Stories about AI are everywhere. Or, as Axios put it, everything is AI now. Depending on which stories you believe, AI will either help save the world or possibly destroy it. So, it’s nice when an article that’s more educational than hype comes along. That’s why Forbes’ recent article "When AI Needs a Human in the Loop" caught my eye. Despite advancements in artificial intelligence, there are many situations where human intervention (human-in-the-loop or ‘HITL’) is necessary to ensure accuracy, accountability, and ethical decision-making.
The article emphasizes the importance of striking a balance between automation and human judgment. It suggests that AI systems should be designed to recognize their limitations and involve human expertise when needed. AI systems often struggle with uncertainty and handling complex or ambiguous tasks. HITL systems incorporate human oversight and intervention to address these limitations and help guide and validate AI-generated outputs.
Incorporating humans into AI workflows adds an essential layer of control and experience. While some fear adding people into the mix adds unnecessary costs, humans and computers together yield better outcomes. AI models can’t make predictions with 100% confidence because there’s no way to account for every possible uncertainty. People with real-world experience give direct feedback to the models when outcomes have low confidence. This helps train the models with a continuous feedback loop that yields better results in the future.
A number of industries have successfully embraced HITL-based AI systems, most prevalently the healthcare industry. One expert explains this approach as “recentering humans in actions and outcomes in the discourse about ever-smarter machines.” They are seeing better outcomes when it comes to identifying potential health issues and more accurate diagnoses. This collaboration also helps ensure AI systems are more aligned with human values to address the growing concern among some that AI might have negative societal implications in the future.
Zippin’s checkout-free platform uses a HITL-enabled AI approach. While perhaps not as critical as getting a medical diagnosis right, we realize the importance of every shopper being billed accurately for their purchase. When the AI-powered sensors come across a problem they don’t fully understand, a select team of experts is alerted to solve the task. This real-time feedback system not only ensures accurate transactions but trains the models as well.
HITL combines the best of humans and machines to yield the best results. AI is outstanding at quickly processing large quantities of data and is a great resource for repetitive tasks. Humans on the other hand are superior in solving problems when data is biased or limited. HITL is a necessary quality control component of any AI system, to keep models learning, just like humans.
