By Ashoke Agarrwal
I coined Concierge Intelligence (CI) as a type of Artificial Intelligence (AI) owned by and dedicated to an individual and fully protective of his privacy. CI would aid the individual in understanding herself better and leading to better life outcomes in health, education, career and relationships – in general, as a putative ad copy would say: ‘Be A Better You’. Further, CI would handle routine tasks like shopping, bill paying, appointments, correspondence and travel arrangements based on a deep understanding of the individual’s preferences and needs and an up-to-the-minute and universal understanding of options. CI would be under the complete control of the individual, who can switch it off and on and decide on the level of access granted.
When I first wrote about CI in Feb 2021, the concept seemed at least a decade or more away. Not any longer. Like the world, I was unaware of the rapid progress of Large Language Models (LLM) technology.
Today, many factors indicate that the first generation of CI is around the corner. A CI prototype might already be in the hands of hundreds of millions worldwide! Let me explain.
For a couple of years now, Apple has been communicating the following:
:: Many of the functions and Apps on its devices – Siri, Keyboard Suggestions, Health, Messages, Mail, Music, Books, and Apple TV – use AI to enhance user experience and utility.
:: Apple puts ensuring user privacy as the highest priority. Therefore, all its AI works on data and software residing on the user’s device, under complete user control, and cordoned off from other entities, including Apple.
The penny dropped when I first read about the Journal App that Apple is readying for release with iOS 17. Journal App gives iPhone users the means to record their day-to-day activities and uses advanced prompt features enabling users to track their emotional state and the causes.
The latest iPhones carry specialised chips that allow the device to run sophisticated AI programs on the device itself. With the breadth and depth of information, the iPhone has about its users, the phone’s processing capabilities and the level of trust Apple had built with its users, all the conditions that make for a CI already exist. Over the next few years, iPhone users, prompted by Apple, will increasingly find use cases for the CI that resides over the phone. With each new generation of iPhones, the CI will get more powerful and within the next decade, Apple will likely brand this as a proprietary feature and build a revenue model around it. CI by Apple could be the next big thing from Apple after the iPhone. If Apple keeps its promise of protecting user privacy, iPhone CI will add to the quality of life and be one of AI’s boons.
While the wizards of Cupertino are coming at AI based on an individual’s shared experiences, the wizard who has given the world Tesla and SpaceX is taking a different tack.
Musk wants Tesla to be the first to launch a fully self-driven car without a steering wheel or a brake pedal to allow a human driver to take control. While many companies, Alphabet being one of them, are at work perfecting AI systems, Musk’s approach is entirely different from the rest.
Alphabet and others are trying to build a self-driving car based on an algorithm that relies on the following:
:: Signals from a hardware system consisting of cameras and radars that transmit in great detail, second by microsecond, the physical environment of the car as it drives through a roadscape.
:: And rules that codify the signals into millions of scenarios and actions that are needed to respond to the system.
The above approach is similar to the early days of Natural Language Processing, which tried to create language models based on the contextual meaning of words and rules of grammar and idiomatic usage.
In one sense, Musk’s flip on the AI needed to build a self-driven car is simple. He believes if humans can drive cars based on just visual inputs, so can AI. So, radars are the first things he has taken out of the equation. His second lead is even greater. Large Language Models (LLMs) like GPT 4.0 work through patterns that a Deep Learning AI system detects from a large enough set of training data without needing an explicit set of rules. Musk’s leap is that he can build AI systems that can operate in the physical world through a large enough training data set. The difference is that in the case of the physical world, the data set is visual.
Every Tesla carries a set of high-resolution cameras. And its software records all the actions that a driver takes. Further, all the data from the cameras and the software systems are transmitted to Tesla’s servers. With millions of Teslas worldwide, Tesla has an ever-increasing training data set.
Musk is not stopping at building self-driving Robocars but is busy building a human-like robot branded Optimus on the same AI principles. The training data for Optimus-like robots will come from recording humans engaged in various activities – cooking a meal, navigating a home, an office or a mall, playing a sport, etc.
Further, in all cases, the training data will be culled so that the robot learns from the best drivers, champion players, chefs, etc. So ipso facto, robots will come out of the gate better than humans because they learnt from the best and have the advantage of being faster, connected and untiring.
Paradoxically, Musk also pays lip service to the dangers of AI and contends that he is trying to build something like Assimov’s Three Laws of Robotics into the AI systems he is busy inventing.
So, between the CI that Apple is fast making a reality and Musk’s promised Robot Intelligence (RI), AI is set to impact the daily lives of all of us significantly.
Another AI revolution is brewing in the scientific field, launching tectonic shifts that will alter human civilisation. But that is grist for another post.
