Stuff That Can't be Done by Humans

  • Kevin MĂŒller

AI in the workplace to help us see the bigger picture: In larger organizations it is hard to know what the other person is working on and what ideas are being generated.

En savoir plus sur le service Moodle de notre agence digitale.

Ideas in the Collective Context

In regards to "learning in the flow of work", we are exploring the use of intelligent chatbots inside the organization's central chat application. Chatbots are a bit controversial, as many people have come to know them as a nuisance rather than something useful. Even though I share a certain skepticism, I will gladly embrace "the machine" in my work-life if its application meets the following criteria:

  • Talk to me in a literate, natural language
  • Respect my input by not repeating unhelpful suggestions, be a good learner
  • Show me who has access to my data and where it is stored
  • Give me the control to opt out or shut you off
  • Do stuff that can't be done by humans

The last criteria is where I see great potential: In organizations of a certain size it is hard to know what the other person in the other location is working on, never mind thinking (granted that the latter is mostly exactly how it should be!). We want to use the machine to contextualize what we are working on and to link discussions with similar topics inside our organization. This will help us avoid working on the same thing as someone else without knowing it. It will also help make our ideas known in a broader context, allowing others to pick up on an idea in order to change or expand it; thus making good use of the collective's experience, intelligence and creative powers.

Imagine the following scenario in a companywide discussion channel in your chat, forum, issue tracker or email application:
Machine: "hey, it looks like your discussion relates to a discussion held here [link to discussion thread]."

This could be very powerful. Coupled with a good enterprise search, that points me to information, people, places and learning modules in the appropriate context it would greatly expand the possibilities and the potential reach of my work.

The Realtime Dilemma

So far so straight-forward? The complexity of Natural Language Processing (NLP) and Machine Learning (ML) involved in understanding and grouping conversations remains a challenge, especially in realtime. Besides the algorithms it also requires powerful machines to process information in a reasonable timeframe. On one hand, if the machine takes too long to process the information the context might be lost to the user or no longer relevant. On the other hand, in order to keep the flow and the focus, interruptions should be avoided. So like the "digest" format of mailinglists, I would prefer the machine to inform me quietly of a possible match in context, idea or discussion. For this I do not require realtime feedback. In an "ask" channel however, instant feedback would be appreciated. Here, rather than linking to similar threads or contexts, the integration of an intelligent FAQ is probalby better suited.

Powers That Lie Dormant

Much work remains to make our own organization improve its use of the human powers that lie dormant in our system. Even in medium sized organizations there is much to be gained through a better understanding of context and more learning from eachother. We are embracing the challenge to build AI supported solutions that help us and our clients make better use of the human powers inside the organization, rather than building better automation to replace the need for human interaction.

Qu’en penses-tu?