One of the most overhyped and overloaded phrases of 2016 has to be …
Almost everywhere you turn someone is talking about bots as one of the faces of AI. At a recent Chatbot conference one of the speakers said “In Sillicon Valley people only invite you to parties if you are working on a bot these days. Apps are so over.” The Economist ran a feature titled “Bots, the next frontier” back in April. My grandma wants to know what skills she should add to her Amazon Echo. Bot marketplaces are popping up on every single messaging platform, from Facebook to Slack, to your local Walmart.
Welcome to the “peak of inflated expectations” on the hype cycle.
So given the next phase awaiting us is the “trough of disillusionment”, let’s explore what it might look like by focusing on personal assistants.
The hype of the virtual assistant.
Amazon Echo, Google Assistant, Siri…. they all promise to help us with everything in our daily lives and give mere mortals what only rich executives can afford. Let’s call our virtual assistant Echoistantiri, after the Greek goddess of AI.
User: Whats the weather tomorrow Echoistantiri?
E: Mostly dry, with a high of 10.
Amazing. And since all you have is a microphone to speak to or a text box to write in, you think anything is possible.
User: How much does it cost to fly to Spain?
E: I have found flights for £50 on Google.
User: Book one right now Echoistantiri!
E: Sorry, I don’t know how to help you with that.
User: Find a present for my niece who is 4.
E: Sorry, I don’t know how to help you with that.
What kind of a useless assistant is that?
And so we enter the “trough of disillusionment”. We smash Echoistantiri to pieces (since the new anti-bot harassment laws are still to be passed) and keep saving up until we can afford a real life assistant.
They would know how to deal with all those basic requests, unlike their virtual replacement.
So why can’t virtual assistants just do anything?
The problem is not so much in them understanding our requests. Natural language understanding technology has come a long way, and extracting ‘intent’ from a sentence happens with good accuracy. In the above present-finding conversation, computers can understand that your intent is to ‘buy a present’, that ‘niece’ is who it is for and that their age is ‘4’.
The problem is in knowing how to perform that task. Virtual assistants can’t possibly know how to perform all the requests users ask of them. Just imagine if Apple was in the business of building mobile apps for all conceivable use cases rather than relying on an app marketplace?
Further, the current mantra in early bots is to ‘make it do one small thing and do it well’. That way the expectations are clear. Contrast the above conversation with this one:
UberBot: Hey, I am Uber, I can help you get a car to your currently set up home address. You can say ‘order an uber’ or ‘how far is the closest uber’.
See what happens there? My expectations are set about the bot’s world and its understanding.
As a user I am not thinking “Aha, this bot can do anything” and I will not be asking it questions like “How many Uber cars are there in London?”. I may try, but if it says “Sorry, not sure about that, I can only help with ordering an uber to your home.”, then at least your expectation was set and you are not disappointed.
Given assistants are by their nature generalists, how can they get out of the disillusionment bucket and onto the slope of enlightenment?
Well first they need to learn how to talk to other bots. Amazon Echo does this to some extent with its concept of ‘skills’. If I install the Uber skill, I can ask Alexa to start the Uber conversation above. But it requires me to install that skill first and therefore know that it exists. It also becomes quite cumbersome to do for all requests I may have.
There is no concept of my virtual assistant somehow querying other bots for what they can do. It could for example come back and say:
E: Looks like you are trying to order a cab, should I pass you over to Lyft or Uber bots?
This interconnection between assistants and specialised bots will be the key here. How do bots expose to other bots what they can do, and then how do they pass the conversation thread and context (such as who I am, where I live) over to them? This is how the usefulness of the whole ecosystem will move forward.
The next step after that would of course be that a virtual assistant can perform my whole flight booking conversation on my behalf. For example, it could ‘call up’ the American Airlines bot, explain that my intent is to book a flight to New York next Wednesday, deal with the admin of providing a name and passport, pay with my card, and then just informing me it has been done.
But we are getting ahead of ourselves and entering the plateau of productivity and it is still some way off.
At Saberr we are super keen to understand what role a team virtual assistant could perform in making that team better, and how to forward what it can’t do to other bots and services, or even a real human.
The plateau of productivity can’t come soon enough!