The debate about whether ChatGPT and GPT-4 are genuine breakthroughs for AI is missing the fact that GPT is now making complex AI-based services easier to use. The implications for the user experience are enormous.
ChatGPT, GPT-4 and, more generally, generative NLP (natural language processing) services have gotten much attention. Some believe they are a real breakthrough for AI. Others wonder if this is just another hype wave.
The reality is probably much more nuanced. Still, it appears very clear that GPT-based solutions can offer interesting new ways to organize and implement user interfaces for complex data.
It is probably still too early to say what exactly will be the impact of ChatGPT and GPT-4 in the long run. But we can already see use cases where these technologies can offer significant value. And those use cases are especially linked to the user experience and interaction with the user.
Let’s look at some use cases.
1. Users getting immediate answers based on public and proprietary data
We can see services where users once had to look at numbers and charts, but now they can ask questions and get answers in plain English. They can also combine these with other data sources and get additional context. For example, where earlier you looked at your wearable app data to see your resting heart rate and sleep score, now you can ask questions like whether your resting heart rate is normal for a person your age, and whether you can do something to improve it.
You can also ask about your last night’s sleep, how it differs from your normal sleep pattern, and reasons why it’s different –all of this based on your personal data (e.g., exercise data, what time you go to sleep, stress level, possible use of alcohol, etc.) and general information (e.g. typical reasons that have an impact on sleep). You can see more examples like this at Prifina’s demo service.
2. Users giving instructions to a service
In this category, we can see, for example, low and no-code services that help you implement automation, process data or tailor services without actual coding. Now you can describe a process or task in English, and GPT-based services can generate the actual code to implement it. In some cases, the resulting code may require some fine-tuning. But this can offer a much better user interface for low-code solutions than earlier services, where users had to draw some diagrams or use some complex macro languages.
3. Services utilizing GPT components to improve usability
Here, ChatGPT-type functions are embedded into a service. For example, it could be a health service that helps diagnose some symptoms – here, GPT is used to categorize symptoms based on descriptions and data from several different sources.
Or, it can be used in a service to evaluate supply chain risks using public data from countries and products combined with risk scores and the company’s own production and cost data. It could also be a user interface component to ask for additional information from the user, or give them additional information.
Usability is a big deal
One can say those are quite simple use cases, and not really a big revolutionary breakthrough for AI. On the other hand, I wrote earlier that the user interface and usability are real bottlenecks of AI services. From an AI evolution stance, the use cases may not be a huge advance – but from a user point of view, it arguably is.
Usability is a really big issue with data analytics. Currently, analytics results often appear in the form of numbers, charts and tables that require much effort to make sense of. If a user wants to tailor them for their own needs, it takes quite a lot of work and often also some coding. But if you can get answers to your plain English questions in plain English and those answers can really ‘format’ the data to answer your questions, that is a pretty big deal.
Actually it goes deeper than usability and user interfaces, because it requires analyzing, processing and combining a lot of data to get all these to work smoothly for the user.
There have been a lot of discussions about the reliability of answers from ChatGPT. That is, of course, an important and concrete question, but it’s a different issue from the technical capability to analyze data and find answers for the user. Obviously, if the underlying data is unreliable, the answers will be too. This must be taken seriously, especially with services that use public data, because we know there is unreliable and intentionally fake data. So, it is important to think about which data can be used and how to qualify data.
GPT puts usability in the hands of the users
Considering my examples above, it is crucial to remember that we’re not just talking about using ChatGPT as it is now, or services that use all the data you can find on the internet. This is actually about using these models and technologies more generally as an interface with all kinds of data: private, proprietary or public.
It can be your personal services that use your health data and data from health care services. Or it can be the company’s internal data with some selected external data sources. And the really big thing is to get immediate answers to serve all kinds of needs.
It will be some time before we see the long-term impact of GPT-based services. But we can see now that it opens new ways for humans to interact with machines and data. This alone is a big thing. It offers more people the capability to get answers and information from data, and it accelerates the ability to implement services that utilize data from many different sources.
In that sense, it’s a pity that many comments around GPT are quite polarized, with people arguing if it’s human-level AI or just hype. In practice, it is better to focus on real things that it can already offer now, and what kind of new doors to development technology and usability it can open.