Would You Get So Excited if Google Duplex Called a Taxi Station to Book a Cab?

Google’s Duplex demo was breathtaking.
A computer that talked like a human, said “umm” and “mm-hmm”, and booked a hair appointment over the phone.
But for the sake of the discussion:

Would you get so excited to see Google’s Assistant pick up the phone and call a local taxi station to book a cab? 

Don’t think so…
I assume your first reaction would be: 

“That’s nice, but didn’t we solve that already with Uber?” 

And indeed it seems like Google’s Duplex can be a fine solution for yesterday’s problem
Today you reserve a table in a restaurant, order a taxi or a food delivery, without involving humans: using dedicated apps.
These apps are optimized for the tasks they do best:
  1. They eliminate the need to make phone calls, wait in line, call again…
  2. They can be work in the background (during meetings, other activities, etc.) 
  3. They maximize the probability to succeed with the booking (by checking availability, finding alternatives)
  4. They provide a unique set of features each domain needs in order to maximize our satisfaction

Let’s talk about booking a hair appointment for a minute. 
Disclaimer: I’m the Co-Founder and CTO at Missbeez, a brilliant mobile app that lets busy people easily order beauty and lifestyle services to their home or office.
Dealing with the business of ordering beauty & lifestyle services for over 2 years now, I’ve learned that many of the treatments have specific preferences and complexities that need to be addressed during the booking process: personal preferences, favorite beauticians, special needs (specific material, color, brand), special limitations (certain availability, sensitivity for a certain material), etc.
It’s hard to imagine a complex booking process made by a digital assistant without having to either provide cumbersome instructions in advance, or interfere in the middle of the process manually. Both options don’t sound very elegant, but hey, both of them are already solved by products that specialize in this domain, such as Missbeez.

Instead of re-inventing the old way of performing such tasks over the phone, Google Assistant should connect with the best apps (or services) for the job.
It needs to act as the pivot for all of those activities, just like Google’s Search already does:
  • Search for ‘Machine Learning’ and google will show you a small summary box taken from Wikipedia. 
  • Search for a certain hotel in London and you will get a quick booking box linked to hotel booking apps. 
Those small boxes (Knowledge Graphs) are based on existing tools and Google connects with the best tools to get the data or perform your task.

Similarly, a truly advanced personal assistant would find the most efficient path to perform a task, and in the case of the hair appointment - use an dedicated service such as Missbeez to perform the booking (in one simple transaction), and avoid the tedious process of calling salons to check availabilities and ask for an appointment.
In the real world, this is what makes personal assistants useful: they know how to get things done, they have their own ways, and they usually do it more efficiently than the rest of us do. This is what I would expect to see form a truly useful digital PA.

But that's hardly the beginning…
Launching an app in order to do “something” is easy. Even Siri can do that.
The main challenge is maintaining the conversation active while performing the task, because when it comes to real-word scenarios - the tasks are not really done in one action.
Many tasks require a sequence of actions to be made and we want our PA to handle them altogether.
This is where things fall short:
Today, the minute your digital Personal Assistant launches an app to perform the first action of a sequence, it basically “dies” and passes the context to that app.
The app takes over, while the history of your conversation with your PA is forgotten. It loses the context and any attempt to continue the conversation from where you left off - will probably fail.

What Google demonstrated with Duplex was a personal assistant that is presumably capable of maintaining a very long conversation without losing the context.
Embedding this capability with the ability to activate different apps and services without losing the conversation context will be a breakthrough!
We’ll be able to maintain long (even parallel?) conversations with our PAs and continue each conversation without recreating the whole context all over again.
We will throw follow-up requests such as:
  • “Hey Siri, can you recheck the availability in that restaurant?” 
  • “OK Google, please notify me when he is 2 minutes away” 
  • “Can you change the due date from tomorrow to Wednesday?” 
  • “Please check the last response we’ve got from him

Keeping the context active and jumping from one thread to another will lead the way to a personal assistant that not only acts as a pivot to “kick off” a task, but also a pivot for managing the whole sequence of actions, required to complete the task.

Can’t wait to hire one of those assistants…