I made this chart to answer two questions. On the left side I asked ChatGPT to rank its most powerful features according to its own estimation. On the right, I asked it to return the top ten uses for the platform.
It is not very dramatic but I still think it is quite telling. We have ten capabilities of ChatGPT on the left, as ranked by itself and ten ways ChatGPT says people actually use it on the right. The lines show connections to the relevant use cases. It is a kind of blueprint of how AI is used in practice. As can be seen there seems to be a distinct gap between what is possible and what the real world uses it for.
A digital tool is not just a set of metrics and features. It needs to be assessed by the interaction between its functional ability and what humans actually want it to do.
On the left, we have a list of engineering achievements. Any one of them would be viewed as a miracle ten years ago. They are triumphs of logic and computation. But the right side is kind of messy. It has fitness plans, document analysis and generation, emotional support, tutoring help and so on, and very high up, meal suggestions. Some of it may seem trivial but some of it isn't. What is fascinating is that this is what we really do when we are given access to some of the most powerful computational processes in the history of the human race.
It should not have been a surprise that people seem to use AI for doing the normal things in their life with more efficiency. It really is a handy to have weekly meal plan done for you in a matter of seconds along with a shopping list and recipes. Anyone with a family has to love that. Anyone without one too. What this sort of behaviour points to is the delegation of chores and resolution of thousands of micro decisions into a few simple prompts. No juggling with shopping lists and dates and so on which are time-consuming and can be a mentally sapping chore. Once people master the prompts for themselves and these efficiencies can be routinised into their daily lives then there will be no going back. It just wouldn't make sense.
But it is still interesting to note that despite all the hype, the real gains for people at this current stage of AI development is in general in the domain of the mundane. No flashing lights, no mirror balls.
Looking at the graph it is hard not to see that the greatest demand for AI utility is in its ability to understand and generate human language. Data processing, image production, and coding uses, while of great importance, play a less significant role to the public at large.
Natural language understanding is the prevalent feature that seems to underpin almost every practical use case: education, customer service, and creative writing to name just a few. Doing complex math and massive data analysis are some of the truly amazing abilities of these new platforms. But what people really want is something that they can interact with in a human-like way that can do useful time-saving things for them.
Intelligence is not about computation. It’s all about interpretation and understanding the way humans interpret the world, which is by means of language. So, it cannot really be a surprise that natural language power of the AI systems has such widespread uses.
It is also clear that the core capabilities of these platforms are being underutilised. Things like image analysis, coding, and real-time reasoning do not extend out much beyond niche groups. I don't think it is because they are all that obscure, but there is a lack of general awareness about what to ask of the system. For instance, creating a simple website is a trivial matter for AI and it could be very handy for the general user to know about. Maybe that will be the next step in the evolution of the technology. Developing a framework for a prototype or a minimum viable product can take a just a few minutes with the correctly phrased prompts. I am sure people will catch on over time. I hope so anyway.
The task, then, is not just to make AI better. The task is to help people think better about what they want to use it for and also have a better understanding of what they could use it for.
This chart reflects a kind of living structure that shows relationships between humans and machines. Probably the most useful takeaway is to look more closely at what regular people are doing with this technology on a daily basis. That is where the action is.
Their individual acts may appear trivial but there is a clear trend towards the delegation of routine work. Outsourcing vital cognitive resources is the most sensible use for the technology that we now have. People clearly want to live and eat better. They also wish to communicate more easily and more effectively, and they don't want to spend a lot of time and energy doing that.