How the AI in our pockets will help us be more efficient humans
Scheduling meetings: a task that humans can find tiring and frustrating. The type of work that should be robotized. That's a big part of Doodle's very own mission, and the chatbot Meekan is there to help achieve it. With its angelic patience and ravishing speed, the task of scheduling tricky meetings is on the verge of becoming the sole domain of AI. Every modern human has to know a little bit about AI. In this post, we share a few of our insights on the ways that AI can truly work for and with humans. We will also talk about the unique challenges that AI faces to be more effective, what parts of AI jive with human thinking, and what becomes an error message. And we'll see how the AI in our pockets is starting to become truly integrated in our daily human lives.
Let us start on a slight downer (but it gets better): right now, in the world of work and of organizing personal lives, many apps propose to help us manage our schedules. Technology's goal has always been to optimize the use of our time and human productivity. And yet, frustration with technology is still an all too common experience, especially when it comes to syncing not just our own calendars but the calendars of a whole team. So, what is going on?
The two main challenges AI must overcome to truly help humans
Two main challenges stand out, that are peculiar to AI, when it comes to automating scheduling. The two main challenges are:
• AI language comprehension, and
• the dispersed and distributive state of AI in our lives and devices as we use them now. Our devices play host to countless little strands of AI, each designed to handle an isolated small task. In somebody's phone, Whatsapp uses AI, as do Tinder, Google Calendar, Yelp, the Maps apps; but the AI of Tinder has nothing to do theoretically with the AI that Google Calendar uses, or Linkedin, or messaging apps, and so on.
So, AI comes in crumbs, in the form of a colony of isolated systems that only occasionally collaborate together. There is no “one world” of AI. This is a problem for an AI trying to help a group of people schedule a meeting.
Here's how AI is already working for you every day
Take a step back to contemplate how many things AI does for us in a day: the maps app can guide us to our job location in any city. It helps us know what subways to take, or which driver to get to pick us up. The list of everyday applications of AI keeps getting longer.
• email spam filtering
• customer service chatbots
• Personal digital assistants
• scheduling meetings
• matches on dating apps
• health and fitness monitoring
• anything with facial recognition, for example on snapchat
• shopping recommendations
• events your friends are attending
• smart homes
• credit card fraud monitoring
These are a multitude of small bits and pieces and individual tasks that AI is employed in. Each of the applications is proprietary to a different agency, business, networking platform, so for example Google Calendar belongs to Google, while one signs in to Tinder via Facebook. But now it gets interesting.
In the past, a major source of frustration with technology was that unless users were on the exact same platform, there would be no way their devices could collaborate. But as the world of work moves further into the industry 4.0 era, this inability to sync became a major hindrance to our productivity.
Nowadays people use apps for a whole forest of small tasks. Deeper and deeper, we now integrate informatics into every aspect of our lives. That translates into collaborations between groups of people as well as groups of devices—and the AI in those devices.
No-one can assume that each individual is on the same type of calendar. So the challenge for a scheduling AI is to make, say, five people's calendars speak to each other. Overcoming the challenge of the “distributive condition of artificial intelligence”, could unlock human productivity 4.0.
Doodle's chatbot Meekan is set to tackle this challenge of AI's dispersal across a whole range of individual applications, furnished by a whole stable of concurrent purveyors, which creates such a challenge to automated scheduling.
So this covers the aspect of distributive nature of AI.
The second challenge is understanding human language without too many hiccups.
This is how AI learns human language
When it comes to training AI in the human use of language, the devil is in the details. If someone's tone is sarcastic, another human would understand that (hopefully) – but AI will take things at face value, and follow a strange path of formal logic and ridiculousness. If, in conversation, a human briefly strays off-topic, as is natural, or conversation is peppered with double meanings, or let's say someone speaks in metaphors: these channels all might lead AI to malfunction.
Unless, of course, it knows how to distinguish the different tones of human utterance. Tone could be informative, sarcastic, dramatic, humorous, factual, you name it.
It's a little known fact, but companies are placing hundreds and thousands of technicians in data centers everywhere, to work on polishing AI's semantic precision. Every word has a semantic cloud of possible primary and secondary meanings –think of the many uses of the word “arm”. But now this is fascinating: A semantic cloud technician will be tagging every word with the many shades of its meaning. Semantic cloud technicians analyze human utterances, so as to show AI how us humans modulate our tone.
That operation is large-scale and complex, as developers shoot for an AI that can really understand how humans communicate, indirectly, obliquely, sideways, and effortlessly navigate the infinitely finely wrought tapestry of human linguistic expression.
We want AI to be embedded, not standing apart from ourselves
AI chatbots such as Meekan or a digital assistant like Siri or Alexa, are trained to ignore everything we say until it picks up something it may be able to help us with. The reason why we have to say “Siri, open Tinder” is because the word “Siri” works as an “on” button, for activation of Siri. The same goes for an email in which you might cc a scheduling digital assistant.
In that instance, us humans are being trained to command AI; but the goal is to get the AI to fill our needs intuitively, without humans having to learn how to operate it. The goal is to get AI embedded in our digital communications seamlessly, helpfully, conveniently, and to iron out silly robotic misunderstanding.
But misunderstandings are pre-programmed, if you forgive the pun. It's hard to teach AI common sense, if you don't know what common sense is. No-one knows what common sense is, although everybody thinks they know. Philosophers have never agreed on a solid way to define common sense. It is impossible to measure common sense. So we are stuck for a while with a major challenge to training AI in really truly understanding what humans mean when they talk.
That covers the second great challenge to AI mentioned in the beginning, which was language comprehension. The two key problems we discussed here have been : language, and the distributive nature of AI applications.
A unique moment in the history of tech
Developing an AI program that can schedule meetings well is a challenge that many developers still categorize as 'impossible'. But crucially, we find ourselves now at a point history when conditions are met for the emergence of new applications in AI. According to the blog wecognize, three key pre-conditions are:
• the fact that vast numbers of people now routinely use hardware powerful enough to run elaborate softwares as well as digital infrastructures, such as social networks and cloud-based services
• we now generate and share never before seen volumes of data. Personal stats, whereabouts, preferences, habits, needs, wants, employment, health, legal and finance data, private thoughts, goals and dreams for the future, entire love stories, falling-outs with friends and the histories of our every social interaction. All this can be used to train an AI in how exactly humans operate, and how they use their words.
• The structures are in place to harvest and store user data. That prepares the ground for technicians to test and develop better, more refined algorithms.
AI or sci-fi?
In the last two years, the business of Artificial Intelligence has mushroomed into a boom and forest of new ventures. Self-styled 'visionaries' base their understanding of AI on Science Fiction, which does not square with the reality of where AI is at today.
Nor does it square with where AI going. Scifi showed us robots with human-level artificial intelligence, intended to baffle naïve audiences. But that kind of naïvete has faded, and nowadays people are much too world-wary, declinist, post-enthused and used to technology and innovation. We no longer feel amazed by a story of android robots falling in love with humans. But what can really amaze some of us now is an AI that will schedule all our meetings with correct intuition.