AI, Assistants, and Bots Demystified

This is the first in a series of posts I’m writing to help demystify the world of AI, bots, and assistants. See the full series.

Everyone’s got a bot, and you can’t go a day in Silicon Valley without hearing about AI or machine learning. All these new terms can get mis-used and confused, especially with so many companies trying to jump on the bandwagon. So, let’s start by defining some of the most common terms.

Here we go:

AI (Artificial Intelligence) is any technology that attempts problem solving in a way that a human might. There are many different approaches to AI, ranging from hard-coded, rule-based systems to statistical systems based on complex mathematics. AI technologies have been in wide and popular use for decades: routing packages for Fedex, predicting the weather, deciding when to activate your antilock brakes, and others. AI is so common that AI researchers often complain that they rarely get credit for it, which is ironic now that companies are boasting that they are using AI.

Machine Learning is a mathematical approach to AI that can simplify “noisy” data into patterns a computer can more easily understand. The information in the real world is chaotic, and people are very good at identifying useful information in a noisy environment — think of driving a car, picking out a voice in crowded room, or even trying to make out someone’s face in a grainy YouTube video. Machine learning enables computers to extract the useful patterns of information, and makes it possible to apply AI to these kinds of problems.

Bots are computer programs you access through a chat interface like Facebook Messenger or SMS. Some people also refer to voice-controlled interfaces like Siri or Alexa as bots, though I think it’s more helpful to only call them bots when they appear in a chat interface. Bots vary a lot. Some have only a few functions while others offer a broader array of services. They may use a lot, a little, or no AI at all. For example, is a customer service tool that provides many different functions but relies on a menu system with little AI. The Domino’s Pizza bot has a narrow focus (pizza everywhere!) but tries to support natural language, requiring more AI. It’s easiest to think of a bot as a program that uses a chat interface to perform a specific function.

Assistants are one application of AI that can perform tasks you ask of it via voice or chat. Assistants are one of the most visible and important AI-based developments because they may radically change how we interact with our machines. Originally, assistants were used to provide voice-control layer for a phone (Siri, Cortana, Google voice), but now, people are starting to use assistants in broader contexts (Facebook M, Alexa). As people get more comfortable with them, we think assistants will jump from being a layer over the phone or other device to becoming a personal proxy for the user, and we’ll explore that trajectory in another article.

It’s important to note that assistants may be made available as bots if they live in a chat interface, but not all bots are assistants. An assistant’s primary goal is to offer a single starting point to do many tasks, whereas bots are more limited in focus. Likewise, assistants that do not run in a chat interface are not bots. Ozlo, the Google assistant, Cortana, and Facebook M are all available as bots. Alexa and Siri are not.

Agents are the hidden workforce that powers many assistants. As an assistant interprets a user’s query, it identifies tasks and subcontracts them to agents: one agent for maps and navigation, another for setting an alarm, etc. Unlike bots, agents aren’t intended to perform in public. They’re the “behind-the-scenes” subroutines that complete tasks assigned by the assistant.

NLP (Natural Language Processing) is AI technology that translates human language into a form that computers can parse. NLP is central to how assistants (and many bots) interpret the words you say or type. NLP has historically been hard because language, especially voice, is very noisy. Fortunately machine learning has made this task a lot easier. As the quality of NLP has improved the last few years, it has also led to rapid commoditization. Whereas Google, Hound, Nuance, and others invested decades of effort into their NLP, platforms like and Microsoft’s Cognitive Services now offer world-class NLP to anyone with just a few clicks. This is one driver of the sudden explosion in bots and assistants.

Conversational Interface is an “inside baseball” term for any computer program that lets you interact with through a back-and-forth dialog interaction, whether through voice or text. Some people think conversational interfaces will replace graphical interfaces (buttons, checkboxes, swipes, etc.) altogether. It seems unlikely. Human language is great for expressing very complicated ideas, but sometimes simply tapping a button is the better solution. The best interfaces will by hybrid interfaces that let you do both, moving between modes as the task demands.

Conversational Commerce is trendy term that tries to make conversational interfaces seem more valuable: by adding commerce! So far, there is little indication that conversational commerce is especially valued by users. Again, sometimes buttons work better. True conversational commerce will emerge when assistants can evolve from simple task planners to true experts that can offer information and advice instantly and on-demand while you shop, allowing you to rely on your assistant to help you make decisions and not just for order-taking.

Have more questions or confusing terms you would like defined surrounding AI, bots, and assistants? Tweet me at @okito.

Also, if you haven’t already, check out our own assistant, Ozlo — the always available personalized expert for food and eating out. Sign up for an invite at You can also follow us on Twitter @TeamOzlo or like us on Facebook. Subscribe to our Medium for more updates like this one.