Why software is more exciting than hardware
Machine learning will do more to transform our world in the next decade than any shiny new gadget will.
Let’s face it. Hardware is boring. Software is where the action is — and machine learning is what makes the difference. Be honest with yourself: when was the last time you were really excited about a new product launch? Many of the defining devices of our age are, well, pretty old now.
You can argue that the main one was the personal computer — which arrived in 1977. The iPhone? Thirty years later, in 2007. That’s nearly a decade and a half ago. Sure, there were both phones and smartphones before that, but it was the inflection point that changed the industry. Even the iPad which finally made the tablet form factor viable is 11 years old, arriving in 2010.
We’ve gone a decade without any form of transformative gadget.
Does that mean technology has slowed down? Far from it. We should have shifted our attention long ago from the gadget itself to what’s happening within it. In fact, we were warned about this a decade ago: software is indeed eating the world. But we’re physical creatures, and the lure of a shiny new device is hard to resist.
Software in everything
What people pay less attention to is the creeping digitalisation of everything. Companies have had a go a making all kinds of things smart: smart glasses, smart speakers, smart headphones. While the results have been mixed, the digitisation of existing devices has been accelerating over the past 20 years. We’ll inevitably see more of that, even though it’s helping to create the supply chain problems we’re seeing globally right now.
But this effort is far from evenly distributed. For an example of that, we need only look at the major crunch point in 2020, as many of us were forced into remote working. We needed good webcams and fast. Many corporate laptops shipped without them, leading to a scarcity of devices, as the manufacturers struggled to cope both with massively increased demand and pandemic-induced manufacturing issues. This led, somewhat ironically, to the best webcam being a device that was basically a decade old.
Even many of those with build-in webcams ended up buying external ones. Why? Many of the built-in ones were, frankly, rubbish. The trend towards thinner and lighter laptops in fact led to webcam quality going backwards, as devices lacked the space for higher quality cameras.
Using software to cope with physical limitations
So, is Apple (for example) busy engineering better webcams into its devices? Maybe — let’s see what emerges in the coming months. But what it’s also been doing is making what it’s got already better with AI:
The on-board machine learning processor in the M1, which Apple calls the Neural Engine, will be working in real-time to optimise lighting and do noise reduction, too.
This is the subtle change that’s happening in more and more devices — they have machine learning capability built in. The M1-powered iPad I’m typing these words on right now can use machine learning to reframe shots on its wide angles camera, acting as a “smart” cameraman for my calls.
The smart software hidden in our devices
The smartphone makers who are serious about photography have been using machine learning to deal with the physical limitations of their devices. Smartphone cameras are too small to deliver narrow depth of field and the glorious bokeh that comes with it, so the software delivers it computationally, using smart software to do the heavy lifting.
Machine learning takes that a step further because it’s really good at dealing with massive volumes of data — like a big photo collection. Photo editing and sorting is a much swifter process than it used to be, with machine learning helping choose the best picture, and helping me extract the best quality of the image from it.
Don’t tell Apple I said this, but the excitement is no longer in the hardware. It’s all in the software, and its ability to take advantage of the assistive intelligence that machine learning enables.
Beyond the screen
And a lot of that will occur away from our daily driver devices. Ubiquity Ventures, for example, has long focused on investing in software beyond the screen: the ability for software to make other devices more useful. We’ve experienced this in the home with smart speakers, watches, and cameras. But these technologies are finding increasing expression in other fields, including manufacturing and bioscience.
Machine-enabled deep learning techniques are yielding discoveries in bioscience because they can handle and extract meaning from massive data sets beyond human ability to process unaided. And, if the pandemic of 2020 proved anything to us, it’s that we could do with learning a whole lot more about biology — and fast. After decades of being afraid of technology threats — like nuclear war — we’ve had a sobering reminder that natural biology can also be a considerable risk to us.
Indeed, perhaps the most significant “feature creep” in our devices has been the ever-growing number of sensors in our devices, from LIDAR in our tablets and phones to the multiple health sensors on our wrists. They help create more data at an exponential rate. But data, in itself, is not useful. A bad case of Data Love needs to be followed by a structured course of Data Insight, and machine learning is the perfect tool for extracting that.
Forget form factors
We have, largely, solved most of the major form factor challenges for modern tech. The new frontier is the combination of both the chips needed to make machine learning fly and the software that can take advantage of it. Steve Job’s bicycle for the mind takes a leap and becomes the sports car for the mind, as its ability to extend our capabilities increases.
Sure, the future may yet bring us some great, transformative new devices. Some think it will be AR glasses. Others think we might need to wait to see what Quantum Computing brings us when it’s available at scale.
But, in the meantime, the machine learning in our software, enabled by those clever little chips on our devices, are helping us make sense of more and larger data sets than ever before. And, as we face global challenges of deep complexity — like the climate crises or viral pandemics — machine learning might be our best ally in spotting the problems quicker and finding solutions more rapidly.