A Product Management Perspective on Four Categories of AI
Originally published on LinkedIn on Dec 18th, 2023
AI was the biggest theme of tech in 2023 and there is good reason for that. After many years of development, a number of foundational AI technologies crossed the hurdle from being interesting experiments to being broadly useful in the real-world. LLMs were the most obvious example with versions of GPT leading the pack. There were other examples including diffusion models behind products like Midjourney and Googles GraphCast weather forecasting model. As a Product Manager in tech, I think a lot about the types of use-cases for AI that companies are considering adding to their roadmaps in 2024. Here is one way to categorize use cases that I have found to be useful:
1) AI as a Summarizer and Translator
Without a doubt, this is the clearest and strongest out-of-the-box application today. The current generation of LLMs can do an excellent job at ingesting complex volumes of text to provide summaries or to translate to other languages or for other audiences. At Tegus, I've seen first-hand how LLMs can ingest transcripts to create short and effective summaries. This unlocks massive collections of interesting content that would not be feasible for humans to read through in full. Across all industries, I believe that there will be a lot of value unlocked in this category in the year ahead. As an example of how LLMs can act as a translator between parties, you could imagine LLMs being used to take the notes from a doctor and then automatically creating a patient-friendly summary with at-home care tips, a justification for a proposed treatment plan for an insurance company, and a more technical set of notes for a specialist who will be seeing the patient next.
2) AI as a Noise Generator
My feelings on these use cases are not as positive. As soon as people realized what this new generation of models would be capable of, the potential to create massive quantities of content to flood communication channels was obvious. Here is an example that I believe we will have to contend with in the year ahead. Many people will still take the time to consider a personal e-mail sent by another human and the offices of many elected politicians will take the time to respond to personal e-mails written in by their constituents. Recently, I saw an ad for a solution that automates the process of writing a personalized e-mail to your elected representative. Although this may be effective for a few months, it seems inevitable that it will destroy any future potential use of that channel. If elected representatives start to receive an order of magnitude more e-mails that are entirely unique and all appear to contain personal stories from their constituents, they will no longer be able to engage through that channel. In general, I expect AI noise to shake-up a number of well established processes and industries. News is another interesting example. AI bots may visit the top news sites that hire journalists who are doing the boots-on-the-ground work of interviewing sources and collecting quotes. They can then re-synthesize those stories into new ones that are factually accurate and unique (based on current standards surrounding copyright and plagiarism.) For every real news organization, there may be hundreds that are knock-offs that pull ad revenue away from the original sources. Of course, organizations will adapt. For example, governments may roll out new methods for authenticated constituents to provide feedback or they may start relying on AI to provide automated responses. News organizations may accelerate their shift from ad-based revenue models to subscription models. I expect that AI noise will create a rollercoaster of short term opportunity and long term disruption.
3) AI as a Creativity Extender
The ability for individuals to express themselves creatively has been increasing at an accelerating rate over the past century. It was not very long ago that if you wanted to make a short video, you had to invest in expensive bulky equipment and then make a manual copy of a VHS tape and drop it into the mailbox every time you wanted to share it. Today, almost everyone has a phone in their pocket that can take video that is higher quality and then instantly share it with the world. The democratization of the ability to use that creative medium has led to a generation of creative influencers achieving the type of reach that would have been unthinkable without the formal backing of a studio just twenty years ago. AI is about to unleash a step-change improvement in the ease with which people can creatively express themselves. As an example, a number of companies are working on models that can generate high quality 3D renderings in a process that is as easy as interacting with ChatGPT. Creative industries such as gaming are now vulnerable to significant changes as anyone will be able to author a game with compelling graphics. Stock photography and graphics design work are two other industries that may see similar types of upheaval as creative tools become accessible to a much larger set of users.
4) AI as a Co-Pilot
This is the category that is the most exciting but it is also where there is the greatest potential for AI to slip into the Trough of Disillusionment. Throughout 2023, specialists in almost every industry experimented with ChatGPT or other tools and were able to demonstrate the risks in depending on those tools as co-pilots in their current state. As one personal example, I showed ChatGPT a picture of a positive COVID-19 rapid-test strip (upside down) and asked what I should expect in the next 72 hours. It told me that I was pregnant and that the next 72 hours would bring excitement, joy, anxiety and wonderment. On the other hand, there are a growing number of very compelling examples showing how LLMs and other types of models can be extremely effective at specialized tasks when proper guardrails, context, and supplemental logic is provided. These examples run the gamut from software teams relying on co-pilots for basic development tasks to scientists relying on models to help guide and accelerate the search for new materials.
As you look forward to 2024, what AI use cases are you most excited about?
*Image credit: Midjourney. AI was not used in the creation of the text.