The AI Hero’s Journey

By: Kieran Penfold

20, November, 2018

Categories:

Artificial Intelligence - Exhibition - Featured -

  •  
  •  
  •  
  •  
  •  
  •  
  •  

by Juan Jose Lopez Murphy

 

Based on Converge Buenos Aires session: https://www.youtube.com/watch?v=-R50scM7mrA&list=PLWWBZaul8AhykP8Z1rczhidq_b3JszQtD&index=14

 

In his book The Hero with a Thousand Faces, Joseph Campbell introduces the concept of the “hero’s journey.” As Campbell explains, a hero’s transformational journey consists of a cycle of essential stages. These stages of transformation are similar to the AI-development process. In fact, five of them are particularly essential in any AI hero’s journey. Stage by stage, we will look at one of my own AI adventures in light of the “hero’s journey.”

 

FIRST STAGE: Call to adventure

In my case, the call to adventure came when an organization turned to Globant with a question: We have millions of chat sessions between us and our customers. What can we do with AI?

 

I didn’t know where to start. But then I thought of creating a virtual responder with artificial intelligence. I would train the AI with my client’s qualified responses, generating suggested answers to save time for employees.

 

To focus my attention, I started at the end goal and worked backwards towards the beginning. I asked who would use the AI, and why, and what such a product would look like. But most importantly, I needed to define what “intelligent” looked like to my client in this context.

 

SECOND STAGE: Refusing the call out of fear of change

I thought of the AI’s task as translating “sender language” into our “client’s language.” But how? I looked to Google and others to see what frameworks were available. Then I discussed this with the client, and we committed to a work plan instead of specific results.

 

This was one of the lessons of this stage: take it on iteratively and incrementally. And in the end, simpler is better. I didn’t have enough time to understand the business domain completely before starting to build a solution, so I had to embrace the empirical nature of the domain.

 

THIRD STAGE: Challenges that arise at the moment of transformation

At this point in my AI journey, I hit a wall. The solution would not work, or if it did, it wouldn’t converge, and if it did converge, the results were just horrible. I wondered if there was a bug in the Google framework, or if it was a question of model complexity or training time.

 

My first lesson here made the way clearer: I had to look at how good our data was, and whether it needed more or better processing. Also, I learned to look to the Open Source communities on this. In the end, I learned that true transformation is a long-term effort; you’ll never get it right on your first try.

 

FOURTH STAGE: Temptations that draw the hero away from the objective

In my case, the client was insistent on a time frame and results, while the endless possibilities of the technology tempted me to distraction. But at some point, I realized that even a fantastic concept does not hold up against stakeholders’ expectations. Using is believing. I had to turn this into an actual application. I also learned that it is critical to prioritize the definition of internal (functioning) KPIs and external (impact for business) KPIs. Letting go of perfectionism, I focused on making the solution run.

 

FIFTH STAGE: Return to normal, with hesitation about fitting in again

In my AI project, this is the stage of the prototype or MVP going live. All too soon, I get feedback from users and systems, and it’s time to get into production. Dealing with suggestions and qualitative opinions of the tool taught me the value of logs: of what’s consumed, what’s generated, what the user does with all this. I adapted questions and hypotheses to the qualitative feedback. Then I discovered “invisible” errors, e.g. the AI’s sentence generator misinterpreting the same word depending upon whether it was capitalized or lowercase. Here, also, infrastructure needs (on-premise vs. cloud), front-end technologies, and my engineering skills came into play. Finally, I remembered that every action must create a better experience for the user.

 

The product development journey, much like the hero’s journey, is marked by challenges. Developing an AI product requires patience, strategy, and like the product you’re building, the ability to learn along the way. Neither type of journey is straightforward but both do make better people—and better companies—of those who endeavor to complete them.

 

By accepting challenges that others won’t, Globant is building what others can’t. We are a leader in cognitive technologies because our teams dare to experiment, to create, to make users’ lives better.

 

Whatever the transformation, we are ready.

 

See Globant heroes Antonio Fragoso and Bill Bronske live at the AI & Big Data Expo.