How AI can fail you and how to make sure it doesn’t

At Data Universe, Benn Stancil took the stage to deliver provocative talk, drawing parallels between the evolution of internet and the current state of AI. 

He highlighted how companies like Instacart, Netflix and Uber redefined their industries with innovative approaches. Just as Netflix transitioned from Blockbuster to streaming, and Uber changed how we think about transportation, AI is poised to revolutionize various sectors.

He emphasized, the success of the internet wasn’t merely about replicating existing systems online but reinventing them. 

What does the internet have to do with AI?

Back when we wrote about the strange new world of the internet, we’re seeing similar parallels with AI.

Surely, we are all aware of the story that unfolded before our eyes. When ChatGPT came out, it was the fastest growing website ever. Nvidia stock is soaring. 

And so everybody has reacted to these last couple years by saying, “What if we took our products and we put a chatbot on it?”

AI product development: Put a chatbot on it

Drawing from historical examples like Webvan, he cautioned against blindly adopting new technologies without considering their actual utility. Just as Webvan failed to replace traditional grocery shopping, hastily implemented chatbots may not address underlying problems.

He pointed out the current trend of applying chatbots onto products and cited examples like Zoom AI companion, Jira Virtual Agent, and Notion AI. 

But we’re learning that AI requires a shift in mindset – it’s not about merely adding chatbots to everything, but about fundamentally transforming processes.

Organizations are also building internal AI-powered chatbots

While initially promising, he warned that many of these implementations might not deliver real value. He likened this phase to the early days of the internet, where enthusiasm often outpaced practicality.

We are still in the early stages of AI development

Early implementations of AI & chatbots are not actually working out because they’re not quite delivering valuable products right now and will required a more thoughtful approach. 

As such he offers a few principles to guide your AI development

Four principles to guide your AI development

1. If you build it, people still might not want it.

Avoid chasing trends for the sake of it; instead, focus on solving genuine problems. 

2. Rather than forcing AI into existing frameworks, consider restructuring the problem itself

Just as redesigning roads could facilitate self-driving cars, reimagining processes can unlock AI’s potential.

An example of this is Tom Cruise in Minority Report where we see redesigned roads for self-driving cars 

3. There is no shortcut to AI

Don’t underestimate the complexity; shortcuts often lead to subpar results. Instead, commit to thorough development and refinement.

An example of this is with OpenAI’s widget library. “Natural language to SQL widget” has 19 lines of code to “write a sql query for this question” that misses the mark on delivering what the user will really need. 

Building effective AI solutions requires significant effort and time in the same way that Simone Biles masters a handstand.

4. AI is best when you don’t see it

Avoid products that prioritize showcasing AI over solving problems. The best AI is invisible, working behind the scenes to enhance user experiences and seamlessly integrate into users’ experiences

An example of this is Google’s magic eraser can erase people from photos and you don’t know it’s an AI-based tool. 

Successful AI development emphasizes problem-solving over flashy technology

By learning from past mistakes and embracing these principles, we can unlock AI’s full potential and create truly transformative solutions.