The basic idea of QS to make systems that help people make personal-data-driven decisions in their life – especially from a wholistic perspective.
However, QS application developers have faced the “Chicken vs. Egg” problem for years: it’s hard to build useful models (apps) without lots of user data, but it’s hard to get users without a useful models (i.e. a useful application). The few successful ones have been narrow in their approach to solving user pain-points instead of wholistic.
With LLMs, we now have models that can leverage any data source (make logical predictions, summarize trends, etc.) as long as the data is expressed in natural language. Additionally, goal-directed AI agents can even use this data to define goals quantitatively, determine helpful actions to take, and measure progress. We also have advancements in data governance (TEEs, trust-less computation, privacy preserving computation).
How do you see AI (especially LLMs and agents) evolving the value proposition of QS applications?
Well timed question, I’ve seen a lot of discussion about this privately and it would be great to get some of it out into the open and shared. One of the main good effects I think is helping to ease some of the process burden we face in learning from our data. To give a simple example, my data often has time stamp issues and other pesky formatting problems. I can now use a column of data as a prompt and say “make these timestamps ISO 8601 conforming” and also do time zone offsets without so much trouble.
I’m involved in some more complicated work with @tblomseth and @jakobeglarsen with developing ways to simulate self-tracking data so that we can create and test visualization frameworks under different conditions. This is more speculative, but super interesting and LLMs are really helpful for trying to get a better picture of dependencies and trying out different solutions.
In my view your question suggests rather more complete solutions involving total modeling than are really available, so I’m a bit dubious there, but for supporting the process of self-research by removing a lot of manual labor I think we’re on the verge of a very big shift.
Thanks for the great overview of the problem! I totally agree that the classic “chicken and egg” problem has been holding back QS applications for a long time.
The advent of large language models is a real game changer — the ability to work with any data source, converting it into natural language, opens up new horizons for personalized and adaptive recommendations. In addition, an agent-based approach, when AI not only analyzes data, but also forms goals, plans actions, and tracks progress, can significantly improve the efficiency and motivation of users.
Modern data protection technologies are also important — without trust in privacy, users are unlikely to want to share their data in sufficient volume.
Overall, I think that with the development of LLM and AI agents, QS applications will be able to move from narrowly focused tools to truly comprehensive decision support systems that take into account many aspects of the user’s life. This will make such applications more attractive and useful, which will help solve the “chicken and egg problem”.
It will be interesting to see how quickly these capabilities will be reflected in real products.