Apple has always said it tries to put the human at the center of everything it does. It is likely discovering that it must do the same thing with Apple Car for a simple reason: human augmentation beats human replacement through automation.
People remain smarter than machines
The Information reports challenges involving the Apple Car project. That’s bad, I suppose, but the emerging truth seems to be that, to coin a phrase from Elon Musk, the need for humans was underrated by those pushing the AI automation envelope.
Look at it this way. Argo.ai CEO Bryan Salesky wrote in 2017: “We must build algorithms that enable our autonomous vehicles to respond to a deeper understanding of the likely behavior of other road users.”
That means smart cars must be smart enough not to collide with pedestrians, cyclists, other vehicles or even unexpected crash barriers or wind-driven obstacles. To achieve that, they must have machine vision intelligence, smart algorithms to make good decisions in an infinity of unpredictable situations (including errors in their own code) and the capacity to measure and estimate things like the trajectory and speed of other road users.
Soft skills are hard to code
To be safe on roads, these vehicles must also emulate the sixth sense most drivers have, that intuition that sometimes warns us when things may be about to go awry. Not only that, but they’ll also need to understand the interactions human drivers use to communicate with others on the road. And, of course, these systems must all be fully reliable in any kind of weather condition, including torrential rain, ice, and snow – and, indeed, when network coverage isn’t available.
Machines must become smart enough to emulate human soft skills — and achieving this seems to be where every automation project is stumbling. We appear to be finding that the limitations of autonomy begin where skills such as emotional and situational intelligence, intuition, communication, empathy, judgement, and others are required.
Augmentation beats automation
This realization is generating a change of approach. Look, for example, at smart manufacturing: While Industry 4.0 focused on replacing humans, Industry 5.0 explores the augmentation of them.
The industry thinks humans working together with machines should be able to achieve more and do it better. Surely this way of looking at things should also inform smart car development.
I admit that I expected self-driving cars to already be on the road. That didn’t really happen, though there are a few such vehicles. But what did happen is that as billions of research dollars were thrown at vehicle automation, researchers identified issues they didn’t expect. Liability and insurance, are an example, along with matters of network, battery technology and the need for networks of charge points.
That’s part of the reason the autonomous vehicles that do exist mostly handle pre-defined routes in semi-private spaces. It’s also clear that every company involved in this work has encountered unexpected challenges. At the same time, most vehicle manufacturers (including Ford) and many tech firms (Apple, Google, among them) are working on the technology.
Problems Apple has faced
The Information tells us some of the problems Apple’s teams have faced. One example is when an Apple test vehicle almost hit a jogger crossing the street at an unmarked crosswalk. The human test driver was forced to slam on the brakes to prevent hitting the pedestrian.
Following the incident, Apple added the specific crosswalk to its database, but even that addition exposes the inherent limitations of autonomous vehicles. And while it is possible that Apple’s AI is simply not as advanced as those developed elsewhere, this seems unlikely, given Teslas were involved in 273 oof 400 U.S. crashes involving driver assist systems, according to the US National Highway Traffic Safety Administration (NHTSA).
Alphabet subsidiary Waymo is arguably ahead of the curve for fully self-driving vehicles, yet as the NHTSA report shows, it also has accidents. As a result, the company only permits “trusted testers” to hail rides in its cars, and those vehicles also carry a Waymo staff member acting as backup to prevent emergency.
The persistence of the human operator means these are not yet fully autonomous machines.
“Instead of being programmed for every purpose, AI should be able to find answers and solve problems independently,” says Inis Ehrlich, German Europe Consultant for Artificial Intelligence.
So, what happens next?
It’s safe to assume whenever these vehicles do begin to appear in greater numbers that deployment will take plave in relatively limited scenarios.This is effectively what is happening as manufacturers deploy driver assistance technologies for specific tasks. But, for the most part, we’re going to need humans with their judgement and intuition at the wheel.
That means smart car development will pivot towards augmenting drivers, rather than replacing them.
With this in mind, it’s more plausible to accept recent reports Apple may license autonomous technologies it has been able to perfect to automobile manufacturers for use in vehicles alongside CarPlay. Even then, we seem to have no immediate ETA.
It looks like putting the human at the center of the experience will be just as important for the next evolution of transport as it was for the emergence of mobile, tablet, and PC. It also suggests human/machine augmentation will define the future of automation. That’s not to say a truly smart car no longer sits on Apple’s road map, but does strongly suggest a semi-autonomous system will arrive first.
Because soft skills are hard to replace.