Product Design at Generative AI’s Frontier
Novel technologies are the latest and greatest, yet have the least amount of adoption because they are so new. They are usually the result of an engineering breakthrough or scientific discovery, and are orders of magnitude (e.g. 10x-100x) better than what came before them. GPT-4 is an example of a novel technology. It’s the most advanced conversational assistant in the world right now. Assistants like GPT-4 have been around for decades, but recent advances in algorithms and hardware enabled OpenAI to engineer a model 10x better than all the rest.
This makes all sorts of new things possible. For one, conversational assistants will never be the same. chatGPT was the fastest adopted product ever, and recently “Voice” was added where users can speak back and forth with their assistant. There is no limit to the conversations you can have and the assistant also can grab information from the web, making it 10x more dynamic and insightful than Siri (and others).
This is just what OpenAI is doing with GPT-4, but others are building on it too. Just in the past year, GPT-4 has attracted thousands of entrepreneurs who are integrating it into their own products via a paid API. Similar to how many different cities are built on Earth, many different products are built on GPT-4.
So what are all these startups building with GPT-4? You probably won’t be surprised to hear that many are building conversational assistants for specific domains. There are coding copilots for programmers, medical copilots for doctors, and legal copilots for lawyers. It makes sense that advances in conversational intelligence lead to better chatbots, but this is not all they will lead to. “Better chatbots'' are most obvious to us, because chatbots are what we’re most familiar with when it comes to language models; however, novel technologies unlock entirely new domains that couldn’t have existed before, and these new domains take longer to discover.
As Chris Dixon points out, when a new technology comes along we first try to carry over use-cases from existing technologies. Over time our design thinking evolves from this skeuomorphic design to native design. By “native design” I mean we consider the use-cases that novel technologies uniquely enable with their 10x capabilities. An example of this is how the internet started as static, read-only websites with content similar to pre-internet magazines and newspapers. It was years later that we discovered user-generated content and social media, which changed the internet paradigm from read-only to read-write.
Here are some early examples of native design thinking in AI.
领英推荐
The point is AI’s design space is mostly unexplored, and it will take decades for us to fully explore. Fortunately, if AI isn’t your thing, but unexplored design space is, there are many other novel technologies for you to choose from. A couple other examples include CRISPR Cas-9 and metal 3D printing. CRISPR is 10x cheaper, faster, and more accurate than previous gene-editing technologies, which is starting to enable personalized medicine at scale. Similarly, metal 3D printing is revolutionizing manufacturing and making it 10x faster for engineers to go from digital designs to physical parts. Printing reusable rockets is one of the initial use-cases of this technology.
As a product designer, your experience building with novel technologies will be like that of a pioneer discovering a new world with little development and few inhabitants. Overall, few products will have been built on the technology, which means established design patterns won’t exist. Also, novel technologies have 10x capabilities than what came before, so it will be like landing on a world with different laws of physics that support new types of buildings and new ways of life. Mental models need to be developed to fit the new paradigms brought about by novel technologies. Designing with novel technologies is more time-consuming compared to technologies that come with off-the-shelf components and mental models; however, you will have the chance to influence how products are designed in your category for years and decades to come.
Also, the tooling and infrastructure that support novel technologies are most immature, and it will take some time for this ecosystem to work through its growing pains. You will often find that the infrastructure you need is missing or under-developed. The main infrastructure component holding AI back right now is a GPU shortage. GPUs are what run AI models, and since the surge in AI products last year, demand for GPUs has far exceeded supply. This has made compute costs prohibitively expensive for some.
There is also the risk that you build your product on infrastructure that becomes obsolete since novel technologies change so rapidly. Right now, OpenAIs models are superior so almost everyone has chosen to build with them, but what if OpenAI gets permanently surpassed by another model provider like Anthropic or Google? Also, a lot can change in 3 to 6 months like technology trends and the dominant products in your category. The fast-paced nature of novel technologies requires that you continuously update your knowledge and refresh your skill sets.
Finally, talent is scarce within novel technologies. Since novel technologies are still relatively unproven, designers and builders are hesitant to spend the time developing an expertise in them. While this makes it harder to form a team, the fact that you are one of a few designers in the field is advantageous to you. It’s easier to differentiate yourself as a designer when your portfolio is aligned with a novel technology.?
This should make job interviews easier for you. If product designers are needed badly enough, and your fit with a company is good enough, you might be able to negotiate an above-average salary relative to your experience level. If the technology goes on to flourish, you can ride this wave and define your career by it. Hiring managers will still be impressed by your experience even if the technology eventually fizzles out and you’re forced into a new field. They will rightly sense that you can see further into the future than others, at least in some domains.