Embarking on the journey of parenthood is one of the most profound experiences an individual can encounter. It's a path filled with joys, surprises, and inevitably, challenges. The first three years are without question the most critical in the human lifespan, mapping a child’s potential for the future, and yet paradoxically it’s the most clueless an adult will ever be within a new role. And this creates anxiety. A need for quick and accessible answers and direction.
Parent Sense aims to provide these answers. Founded by Meg Faure, Parent Sense is a global mHealth business, offering scalable, digital solutions for parents by leveraging generative AI. Meg is an Occupational Therapist with over 25 years of experience, helping parents navigate this critical journey of raising a baby. She has published eight books, thousands of articles, recorded more than 150 podcasts and hosted weekly online chats with mothers and fathers since 2008. With the advent of accessible chatbots, Meg sought out my expertise to develop yet another scalable solution to add to her already comprehensive parenting solutions.
I began my journey in Generative AI at the beginning of 2021 whilst working for Parent Sense. Observing the scale of the textual data developed by Meg over a 25-year career, I could not help but consider the opportunity to develop a conversational chatbot that could deliver high quality guidance to the global parenting community. Imagine it’s 3am and your baby won’t stop crying. Instead of grappling for answers in parenting books and online forums, you have access to an on-demand chatbot, designed to support your specific situation, and answer your immediate questions. The critical issue was that the chatbot answers had to be reliable.
The state-of-the-art model, back in 2021, was called BERT, a large language model developed by Google. I attempted to train this model on all of Meg's parenting data, but it was unsuccessful. The model was not good enough at learning the representation of the data, which, at the time we attributed to a lack of training data. We decided to put the project on the backlog and focus on revenue generating activities. A few months later, I left Parent Sense to joi a data and cloud company led by Julian Box. During my time at there, I learnt about a privacy-first, platform-agnostic approach to software development, reflected in the way we develop our PrivateLLM technology at Clairo AI.
Fast-track to the end of 2022, ChatGPT broke into the market and a paradigm shift ensued: the world changed their perspective of AI. Suddenly, everyone rushed to adopt GenAI. Meg Faure was no exception, testing ChatGPT on a myriad of parenting questions. Within hours her delight dissolved into dismay – the answers were terrible, dangerous and irresponsible. As a passionate leader in the parenting domain, Meg felt a strong sense of urgency to protect parents from this easily accessible but ill-considered content.
Her dismay was driven by two factors: firstly, ChatGPT had access to a myriad of conflicting and often erroneous parenting beliefs or trends. Secondly, ChatGPT was prone to ‘hallucinating’. In GenAI, "hallucinations" happen when the AI chatbot creates incorrect responses, either fabricated or very unlikely, which don’t directly answer the question or make sense based on facts and logic. When you ask ChatGPT a question, it is possible that it hallucinates due to the nature of the GenAI model predicting words in a sequential pattern, or sentence. We discovered this while testing ChatGPT with complicated questions on parenting. I then dug up the Parent Sense datasets from the archives and started fine-tuning the OpenAI models to learn about parenting. Again, fine-tuning was unsuccessful due to our lack of training data. And the models still predicted and delivered incorrect facts.
The research did not end there, our next step was to ingest the correct information into the prompt for the GenAI model when the parent asks a question to the chatbot. This technique is called Retrieval Augmented Generation, or RAG. RAG has become a popular technique across most GenAI platforms as companies can safely leverage their own data to generate personalised responses. The GenAI model therefore shifts from being a source of holding knowledge, to an engine for generating outputs. The value in the approach becomes the way in which the systems search for the correct information which holds the answer, based on user inputs. Using RAG, companies can create their own GenAI “agents”, which are experts in a specified domain, or area of the business. For example, a company may have their law agent, or marketing agent. We used this technique to create a parenting agent, aptly called aiah (an affectionate name for a parenting assistant using AI). aiah quite literally takes the guess work out of parenting.
After a period of development, aiah was ready make her mark on the parenting landscape. But to be useful, she had to meet parents where they are at. This meant interfacing aiah through three channels. The first channel is the Parent Sense mobile app, where parents have a suit of features delivering content customised to their baby’s age, including the chatbot. aiah’s second interface was through WhatsApp, giving users even easier access to the chatbot. aiah’s third interface (in development) is a reactive webapp, allowing access to on demand parenting advice on a zero-rated web browser – a cost effective solution for Africa. All three of these interfaces connect to aiah through APIs, allowing Parent Sense to maintain a single backend.
What sets aiah apart is its combination of deep domain knowledge, securely ingested into a private framework, ensuring that all proprietary insights remain confidential and protected. This approach not only safeguards the valuable parenting wisdom accumulated by Meg Faure but also assures users that their interactions are private. Furthermore, aiah's dedication to delivering highly accurate, domain-specific answers without succumbing to hallucinations or inaccuracies addresses the critical need for reliable information in the sensitive context of parenting.
The development of aiah by Clairo AI marks a significant advancement in creating tailored, user-centric AI solutions. By emphasizing data privacy, precision in content, and broad accessibility, aiah represents a pioneering effort to meet the nuanced demands of the parenting community. This initiative not only showcases Clairo AI's expertise in leveraging AI for specialized domains but also reinforces the potential of generative AI to transform and enrich our everyday lives with trustworthy, on-demand support.
