The Future of AI: Show, Don't Tell
Chat Interfaces are all the rage, but the future of Human<>AI collaboration isn't a long text thread.
As AI has arrived and started to boom, the interface of choice has been the text thread.
This has led to some questions like “Is the modern UI dead?” or “Is ChatGPT the next app store?” or “Can I build a startup that is just a wrapper around ChatGPT?”
While I’m sure some of these are more engagement bait than actual cogent points, there is clearly a broad question of how user experiences will evolve in both business and consumer software. More plainly:
“How will humans interact with AI in day-to-day technology use?”
Let’s take a look at that question from a first-principles perspective.
1. AI will impact everyone’s life every day
There is a school that believes AI is still hyped up and that it won’t be the game changer that many think it is. While we hear about all the companies rushing into AI, many companies are not, and even more companies are simply putting resources toward it without applying any fundamental changes to the business.
On the personal front, most people don’t actually use AI that much. An April 2023 survey shows only 5% of people used ChatGPT everyday. While that number has probably gone up with additional tools, the reality is that most people are still living and working in 2024 pretty similarly to how they did in 2021.
That said, it’s clear to almost everyone who has used best-in-class models (GPT4 or similar) that AI is already an incredibly powerful tool that embeds the whole world’s collective knowledge and makes it as easy to access as it is to text a friend.
Far from a novelty, I’ve used it to:
save countless hours on software development work (an example detailed here),
to handle language translation with my in-laws,
to draft legal docs,
to improve my writing,
to discuss health issues,
to learn about gardening, science, history, and more,
to make meal prep recipes and grocery lists
to come up with science experiments we can do with our nieces and nephews (here’s how),
to answer all sorts of “search-type” questions like how cable cars work, who is on the $50 bill, and more
to generate logos and other visuals for blogs or products,
and so much more.
Now, does it do any of these as well as an expert human? No. I’d trust a great engineer more than GPT, we’ve still had a great lawyer help with legal docs, and I’d still rather have a great chef determine our meal prep recipes. But all of those things are quite scarce (and expensive)!
But GPT is about as good as an average professional in most things. This is more-or-less confirmed by the various papers on how GPT4 has done on various professional benchmarks. It’s not perfect, but neither is the average professional!
The reality is this: for almost every human, AI can give you very quick access to professional-level performance. As most humans are only a professional in one thing, this is immensely powerful.
Abundant professional-level work across any knowledge-based industry should absolutely change your life. Dare I say, if it isn’t changing your life, you might be doing something wrong.
2. Big Change Will Take Time
New technology never permeates the globe overnight. The internet became broadly available in the early 1990’s, but didn’t change things until the 2000’s. Mobile phones followed a similar curve of a decade or more before causing dramatic “how did we get here” level change.
Similarly, we should expect AI to take about 10 years to really change the world. People need to build, iterate. Companies need to grow, scale, and adapt. All of these things take time.
Like many exponential changes, things will happen slowly… and then quickly. We’re in the slow phase now where AI is being used by more and more people and being increasingly relied upon for providing that base-level competence across many broad areas. Forward-thinking professionals have started to use AI to improve their own skills, cover their weak spots, and improve the service that they provide.
But the killer apps aren’t really out there yet. As Sam Altman himself noted, ChatGPT is a “horrible product” in part because a chat interface is incredibly clumsy for most things, even if it is universally applicable.
That brings us to the third point:
3. AI-Enhanced User Experiences are the Future
The view that anything and everything will be handled through a chat interface I find very short-sighted, and there are A LOT of people who seem to be working with that assumption. The game toward being “the best chat interface” will have a few winners, sure, but it is destined to be a battlefield littered with ghost projects.
Instead, let’s approach this from a greenfield perspective: what would the ideal user experience look like for, say, a Business Intelligence tool in an AI-powered world?
Such a tool would likely:
Have standard visualizations that allow users to get accustomed to the data being displayed.
Enable dynamic splits of that data in near real-time.
Provide AI-generated questions and observations that a user could click on to trigger a deep-dive.
Have proactive AI-driven alerting that would monitor trends, try to handle/filter out false alarms, and then raise the alerts with the appropriate person within the organization.
Have an agentic “AI organization” that could be tasked with doing deep-dives on key questions, e.g. “Why were my sales down last quarter?”, and generate a detailed powerpoint-type presentation with key takeaways, visuals, and recommended actions all highlighted.
And I’m sure much more, you get the idea: picturing a BI tool that is simply a chat interface is crazy and leaves so much on the table, even if it generates some visuals along the way. Humans are visual creatures, and the cliche that “a picture is worth a thousand words” is true.
Similarly, having some kind of AI-personal assistant without a visual interface to manipulate my tasks, needing to text my shopping app or my music app, and more simply don’t make sense. I don’t need to chat with a weather app, I just need to get the weather for my location!
The best AI-driven applications will have AI as a strong-but-mostly-silent partner behind the scenes, assessing the data and then interjecting when it feels like it can be helpful. When it is visible within the app, it should not be a separate chat window, but rather feel integrated within the app itself.
To take an example I’ve thought a lot about: an AI project/task manager should provide options that would call the AI, which could then generate previews of the proposed updates live for the user in the UI, who could then approve or reject those changes. By integrating with other services, it could then act as a passive “task checker” and intuit that the e-mail you sent titled “Summary of Q3 Earnings” completes the task to “Mail out Q3 Earnings.” And so on.
4. What does that mean for the Humans?
Of course, this leaves all of us wondering what this means for all of us humans. Will AI take our jobs?
Let’s examine a great visual from my former colleague Sarah Hinkfuss, now at Bain Capital Ventures:
Judging by the companies that I’ve worked with, my estimation that white collar knowledge work would break down to roughly:
50% of jobs fall into L1/high L2. These jobs should see increased efficiency from workers and run a bit leaner, but won’t see substantial job loss.
30% of jobs fall into low L2/L3 range. These jobs will change largely to “AI babysitting” where a human is still needed, if only for legal & compliance reasons. You will see substantial job loss as well as these roles become even more commoditized.
20% of jobs fall into L4 range. These jobs will go away, even if it takes time for the tech to get there and for organizations to adapt. These organizations will still have a manager/director-type human, but they will be overseeing a system of AIs rather than a system of humans. These jobs will all go away.
Taken together, this assessment would say that we’re roughly going to lose ~30% of today’s knowledge worker jobs over the next, say, 10-15 years.
That seems like doom and gloom, but it forgets an important piece: all of the new jobs created by AI! That isn’t just software companies: it’s all of the robotic companies that spring up, new businesses and industries made feasible by access to cheap knowledge. A lot more may be able to go into creative work and pursue new opportunities in that realm. And the list could go on.
AI may feel different than past technological innovations, but humans in a free market are quite adaptive. The best advice I can give (and what I’m trying to do) is answer the question, “what can I do substantially better than ChatGPT?”
That’s all I’ve got, but I’d love to hear your thoughts: where do you think AI is going to take us? Leave a comment below!
"we’ve still had a great lawyer help with legal docs" I think i know who this is!