The Mac's mouse, the iPod's click wheel, the iPhone's multitouch display, and the Apple Watch's digital crown are all part of Apple lore in which a new device class mandated a new user interface. But there's a significant exception to that established pattern: Siri. The voice agent emerged as a way to control some of the iPhone's features but was never a way to completely control it the way Alexa served as the Echo's main user interface. Rather, it could retrieve bits of information and complete simple tasks online.
Now, an app called Natural seeks to go beyond what agents such as Siri and Alexa can achieve in terms of transactions while remaining wed to the smartphone's -- or any connected device's -- touchscreen. A request using Natural's "generative interface" begins similarly enough to those of other voice agents, with a verbal or typed command such as, "Order some Chinese food" or "Book me a flight to San Francisco." With Natural, however, the query remains editable. You can change or add to it at any time -- tapping "Chinese" in the first request and replacing it with, say, "Mexican" -- or add when you want your San Francisco flight to take place or what your ticket price range is. Natural then instantly updates the options presented. There are no extended spoken exchanges; indeed, the app never speaks to you at all.
But the more meaningful difference is the degree to which Natural can handle the completion of the task. With other agents, ordering food or an airline ticket goes only so far before you need to transition to an app such as DoorDash or Expedia. That's not the case with Natural. In fact, if you want to order from a restaurant that uses Grubhub for delivery and you don't have an account, Natural will create one for you behind the scenes, optionally deleting it at the end of the transaction. According to Jerry Yue, the founder of Natural developer Brain.ai, it does this without any access to APIs. It has simply figured out how to navigate the service.
That may make it sound as if Natural is trying to make apps obsolete, but Yue says the company plans to allow other apps to tap into Natural's AI capabilities -- apps coming to you versus today's model of you seeking the right app. That said, Natural could go a long way toward bridging functionality for devices that don't run an app-rich OS such as Android or KaiOS.
The app offers a streamlined approach to transactions such as ordering lunch, buying gifts, and booking travel at launch. However, the web has allowed us to complete those tasks with a bit more intervention for decades. So, how about something that's more in line with the raised creative bar of GPT-3? Yue offers a taste of the technology's underlying power by demonstrating how a future version of Natural might be able to share a file via Dropbox and draft a range of emails, including a multi-paragraph cover letter that highlights certain skills in your background. All one has to do is fill in the blanks provided for certain details.
Natural lives in the cloud and can be accessed from virtually anywhere after voice verification. Yue says the company is committed to protecting privacy, starting with local voice processing. The product's guiding philosophy, he says, is "that tech should be like the air we breathe. The air is transparent and doesn't distort your vision, and everything you see is real. You can't sponsor air and make people see things you want them to see." (Well, at least not until augmented reality makes considerable progress.)
Brain.ai has made Natural available on iOS for free and promises an Android app soon. While there's no effort to monetize the app, for now, Yue says that potential revenue streams could include affiliate relationships and a premium tier.
Previous and related content:
What is AI? Here's everything you need to know about artificial intelligence
An executive guide to artificial intelligence, from machine learning and general AI to neural networks.
Alexa is becoming the anti-Windows
While Microsoft has staked out the high-productivity state of "flow," Amazon is focused on when you're not focused. Can the complementary approaches drive cooperation among the cloud competitors?
OpenAI's gigantic GPT-3 hints at the limits of language models for AI
The California research outfit OpenAI is back with another gigantic deep learning model, GPT-3. While it shows that bigger can be better in natural language processing, it also points to a potential absolute limit on the whole practice of language modeling.