This post outlines three ways to build an A.I. without going broke.

For the purpose of simplification I will focus on three common use cases:

  1. User Data — learn insights for improving experiences — behaviour profiles
  2. Conversational Interface — customer support assistants (Siri, Google Home, Alexa…)
  3. Image Classification — use image data to discover insights automatically (used in self driving cars, Apple FaceID)

To build any A.I. system what you need is data which has been labelled properly. For example having a description and category for each data point of credit card purchase items:

category: food, description: mcdonalds, price: $4.95, timestamp: 04092018 12:10:00 PM

Without this your computer will not be able to learn anything from what you provide it…

Garbage in garbage out!

How to get started

Do everything yourself with free web/open-source tools!

You would be surprised at how many companies are focused on building the tools to make building an A.I. as simple as using Photoshop!

For the three use cases above here are some tools I would use to get a prototype out with less than 10 hours of effort on each and use all of them without paying more than $1!

1. H20.ai — Driverless A.I.

credit: screenshot from h20.ai website

You can import a csv of data and it will start to derive insights for you based on that data, think…import your credit card data and find some personal insights.

Pros:

  • Desktop app for mac and windows; tutorial videos to get started easily
  • Easy import of comma separated value files (csv’s)
  • Point and click interface

Cons:

  • Difficult to have an engineer replicate model code from initial application
  • Limited flexibility to integrate with existing services (unless you are an engineer)

 

2. Dialogflow

 

credit: screenshot from dialogflow.com

 

You can build a conversational chat interface and integrate it with slack, messenger, telegram, twitter all with a few clicks amazing right?

Pros:

  • Web application and drag n’ drop interface; sample projects to get you started quickly
  • Integrable into many messaging apps telegram, slack, messenger…
  • Additional services to make your system smarter such as conversation flows and message variation types
  • Integrable into any source project

Cons:

  • Limited access to source code for replication if needed, but are you really going to do this?!

 

3. Google CoLaboratory Notebooks

 

credit: screenshot from tutorial notebook

This tool is a bit harder to use, in a previous post I put together a tutorial to get people started here. With these notebooks you are able to get a development and training environment up in minutes and start classifying data — or images in under an hour.

Pros:

  • Easy installation of support libraries needed for image classification such as keras, tensorflow and openCV
  • Code level control and shareable amongst your development team
  • Integrable into your existing development cycle

Cons:

  • Best for developers and engineers
  • May require additional development time should you need to implement a native solution on your own server

 

With these tools you should be able to build a prototype for your use case given you have the starting data in less than 10 hours. You should even be able to test it with friends, awesome!