It’s a bright day in Afghanistan. A small group of mujahidin is trekking through the mountains. They carry their Kalashnikov rifles on their shoulders, but they are not especially worried. The nearest enemy unit is several hours away. So high in the mountains, they would see them coming from a long distance. There are other dangers, though. Several thousand feet above their heads, a small unmanned plane is cruising unseen. After only a short observation period, it fires a Hellfire missile at supersonic speed. The militants have no time to react. A few heartbeats later, a massive explosion blows them to bits.
Does it matter for our unfortunate mujahidin’s if a drone or a killer robot fired the missile? They are dead either way. The only difference is who pulled the trigger: a human or a machine?
If you are looking for ways to use AI profitably in your business, that distinction should matter to you.
- AI Automation
A killer robot is a very straightforward example of automation. Human makers design a machine according to a set of specification, then gets out of his way and let it do its bidding. In the case of AI automation, the device makes its own decision. Human beings don’t necessarily wholly disappear from the picture, but their roles are reduced to training, maintenance, and monitoring. Killer robots replace the soldiers, but the drill sergeants, mechanics, and generals are still needed.
Killer robots are easy to explain but harder to build. Despite years and billions of dollars of research, no military has deployed full lethal autonomous weapons — the technical name of killer robots — in any meaningful capacity as of 2019. The calls to ban those types of armaments may have slowed down the R&D. However, I’m prepared to bet that the murky ethics of killer robots are not the main impediment. It’s more likely that technical and economic challenges are preventing fast development and industrialization.
The terminator delusion
There is a common misconception that AI automation means superhuman performance. I call it the Terminator delusion. In Sci-Fi stories, AI and robots are often all-powerful, near-invincible weapons of mass destruction. It’s been a common theme in literature that dates back from the legends of the golem and Mary Shelley’s Frankenstein. Popular authors portrait humanity’s creations as overtaking us with their superior strength or intelligence. It’s fear of technology, taken to the most extremely paranoid conclusion.
The few high profile victories of game-playing AIs have alimented that delusion. Popular media sees the victories of Deep Blue, Watson and AlphaGo over world champions as milestones toward the inevitable machine take-over.
However, the reality of AI software in 2019 is that superhuman performance is the exception, not the rule. AI systems today are “narrow AI” with a severely limited range of capabilities. In another story, I’ve compared the intelligence of these systems to that of a dog that is taught special tricks. Another way to look at it would be to think of narrow AI robots as lobotomized humans. Take someone with the ability to perform some task well — for example, play chess — and destroy all the parts of her brain that are “unnecessary.” Personality, sense of humor, drives, and desires, everything is stripped away. The unfortunate lobotomized person might still be able to perform her one task very well if nothing changes. But her inability to learn quickly, her lack of creativity and adaptability will handicap her.
Today’s killer robots would not be Terminator-style near-indestructible super soldiers. The Walking Dead’s zombies would be a better analogy. Think cheap and expendable dummies.
How can you use AI innovation for your own business? How would you use your killer robots?
The VC Michael Harries has a helpful way of thinking about the potential of AI automation: “How might 1000 smart interns change your business?”. Think of what you could do with so much cheap labor available. Note that 1000 neurosurgeons are not an option. We’re talking about giving you access to a lot of cheap, barely legal, inexperienced interns distracted by an overactive libido. Their productivity can be extremely beneficial, but you might think twice about giving them the password to the bank account before going on holiday.
Many companies have followed the 1000 interns route with success. At Facebook, one of the first widespread use of AI automation was a function to tag people’s faces in pictures automatically. Marc Zuckerberg didn’t hand over his job to an AI. He didn’t roll-out AI to take care of any essential or strategic task. Instead, he let the machines do a job that a young intern could have done: tagging people in their party photos. That might sound trivial and low value. Each tagged picture might only add a fraction of a cent to Facebook’s bottom line. At scale though, those cents add-up and they start to matter.
If you can’t think of a task where 1000s of interns would help, it means AI automation is not what you need.
2. AI Augmentation
Drones are different from killer robots in that they still have a human pilot, who is notably tasked to take the kill decisions. We can think of the drone as an extension of the human body. The machine augments the pilot’s capability using AI, robotics, and, of course, plane technology. Using the drone, a human operator — alone or in a small team — can target and kill enemies from 1000s of miles away, with incredible power and precision.
If you are after superhuman performance, AI augmentation does the job. Reaper drones are an excellent example of AI augmentation. The machines might not be fully-fledged killer robots, but they do include their fair share of automation and robotics. The human pilot is still in charge. The drone can’t pull the trigger on its own. The pilot still has to decide herself when to fire. However, her overall performance is increased to superhuman level thanks to a whole host of robotic processes that control flight function, targeting, and more.
Using technology to augment human performance is nothing new. The first scribe who used a clay tablet to write down his accounting ledger was using technology to expand his mental performance. Much of today’s information technology follows in the footstep of that first, exceptionally powerful idea.
Drones are in some ways less impressive than killer robots. Those are just an overpowered version of the remote control toys you had as a kid, right? However, today’s drones are many times more capable that even some of the deadliest robots dreamed of by Sci-Fi writers. In the 1984’s movie, the Terminator could walk and — almost — talk like a human. His equipment was relatively puny guns purchased in a Californian gun store.
In contrast, a USAF MQ-9 Reaper can carry 3,800 pounds of ordnance at over 300 mph, 50,000 ft high in the sky, for 23 hours, over a range of 1,150 miles. The Extended Range version has a sensor ball that can automatically identify threats, track 12 moving targets at once and fires three missiles per second. The Terminator is hopelessly outgunned.
Gary Kasparov, who was chess world champion in the 1990s, owes much of his notoriety for having -narrowly- lost to IBM’s DeepBlue in a highly mediatized chess game in 1997. DeepBlue was, of course, a killer robot for chess: it’s an example of AI automation.
When AI research started in the 1950s, solving chess was seen a highway to solve human intelligence. Chess, with its demanding strategy and tactics, was seen as the pinnacle of human expression of intelligence. That view turned out to be wrong. Chess grandmasters are not, in fact, necessarily smarter than other people. The game is also easier for machines than it is for humans. A program can memorize 1000s of openings in a simple look-up table. After the opening phase, when the software is “out of book,” it plays by finding the best move for each turn. There is no consideration of past moves or overall strategy. Playing at grandmaster level chess becomes a straight forward search problem, solved by evaluating millions of moves per second. Far from being the pinnacle of human intelligence, chess is uniquely suited to computers’ strengths.
Since the 1997 DeepBlue-Kasparov game, no human player can rival the best chess playing software’s. What is little known, however, is that after his defeat, Kasparov started organizing tournaments with humans playing together with chess computers as a team. The question was, who’s stronger: a human grandmaster, a chess computer, or a side of human player and computer? The result, which wasn’t evident at the time, is that teams of humans and machines outperform humans and computer only competitors. In fact, even a relatively weak human player can significantly improve on the performance of the chess-playing algorithms, provided he or she is using a robust decision process. Both AIs and humans have weaknesses, but in the ideal conditions, they complete each other well.
The takeaway from the chess world is: if you want super-human performance, use AI to empower humans. Drones are stronger than killer robots. Keeping the drone’s pilot in the decision loop increases performance over pure automation.
The augmentation route can also have tremendous value to your business. IBM’s Watson medical diagnostic tools are a famous example of AI that is used to augment the doctors’ performance. In the medical world, there can be no question of automating away the physicians. However, intelligently deployed AI can significantly improve the accuracy of diagnostics, provided that trained medical practitioners stay in the loop and take the ultimate decisions.
The underlying technology for both automation and augmentation is the same. Both can be done with success. However, the systems are used very differently. The differences in usage have, in turn, a lot of impact on the design and the engineering options. So think carefully: would your business benefit more from automation or augmentation?