A diagram to help explain AI

This diagram was presented by Daniel Hulme, of Satalia, at the AIBE Summit 2017 and helps to describe the AI loop.
The following is our interpretation:

The Data, Insight, Action, Adapt cycle is commonly applied not just for AIs but for the general collection of data and how to make use of it. This original purpose shouldn’t be forgotten and should be directly applied to AI applications. Each step of the cycle concerns not only your input and response but also that of the AI. Let’s look at each individual step and how it impacts not just you but also the AI itself.

Data

An AI is only as good as the data you put into it. Data is a broad term covering a wide assortment of different facts. Concerning AI, such data is collected in databases for access and can concern sales, customer data, feedback, yearly performance, etc. Basically, anything that can be quantified and recorded can be used as a data set for an AI. To simplify you provide the information and the AI uses it to form its conclusions and reactions.

Insight

An advantage of AI is that it can ‘read’ through a very large data set far more quickly than a person could to draw its conclusions or responses. A well developed AI will take this insight and use it to guide future output and build upon already gathered results. The information an AI provides you can tell you very valuable information which can be used to shape company plans, financial strategies, product releases, customer interaction, and more.

Action

Once an AI has reviewed data and generated its results the next step is action. While it may be tempting to take any AI generated results and start applying them it is important to review the information before taking action. Remember, an AI is only as good as the information it has access to so before using AI generated information to guide policy (or direct implementation in cases of customer-facing AIs such as chat bots) review the results generated. Assure that important data was not missed and that the AIs processes are consistent and once reviewed these results can then be put into action. From the AI side, many AIs will build upon previous actions. For example, AI chat programs that interact with humans will often learn new words and form new sentences over time. The AI bases this new behaviour off of past results and newly gathered data gained from the real-time interaction.

Adapt

Lastly, remember that we live in a fast-changing world. Not only is AI developing rapidly but data changes rapidly as well. Real-time results that are accurate to current ongoing events are absolutely essential for success. Therefore don’t be afraid to change methods, modify programming, or introduce new data sources into your AI software. A well maintained AI will also adapt itself overtime learning new ‘skills’ and building on its possible results. A fun example of this is as of 2015 Google’s DeepMind AI had learned how to play video games.