People are fond of listing things that computers & robots can’t do; like creating original artwork, composing a symphony, appreciating fine wine, or cooking a tasty and nourishing meal from a handful of ingredients. We’ll temporarily ignore the fact that many people can’t do those things either (and some machines can), and look at the equation from the other side. Modern machines, with advanced sensor technology, huge processing capacity, and sophisticated learning algorithms, can do many things we can’t. Machines can fly at twice the speed of sound, manufacture a silicon wafer, work 24x365 without a break, and perform millions of calculations per second. Very soon, we might add ‘safely drive a car’ to the list of things machines can do. It’s clearly something that humans aren’t very good at – there were 35,092 traffic accident deaths in North America in 2015, and as the NHTSA Administrator, Dr Mark Rosekind said, “...people die when they drive drunk, distracted, drowsy, or if they are speeding or unbuckled.” Have you ever met a drunk, drowsy or distracted machine?
If you need further evidence on machines superiority, then take a look at the analysis conducted by William Grove, who reviewed 136 separate studies of experts vs algorithms, and concluded that on average, the algorithms outperformed clinical experts in 33 – 47% of the studies. Those are amazing statistics, when you think about them. Plus, we’ve reached an age where the power of algorithms is leaping forward, significantly increasing their capabilities as we learn better ways of modelling, and have larger datasets from which the algorithms can learn.
Still not convinced? Look at this detailed report by the Journal of Defense Management on retired U.S. Air Force Colonel Gene Lee, who in 2016 flew a series of flight combat simulations versus an AI “genetic fuzzy” algorithm developed by a doctoral graduate from the University of Cincinnati. Colonel Lee has flown thousands of real life air-air intercepts, and was an Air Force Battle Manager and adversary tactics instructor. The AI algorithm shot him down every time. Lee, quoted in Popular Science said it was “the most aggressive, responsive, dynamic and credible AI, I’ve seen to date.” The aircraft flown by the algorithm was a traditional plane, i.e. constrained by the physics of having a human pilot inside. The latest generation of planes (drones) are designed to have no physical pilot onboard and are therefore lighter, faster, can pull more g-force, and have higher performance envelopes. These drones are still, today, controlled by remote pilots, but I wonder for how long, as the pilots are probably inhibiting the planes performance! It already seems clear that a fleet of fast and highly maneuverable autonomous fighter-drones, controlled by AI, in a mission against traditional pilot-based aircraft, would be a mismatch. The machines would win. [Update: the Russian Air Force has announced plans to have hypersonic swarms of fighter jets in service by 2025]
Why don't we let machines do more?
Given that machines have so many positive traits and several advantages over humans, why don’t we let them take more of the strain in our organizations, and make more of the decisions? Fear is the simplest answer, although we can’t blame Cyberdyne Corporation and their line of Schwarzenegger look-alike terminators for everything. While science-fiction remains a keen advocate of the dangers of giving machines too much power or intelligence, there are very few professional AI researchers who believe in the mid-term that machines will become self-aware and attempt to eradicate humans. Our fear is rooted in more immediate concerns, like what happens if the machines go wrong and damage our business? Or the fear of losing our jobs as our roles are replaced by more efficient machines - a very real concern for large swathes of the population, which perhaps calls for a broader, political solution (back to you Schwarzenegger).
There are, however, huge dangers in procrastination and delay. Technological breakthroughs are an inexorable march forwards, and once new technology is available, someone will use it to their advantage – whether that’s a nuclear weapon, the printing press, a swarm of weaponized drones, or an advanced manufacturing process. Some industries are already taking advantage of new machines, while others nervously watch their competitors but remain locked in a traditional paradigm, waiting for someone to signal the first move.
The Industrial Internet of Things (IIoT) is an interesting case in point. IIoT is founded on a large numbers of machines becoming smarter – possessing new sensor technology, low power connectivity, edge processing capabilities, and access to semantic knowledge and analytic resources in the cloud. Many of the initial solutions built for IIoT have been targeted at predictive maintenance – essentially analyzing sensor data to identify future issues with individual machines (take a look at NarrativeWave who are doing this with turbines or Osprey Data analyzing oil industry pumps, for two good examples). We believe the next wave of change though, the disruptive transformation of IIoT, will occur when organizations allow their machines to collaborate to solve problems. In the natural world, vast numbers of complex problems have been solved via cooperation; whether that’s termites building self-regulating towers to maintain a perfect temperature, bees exploring and exploiting natural resources, wolves hunting in a pack, or humans working in teams.
In the machines vs humans debate, creativity is a key human skill that machines have yet to fully develop, but it's not necessarily the primary difference. Where humans really excel against machines is at being highly adaptable, general purpose creatures - and in teamwork. Most of the machines we've constructed to date, especially the advanced ones, tend to be built for a single goal, and they do it alone. The algorithm might be able to beat Gary Kasparov at chess, or Lee Sedol at Go, but can it get the team together, drive them back to a hotel, open a handful of champagne bottles, and organize a dance game to kickstart the party? Most of the machine-learning (ML) systems powering intelligent machines are use-case specific, “show me a picture and I’ll tell you if it contains a cat”, and while the ML approach can be very general purpose, they still need significant training or set-up to achieve something new. While the quest is on for a master algorithm, that can act and learn more like a human, it may be some time before we reach this unifying goal.
The future today
We don’t need to wait for this next generation of software intelligence, though. We can make huge progress now, because we already have the tools required to enable machines to work as a team. Sure, they'll be able to do more when they get smarter, but ants & bees are not the most cognitively capable animals on the planet, but they can be highly effective by deploying swarm intelligence. In the second part of this blog post, I’ll look at some examples where swarms of machines can deliver more than human experts + cloud based analytics. If you think what teams of people have achieved historically, imagine what our machines might accomplish.