Last week, a consortium of large companies, including P&G, DEME, AB Inbev, Puratos and Solvay, invited 20 AI companies from all over the world to compete at the inQbet hackathon. Faction XYZ won the first prize with their AI lab assistant.
Current situation – Paper note taking
At Procter & Gamble, R&D Lab Assistants are required to take precise notes on the actions they perform while doing an experiment or mixing a batch. They take notes in a paper lab notebook. After the experiment is done, lab assistants should enter the data from the paper lab notebook into an electronic lab notebook, where it is stored for audit, archive and data collection purposes.
Problem – Digital is needed
Only 18% of the electronic lab notebooks are filled in. That makes a huge 82% incompleteness rate. Workers don’t have the time or neglect entering the information in a digital format. This leaves P&G open to liability and patent discussions, not to mention a missed opportunity to collect valuable data in an usable format that could lead to new insights.
Solution – Artificial Intelligence lab assistant
By using Artificial Intelligence, we can have an AI lab assistant that listens to the worker as (s)he narrates what (s)he is doing. The AI automatically transcribes the text and extracts the relevant information such as the different ingredients used, their exact weight, as well as actions like mixing and checking for dissolution.
For the prototype, we used a speech transcription model based on the Google Cloud Speech API. The NLP models were rather basic for the prototype, and will be improved in a next iteration.
- Custom speech recognition model
- Making the AI conversational
- Better training of entity extraction
- Field tests
See the prototype in action
Our Head of Machine Learning Jos Polfliet accepting the 1st place award at the inQbet hackathon.