Oftentimes, there’s a gulf between how it’s designed to work and people’s expectations. However, that’s changing as we see new changes forthcoming, especially as machine learning is gaining traction.
According to Stanford University, machine learning is “the science of getting computers to act without being explicitly programmed.”
How do Machine Learning and AI differentiate?
Artificial Intelligence refers to a field of study aimed at creating systems that replicate some or all traits of intelligence. Machine Learning refers to a set of technologies and approaches for creating systems that can be trained to mimic specific information processing tasks. Since ML can be used to mimic tasks that require intelligence ML systems can be AI systems. For the purpose of this article they can be treated as synonymous.
As Oak Intranet’s Senior Data Engineer involved in machine learning, I’ll be talking about where we see intranet search going in the future. Or, more accurately, how it fits into a wider set of concepts with examples of major features, and hopefully give some insight into how we see Oak’s own search features progressing.
Having looked at the sort of things that people search for on their intranet, queries tend to be for functionality rather than content. Like, “How do I make a page?” or even direct commands such as, “Change my password”.
This isn’t that surprising as an intranet, like Oak, is a bundle of functionality, and content is just a special off-shoot of that. It’s perfectly natural for someone to expect the search to naturally be a means to get to functionality.
Before continuing, I want to make a distinction between ‘classical search’ and ‘future search’:
‘Classical search’ refers to the old-school approach to implementing a search engine, where it’s content-focused. It aims to present you with a set of content items that’s as small as possible while containing all the things you would want to find. Technologically, it’s just a database query or a set of queries, with some fuzziness (like case insensitivity or phonetic hashing), and a post-query ranking algorithm. The query is supposed to return candidates, which are then ranked, maybe filtered further, and then presented to you.
If we accept the idea of search as a way to do things, then this classical approach is only a partial solution to a real problem, you might want to do more than view a list.
In comparison, ‘future search’ is what search will become. People rarely want to search again after putting in their query. Nor, do they want to browse list pages. They want to get stuff done. There’s an end-goal, and that’s where future search differentiates.
People want to get stuff done. It's an end-goal, and that's where future search differentiates.
Search will reach a point where it will interpret your interactions in terms of goals, take a typed sentence or a spoken phrase, and deduce what it is you’re trying to achieve.
If you were to search for, “Change my password”, the goal is very clearly to not list documents about changing passwords as it’s a well-defined action to actually change your password. The search engine would have a natural language processor, or NLP, detect that this is a request and lead to a well-defined application action that gives you a link to that part of the application. Or, better yet, sends you directly to the page itself.
Classical search has an intermediate step built in at a design level: it aims to get you to a results page and as such is inherently useless as it doesn’t provide functionality — it takes you to functionality via a database of content. This is a limitation in its ability to deduce and fulfil your goal. It can get you close to what you want but you need to complete the task by looking for what you want on a list page containing results you never would have picked. Essentially, the search engine has given up and is requesting your help.
In the future, search will refer to a domain that contains a complete specification of what can be done. So, if you were to query, “Change my password”, it processes the request within this domain and ends with a complete result. It would only present a list page as a fall-back in case it can’t send you to the correct destination or if you did specifically ask for a list, such as “news articles about cats”.
Currently, search treats every search you perform as a new task. It might give you suggested searches based on previous queries, but any follow-up is treated as if it were unrelated. However, ‘future search’ will remember as you go along. The act of searching becomes a process in which each search is influenced by recent searches, both short and long term, until cleared by you.
For example, you search via your intranet, “Get me everyone in marketing”, then for, “Provide their email addresses”, the search engine would deduce that when you said “their” you meant those returned in the previous search. This type of behaviour is only possible if the search engine has a type of episodic memory of previous interactions; otherwise your goal would have been impossible to deduce.
Join us in Part Two to find out more about what we can look forward to in the near future and what that means for the workplace.
Marc Hall has been at Oak since 2013, and is the Senior Data Engineer and Developer. He's involved in Machine Learning and applies it to future intranet development.
Marc enjoys gaming, reading, and Muay Thai.