Why AI Needs Human Supervision: My Journey to GetJenny

Joining a new company is always exciting. Doubly so when you join a great startup in the middle of a success story. As a co-founder of two startups in my past, the feeling is both familiar and thrillingly novel, as GetJenny already helps companies like Ukko, Colliers International, Terveystalo and A-Lehdet improve customer experience with conversational AI.

On the other hand, this is a somewhat similar situation to my previous assignment. There I entered a company of comparable size and helped grow it into a leading international player. It is the now-too-rare thrill of embarking on a journey. The destination is new, but the excitement and the experience is familiar.

At this stage I am still learning the ropes, digesting information and getting to know the people. The welcome has been very warm and the enthusiasm palpable. I, for my part, hope to bring in relevant experience from my past.

For example, though I have not worked in GetJenny’s field of improving customer experience with AI chatbots, I do have extensive experience from another angle.

I helped build technical solutions that measure a company’s customer experience through media data. I learned how to think of the interaction from the customer’s point of view, and see it in terms of actions that deliver value to the customer, positive or negative.

Natural Conversations and AI

Natural language is a keen interest of mine. I have studied computer science and linguistics, and dealt with natural language as a raw material throughout my career..

For example, while at university, I was involved in several EU research projects, where the goal was to retrieve, machine translate and present documents relevant to the user’s search intent. Later on, in an industrial context, I adapted the same basic task to deliver actionable insights to companies. Intent and action.

I am happy that at GetJenny I have a chance to take an even deeper and broader dive into what is often called Conversational AI. I get to work with some of the brightest minds in intent-focused augmented intelligence to understand what users want and present them with relevant answers to their questions. See more detailed explanations of Natural Language Understanding in practice.

The Importance of Supervised AI

While on the subject of AI, it is constantly improving with exciting new approaches like GPT-3, to pick just one prominent example. I definitely take the side of Janelle Shane, keeper of the entertaining and educational “AI Weirdness” blog about how AI needs supervision.

Source: @JanelleCShane on Twitter

I have worked in the Augmented Intelligence framework for well over a decade before joining GetJenny, and am still of the opinion that while “unsupervised” AI already delivers things capable of changing our world, there are many contexts where human control is essential.

Fluency in language is not enough when understanding is required. Replying glibly but not thinking at all of what the customer intends, and not thinking in terms of actions and consequences can even be dangerous. Because errors are too costly. Because explainability is needed.

When I first talked with GetJenny I was pleasantly surprised to find that the company straddles both worlds -- supervised and unsupervised AI. You need to feed off advances in unsupervised modelling, but you also need to tune it to your specific branch of knowledge needed, and you need to control that knowledge by human means if you want to avoid ghastly mistakes. Mapping that knowledge to data is where the future of AI lies more and more.

And I feel I am at the right place to do my own little bit of tinkering.