is Smart Content. The definition of Smart Content is:
- Content that is modular and componentised rather than large, monolithic documents
- Content modules that are described with robust metadata that describe the context for which the content is useful
- Content that uses terminology consistently and adheres to a well-defined terminology hierarchy
- Content modules of the same type that are ordered consistently from one document to another
- Even better if the content itself uses rich semantics to communicate to a computer all the knowledge that the author of the content, the subject matter expert, knows about that content – not just the words to be read but semantics that can unambiguously declare if the word "Chicago" is meant as the city, the style of hotdog, the style of pizza, one of two different baseball teams, the musical, the manufacturing company, etc.
How do You Generate Smart Content?
The attributes that define Smart Content are achieved by authoring structured content, which requires XML. For decades, many promises have been made about XML and what it makes possible in terms of content reuse and automation. In some areas such as tech pubs this has happened, but when it comes to mainstream content creation (by non-technical authors who create content that feeds AI systems) XML is no more prevalent than it was 20 years ago. The key is to Smart Content is to identify an XML-based content schema that allows non-technical authors to create structured content in a familiar environment, such as Microsoft Word.
Smart Content creation is typically a key component of a content automation solution that can not only power AI but also transform the overall content lifecycle. Content automation enables omni-channel publishing, reduces time-to-market, ensures content compliance, and ultimately drives increased customer engagement by delivering the right content to the audience at the right time.
Dave White, CTO, Quark Software
Image Credit: Sergey Nivens / Shutterstock