Why Implicit Signals Are Not Enough To Build Proactive AI Assistants
Modern generative AI has the potential to become the ultimate digital assistant—one that can be proactive and bring you things you didn’t even know you wanted—but the only way it will be able to do that is if it can understand the difference between you and the billions of other humans on earth.
We’ve had nearly a decade of digital assistants like Google Assistant, Alexa, and Siri, and they have still failed to evolve beyond being tools for setting timers and turning off lights.
Why can’t I ask Google Assistant to book me a flight to New York and have the ticket appear in my inbox? Why can’t I tell Alexa that I want to have dinner with a friend who will be in town in a few weeks and simply have a reservation appear on my calendar?
These would be the most basic tasks for a human assistant, but they remain impossibly out of reach for even the most advanced digital assistants. The reason for this is not a lack of technology—Google can certainly figure out how to automate the booking of a flight—but rather the fact that these assistants actually know very little about us, the individual humans who are interacting with them.
Let’s think about the kinds of things Google Assistant would need to know in order to book a flight end-to-end. It would obviously need to know your destination and travel dates. But it would also need to know things like what airlines you prefer, how much money you want to spend, what kind of seat you like. It would need to know that I, for example, will usually book Premium Economy—unless the only options are middle seats, in which case I’d rather just sit in normal Economy. Or it would need to know that I tend to prefer Delta, but I’m not going to go out of my way if it’s significantly more expensive or inconvenient.
These are the kinds of things where a human assistant would ask me questions to get to know me and my preferences, and then would generally be able to operate autonomously. But Google Assistant has no way of getting this data.
Indeed, it might seem as though companies like Google and Facebook know everything about us, because they can watch what we do, access all our messages and emails, etc.—but in reality this is a massive misconception.
What Google and Facebook do is build profiles of us by collecting implicit signals, which are things that can be observed. For example, your search history, YouTube watch history, credit card statements, location data, etc.
This data is powerful (as evidenced by Google’s market cap), but it has some fundamental issues that prevent it from being truly useful:
Narrowness: In reality, only a small fraction of who we are is represented in our digital footprints. Think about all the beliefs, preferences, tastes, and dreams that you have never expressed, but that make up the context that informs every decision you make.
Ambiguity: Without understanding the intention behind our behaviors, implicit signals are effectively useless for anything other than generic recommendations. Does searching Google for refrigerators suddenly mean I am deeply interested in appliances, or does it simply mean that my fridge broke and I need a new one?
Creepiness: Because this data is collected through observation, it feels creepy to use—even when the results might benefit the user—and is subject to increasingly tightening regulations around the world.
It is clear that implicit signals are not enough. In order to build the next-generation of AI-powered digital assistants, we need to radically evolve the way we think about personal data by focusing on explicit data, rather than implicit data.
Explicit data is data that is provided directly by users. It can be used on its own, or can be used in concert with implicit data in order to disambiguate that implicit data and make it useful. Modern artificial intelligences are then able to leverage this explicit data in a way that previous-generation systems never could.
We are on the cusp of a revolution in the way we interact with technology. We simply need to build and curate our datasets, and AI will be able to make our lives better in ways we can’t even begin to comprehend.
- learn more about YouAi
- learn more about Koji
- follow me: instagram, twitter