Artificial intelligence · Engineering

How difficult is it to start baking AI into your product?

Maja Rašić QA Tester

March 25th, 2017

Our goal this year has been to leverage our data to personalize our product. One way we’d like to further that goal is by using AI to give our users relevant info and features. Is this something that front and back-end developers can generally handle? Or do businesses typically hire a specific person to handle AI integrations?

Reyner Dsouza Big Data Strategy Consultant and OD Expert

March 27th, 2017

AI is a highly misused term like data science. Ideally any person who is innovative and builds a feedback look to tweak your underlying business rules as per the learnings understood can implement the project and term it as an AI project. Before I comment more on your question I would like to know if you are looking for something like a smart recommendation engine or something more than that?

a a .

March 28th, 2017

Ok, you don't need AI, and no, front and back-end devs are not classically trained in AI.

There is no bachelor in AI. There is PhD in AI though.

If you want improved relevant information and better prioritise features, you should start with tracking better meta-data.

To answer your question though, it is not difficult, just takes time. Also, this is not a problem that AI can solve.

Remember, when you pick up a hammer, everything looks like nails.

Stephanie Wagner Founder at Agile Bloom, LLC

Last updated on March 26th, 2017

There's some free Udacity courses available, the prerequisites mainly being a couple years of programming experience, linear algebra, and calculus.

There was one that I took (I forgot what it was called...but it was free and the instructor was from Google) that was really good and he explains the approaches to use when using machine learning to solve a problem. I would say find a course that has most/all these topics: Classification, Regression, Clustering, Supervised and Unsupervised learning. Once you get a grasp of the basics, you can get started on using AI/machine learning to solve certain problems for your app (try TensorFlow and Google Prediction API).

There's definitely a learning curve, so if you need something right now, then maybe hire a guy. You don't need a PhD or anything to do this stuff on your own, though.


March 28th, 2017


Not knowing anything about your product, let me generally say the Personalization is typically a data-driven operation, whereas AI is typically model-driven. More specifically, in Personalization we use various methods to find a causality relationship between various elements that are known. AI techniques can then generalize this causality relationship to include "like" elements that we may not have seen before.

A classical example is that I touch a flame and my hand burns. So the causal relationship - which data mining will find - is between flame and burn. AI techniques can be used to generalize "flame" to anything that is "hot," therefore creating a model that will trigger when data elements (hand, heat) are present.

The key difference between data-driven and model-driven is that data-driven strategies are typically quantitative and, therefore, require large data sets. AI models, however, can be developed using small data sets and qualitative techniques.

Either way, a feedback loop will refine the results over time.

For most personalization projects, all you need is data mining and/or what's being referred to Big Data Analytics.

For either type of project, you will need a statistician on your staff. You will need a theoretical computer scientist or a mathematician if you are blazing new trails, such as for example, we now see in self-driving cars, personal assistants, etc.

My recommendation would be for your engineers to use an existing library that reduces a lot of common Data Mining or AI algorithms to API calls. An excellent environment for personalization is Spark, coupled with Spark's Machine Learning library. Using this environment, most experienced engineers can implement a very effective Personalization strategy that will improve over time. There is also lots of good Open Source tools coming out of Facebook, Google, Twitter, and PayPal.

Good luck,