The news coverage was relatively small and overshadowed by other announcements that Amazon made at their annual re:Invent conference recently, but when I read about their new DeepComposer, “the world’s first machine learning-enabled musical keyboard,” I had one of those moments of clarity in which I could see the future potential of AI.
DeepComposer is a two-octave piano keyboard that allows you to play or import a melody, then select the music style (classical, jazz, rock, or pop) and types of instruments you want to include. Using machine learning (trained on huge numbers of examples of each genre), DeepComposer generates a fully fleshed-out musical arrangement of your original melody, turning your two-finger version of Chopsticks into a symphonic piece played by an entire orchestra. Or a jazz/rock/pop band.
What struck me about DeepComposer, beyond the evident power of the computing that must be necessary to accomplish something like this, is that this is an example of one of the best models for the future of human/machine interaction: collaboration. A “cobot” or collaborative robot is a machine that works in close proximity with human workers. A human generates the initial core creative idea, then a trained AI agent fills in the gaps and does the work necessary to develop the first germ of thought into a finished product, freeing the human to continue innovating.
This Collaborative AI model is appearing more and more in different contexts. I believe that leaving the heavy-lifting work to computers is indeed the future, where the “unassisted human” will be an obsolete rarity by the end of the next decade. AI is excellent at capturing and analyzing facts and data, while humans still far exceed AI capabilities in bringing context, insight and creativity to that data. Systems that specifically acknowledge these two different skill sets and break tasks into AI-relevant and human-relevant portions are more likely to find solutions to problems that are both practical and inspired than systems that don’t.
Examples of collaborative AI abound
The company x.ai has built an AI meeting scheduler that it calls Amy Ingram. When you need to schedule a meeting with another person, you simply send an email to them and put “Amy” in the CC field. Amy will pick up the task from there, checking calendars and writing emails until “she” identifies a time and place agreeable to all, at which point “she” books the meeting.
Because Amy deals in natural language, meeting invitees may never be aware that Amy is not a real person. I know this because it happened to me! The meeting organizer did not have to do anything beyond send the original email. Yes, computer, please take care of the tedious, repetitive parts of my job for me, so I can spend my time thinking about what I’m going to say in the meeting, instead of playing ping pong with other people’s calendars.
One of the tricky things about collaboration with another party, though, is that sometimes emotions and other human elements need to be taken into consideration. In this regard, Facebook’s recent announcement that it has taught AI agents to collaborate with each other to play the card game Hanabi (which means fireworks in Japanese) is significant.
Hanabi is a game in which you can see other players’ cards, but not your own, and players try to work together to achieve the highest score possible from the layout they make on the tabletop. The game involves a lot of subtle communication and figuring out the other peoples’ perspectives and state of mind in the process, which makes playing well a notable achievement for a team of AI bots.
Until now, we’ve been delivering commands at a machine to help us with mundane tasks and calling that collaboration. But it’s now more possible than ever for future AI to read the task, the situation, and even us, and to suggest context-sensitive improvements which may then further spark our creative ability. If we get to that point, we may begin to consider AI and cobots as trusted colleagues and not just computers anymore.
Destroying the “evil robot” narrative
Is this really a large leap from where we are now? Today’s predictive and automated network management tools are practically trusted colleagues already and the addition of elements like voice automation and contextual or even emotional awareness will only increase that sense. And future generations, who will grow up hearing bedtime stories from Alexa and having their emails written by a more advanced version of “Amy” won’t think it odd at all to work with cobots.
Science fiction abounds with examples of power-hungry robot overlords and resentful AI slaves, but I think that what we’ll have in the future is neither one. Instead, bots will be more like the droids in Star Wars that act as companions and assistants. We’ll build on the collaborative AI foundation that is appearing all around us, in fields from network management to scheduling meetings to music composition and beyond. With our capacity to create and invent liberated by AI’s ability to take care of the little stuff, our human-machine partnership will accomplish far more than either one of us on our own can dream of.