As we’re entering an age where AI is becoming available for all, the question on everyone’s mind is: “What’s the best kind of AI for business?” Often the perceived quality of AI is derived from the accuracy of the algorithm, the amount of required training data or the speed of training time. While these certainly impact the overall efficiency of the AI, they miss the core point: The #1 quality of business-driven AI should be focused around workflow, in other words, the ways AI affects our ways of working in an organisation.
Increase efficiency of every human expert
Human-in-the-loop (HITL) is a philosophy of AI which focuses on creating workflows where an AI learns from the human operator while intuitively making the human’s work more efficient. The machine performs an action, asks a human expert for input, and learns from the response it gets. An example of this is offered below. Ideally, the interaction process not only makes the human expert’s work more efficient, but also captures the combined intelligence of every expert interacting with the system. This way all the tacit knowledge of the human experts can become part of the same shared system.
Case: Increasing quality and efficiency with Human-in-the-loop Document AI
An example of a Human-in-the-loop AI augmenting existing workflows can be taken from the world of law. Lawyers today browse through hundreds of pages of documents daily, looking for indicators of risk. While detection of certain risk types can be trained to an AI using various methods of Natural Language Processing, the process still heavily relies of human senior level expertise to interpret the text. A mistake in recognising the risks can lead to expensive costs in business damages.
After pre-training the AI, the system starts to produce suggestions of risk-indicating sentences within documents, forming risk summarisations of extended documents. The human will then review the results, give “thumbs up / thumbs down / change risk type” -feedback on each suggestion. This feedback will further teach the AI to improve its performance. Once entire legal departments are operating around the same learning system, it begins to rapidly capture the collective intelligence of the entire department. Thanks to human supervision, the same document-revision quality is retained as before, only executed faster with each feedback round.
Capture tacit knowledge from any task that requires human expertise – in a single system
The Human-in-the-loop approach can be applied to almost all business processes where human experts apply their silent knowledge to perform a task. It can range from customer service (reviewed reply automation) to medicine (X-ray image revision), manufacturing (visual quality control) and finance (fraud detection).
In addition, the system can also be used to educate junior level employees – let’s assume that a X-ray based lung condition detection system is initially trained with only senior level specialists. The system can then create suggestions for junior level doctors working on X-ray imagery, pointing out suggested anomalies that might have been missed without prior experience. Even still, the junior doctor retains the power to either accept or reject the suggestion, using their own expert intuition to make the final call.
To summarise, the purpose of Human-in-the-loop AI is not automate, but to augment every single expert knowledge worker within an organisation. It captures essential information, learns from the feedback it gets, and begins to serve human experts through increased quality and faster speed of executing operations. It fits nearly all human expertise centric processes within organisations, and is ideally a natural, subtle part of everyday workflows in an organisation.
If you would like to know more how Human-in-the-loop AI could become part of your workflows, schedule a meeting with us here.