Artificial Intelligence, or AI for short, has been a topic of conversation, research, and action movies for decades. Whether it’s HAL refusing to open the pod bay doors, Will Smith’s killer robot swarm or the ever congenial Data aboard the USS Enterprise, the images and ideas we have can be a barrier to appreciating how to leverage AI in 2017.
Last year, Salesforce introduced “AI for Everyone - Salesforce Einstein” and, with the updates released for Spring ‘17, practical AI is now available across all of its product portfolio: Sales, Marketing, Service, Platform, etc. Back office or front office, no part will remain untouched by the now present availability of some form of AI.
What It Is and What It's Not First and foremost, business leaders need to understand what AI is now as well as what it will be in the future.
The term “Artificial Intelligence” can be misleading. The tech and the value it brings is better described as “Machine Learning,” the ability of a system to take examples and learn to find correlations and patterns.
This is a bit different than the “brute force” approach of first 20th century chess-playing computers(1). Speaking in very general terms, these early systems looked forward into the 1 billion future possible positions on a chess board to assign values to a piece or move at a point in time, then repeating this process in a linear fashion as the game move along(2).
Machine Learning (ML), by contrast, describes the ability of an artificial process to detect patterns given a set of data and to do so at scale. The data may seem unrelated, opaque and overwhelming to the human, whereas these systems can now examine the seemingly infinite combinations of the data points and cluster these to reveal commonalities and possible correlations.
And this is putting it too simply. Today’s announcement(3) that Google’s AlphaGo has again defeated a Grandmaster at Go demonstrates how AI can analyze its opponent's behavior based on previously "learned" patterns through immense data sets, to pick up on the opponent's playing style, timing, aggressiveness of decisions, or (and this was the last hurdle identified) even changes to opponent's playing style through the course of a match(4).
The “learning” part of machine learning occurs as a human or artificial entity provides feedback to the system. This feedback can take the form of a priority weighting on known or revealed aspects of the data or even just examples of the desired objective. From this feedback, the ML system learns to improve its results for all future operations(5).
What’s Available Today Salesforce Einstein now makes machine learning AI available now for:
Database data (ex. Yearly Revenue = 1.2m; Market Value = 200b)
Image data
Textual data (aka. Natural Language Processing or NLP)
Let’s look at some examples of these features in a rough order of sophistication.
Sales Cloud Einstein includes Automated Activity Capture, an email and calendar integration with Google™ or Microsoft® Office to report on key activities and reveal patterns. The current version is more along the lines of Business Intelligence where the human is responsible for look for the patterns given the sophisticated reporting, however, it’s built on the Einstein ML architecture so the learning aspect will grow soon and exponentially.
Two additional examples are included in Sentiment Insights, part of Marketing Cloud Einstein(6). First is “Leadsift” which allows for the examination of social media data and other language data sources to score potential customer. This is an application of ML, specifically NLP to create sophisticated categorizations to rank and prioritize marketing activities in ways previously unavailable.
The second Insights example is “Lymbix Emotion Analysis.” This feature reports the tone and sentiment of multi-language conversations using Natural Language Processing (NLP) combined with adaptive learning from crowd-sourced data to identify over nine sentiment types. Here we see the learning aspect front and center as the system continually monitors, re-evaluates and improves its scoring.
These are just a few of the built-in features Salesforce has released to date(7). For a comprehensive audit of each Einstein feature, please navigate to Blog --> Einstein. The thing to note here is that as built-in features, these will be things that dramatically enhance the Salesforce products. These are things to look for, expect and start planning now to leverage. But they are not extensible. Not yet.
Many, many plans are on the roadmap to make Einstein’s ML power available for admins and developers to leverage. A few tools are available that makes AI accessible in Salesforce today:
Einstein Data Discovery easily imports and automatically examines variable combinations in datasets of many formats. There is a step-by-step wizard for setting up the data, selecting actionable variables and relationships of interest, then allowing different scenario & aspect selections to observe correlations. The user can then save the results and loop this flow to allow the engine to learn and improve the analyses. It even has as export to PowerPoint option.
Einstein Vision is a RESTful API just released and generally available to use AI to analyze image data and apply learning models for sophisticated classifications of all kinds. Einstein Vision is available at https://api.metamind.io or as an Heroku Add-on(8). Watch for other APIs to become available very soon. Salesforce has also released and supports the PredictionIO framework (9) to enable developers to create their own AI containers and services, now available as an Heroku buildpack(10).
Think Differently
Just as every team member knows how to use email, so also should they now understand how Machine Learning could (and eventually will) improve their business, to appreciate what kind of tasks ML is going to be much better at than the human(11). AI expects the user to point it in the right direction and let it do the discovery by providing the system with the critical data, a critical mass of this data, and the expertise to provide the system with the right feedback. The user is responsible for asking and answering the following types of questions:
What data is critical to our business?
What kind of information drives a process?
Does that data or process drive one of your organization's core strengths?
Are there factors that you suspect are impacting performance but the combinations and correlations are very difficult to tease out?
Successfully leveraging of AI, therefore, will involve a keen understanding of the business environment, an awareness of the quantity, priority, and possible interrelationship among data available (however unrelated some may seem), and a drive to realize the impact of future actions.
There may be a time when algorithms replace humans altogether but it’s up to you to imagine how AI can improve your business today. Chances are it’s already here.
1 See Turing’s Enigma-breaking bombe machine loosely portrayed in The Imitation Game
2 For a fascinating look at how AI is enhancing the play of these systems, See Google’s AlphaGo https://www.wired.com/2016/03/googles-ai-wins-fifth-final-game-go-genius-lee-sedol/
3 http://www.reuters.com/article/us-science-intelligence-go-idUSKBN18J0PE?il=0
4 And, of course, the AI never gets tired or intimidated!
5 https://www.salesforce.com/video/297129/
6 https://resources.docs.salesforce.com/rel1/radian6/en-us/static/pdf/Radian6Insights.pdf
7 https://www.salesforce.com/content/dam/web/en_us/www/documents/datasheets/cheatsheet-spring-2017.pdf
8 https://elements.heroku.com/addons/einstein-vision
9 Based on the Apache open source project
10 https://www.jamesward.com/2016/06/14/machine-learning-on-heroku-with-predictionio/
11 https://www.npr.org/player/embed/524731597/524879464
