top of page

Human in the Loop - A Path for Automation

Writer's picture: Robert NesslerRobert Nessler

Updated: Dec 11, 2023




What is Human in the Loop?

Human in the Loop, sometimes referred to as “HITL”, represents a strategic blend of human oversight within predominately automated processes for the purposes of enhancing system reliability and evolving automation technology. In the case of AI, HITL is often regarded as a form of active learning to correct and reinforce outcomes in a supervised machine learning model.


Military and Automotive Examples

The strategic importance of HITL can be illustrated through diverse applications. For instance, the autonomous capabilities of drone strikes in military applications are complemented by human decision-making, ensuring ethical and strategic considerations are accounted for in critical moments. Similarly, in the automotive industry, Tesla's “full self-driving mode” exemplifies the importance of HITL, where the driver, despite the advanced automation, plays a crucial role in overseeing and intervening when necessary. This integration does not undermine the autonomy of the system; rather, it serves as an essential feedback mechanism, enriching the system’s learning process and improving overall reliability.


HITL for Voice AI

Back in 2018, I worked at a company named kea (that coincidentally our very own VoiceBite co-founder, Arjun Vasan, was also a co-founder of). We began experimenting with HITL for our Voice AI Restaurant ordering services, offering a glimpse into the practical application of HITL in customer service automation. 


Initially, customers would leave their orders as voicemails, which were then processed by offshore agents. This evolved into a more interactive system where agents could listen in on calls in real-time, interject if the AI struggled, and take over the conversation. However, this direct and obvious intervention was not well received by our end consumers. Admittedly, the whole experience was strange and can be only defined as uncanny.


Our breakthrough came with the implementation of a hybrid model using a predictive soundboard, reminiscent of 90s and 2000s prank calls. This allowed human operators to discreetly assist the AI without revealing the human-AI handover to customers. This approach not only preserved customer experience but also enhanced the efficiency and accuracy of the AI system - as suddenly we had built a feedback system that would improve itself over time.


HITL as a Strategic Asset

Having been in the Voice AI space for the past 5+ years, I have seen these technologies evolve significantly where it’s difficult to tell whether you are speaking solely to an AI or a human with a synthesized voice. At VoiceBite, we view HITL as a path toward automation, and not a crutch to fallback on. It’s important to make this distinction, as one approach lends itself well to growth and scalability, while the other will quickly run into issues as more instances of the AI are deployed in the real world. (Hint: my co-founders and I have personally seen over-reliance on HITL become the pitfall of companies as quickly as it helped springboard them.) Embracing HITL is, therefore, not just an operational necessity but a strategic imperative in the quest to build the best AI experiences for end consumers.


(Notes: I wrote the initial version of this post, ran it through Chat GPT and then revised it myself - I was the human in the loop, if you will. Happy to send my original version if anyone would like!)



If you’re interested in learning more about VoiceBite or booking a demo, feel free to shoot me a message at rob@voicebite.ai.

105 views0 comments

Recent Posts

See All

Comments


bottom of page