AI Assistants and the last mile problem

In five years, you will not be buying phones for the power of its camera or how sleek it looks or a foldable display.

In fact, you will not be buying a “phone”. You will be hiring a digital assistant.

An assistant who will be available to you through an ear piece, a flat piece of glass (“phone”) in your pocket and in your car.

What should a good assistant do?

Say, you hired a real, flesh and blood human assistant. Let’s say her name is Amy. How would you interact with Amy?

Amy, I want to see my dentist this month. Get me in.

or

Amy, for New York, can you get us into an Airbnb instead of a hotel?

or even this

Amy, can you find us a good place for dinner for Jill’s birthday?

Each of these seemingly simple tasks is very complex to do for a digital assistant. It requires multiple layers of context and third party interactions.

Getting the job done

Your assistant, human or digital, will need to do three complex things that each have varying degrees of difficulty.

1. Understanding language: Your assistant will need to understand what you are saying.
2. Finding context: Your assistant will need to ‘know’ who your dentist is and that “for New York” refers to your upcoming trip to New York. Your assistant will need to reach into your email, your calendar and your records to find and remember things about you.
3. Working the job: Your assistant will need to reach out to your dentist or your Airbnb host or a restaurant, to book your root canal, get you a house to stay in or that perfect table for two by the window.

‘Working the job’ is the Last Mile.

The last mile problem

Exponential improvements in natural language understanding will take care of (1). Your assistant will be able to figure out what you are saying (more or less).

Calendars, emails and other personal information can already be found in digital repositories. These repositories provide varying levels of access to digital assistants. This handles the data access part of (2).

Now, make no mistake, “context” is hard and determining and maintaining multiple layers of context is very complicated but it will be solved algorithmically.

This brings us to the last mile: working the job

How will Amy ‘talk’ to my dentist, negotiate an appointment and book me in?

How will Amy work out that perfect table by the window at my favorite restaurant?

Most small and medium businesses are analog / human at the points where they interact with customers.

Can we expect AI assistants to converse with humans for the last mile ?

While understanding human speech to determine intent is difficult to say the least, generating human-like speech to carry out a back and forth with humans is orders of magnitude more complex.

The complexity of the last mile comes not from algorithmic complexity, but from getting a variety of actors to interface with digital entities like AI assistants.

Handling the last mile

Businesses like medical offices and restaurants will need a pure digital, unstructured interface to carry out tasks from digital assistants.

Today, a business has a phone number (human interface), email (human interface), fax (human interface). All these interfaces to customers are manned by humans and therefore constrained by human work conditions.

Yes, business have websites but websites are poor interfaces for digital assistants. Most businesses do not have websites through which you can handle tasks.

The best digital interface will be one that can accept queries and commands in plain text and provide responses or perform actions. Such a pure unstructured digital interface will be a chat bot.

A dentist can deploy a chat bot that interfaces with her scheduling system, her patient records.

The retailer can have a chat bot that connects to their supply chain system and respond to queries about shipping times.

The restaurant can have a bot that responds to queries for reservations and information on specials and menus.

With such digital agents representing businesses, a digital personal assistant will only have to discover the interface. It can then issue text commands to carry out tasks.

A text conversation between digital agents could dispense with the ‘nicety’ overhead of human conversation and get jobs done fast.

The Assistants are coming.

But Assistants are going to need business chat bots to take them the last mile.

AI Assistants and the last mile problem was originally published in Hacker Noon on Medium, where people are continuing the conversation by highlighting and responding to this story.