When even Elon Musk – normally a champion of the human ability to improve its condition through material progress – backs the robots, we know we’re in trouble. But barely a year ago, back in 2017, he became fear-monger-in-chief of the artificial-intelligence apocalypse. “What’s going to happen,” he said, “is robots will be able to do everything better than us … I mean all of us.”
But a lot can happen in two years. A report just published by PriceWaterhouseCoopers should go some way to calming such hysteria. It argues that AI will create slightly more jobs in the UK (7.2 million) than it displaces (7 million). So rather than lamenting an apocalyptic tomorrow in which we are ruled by our robot overlords, a more useful way to think about the future would be to consider how we can interlace the strengths of machines with those of humans.
John Giannandrea who left his role as Senior Vice President of Engineering at Google, to head up AI at Apple, is already doing so. “There’s just a huge amount of unwarranted hype around it right now,” he says, “[much of which is] borderline irresponsible.” We shouldn’t be using it to match or replace humans, but to make “machines slightly more intelligent — or slightly less dumb”. He isn’t dismissing the potential of computers to radically alter the way we work, but is thinking about the ways it will do so in a slightly more nuanced fashion.
He knows that the better we understand the differences between the way people think and the way in which machines calculate, the better we can assess how to work with them. For example, unlike machines, humans typically lean on a variety of mental rules of thumb that yield narratively plausible, but often logically dubious, judgments. The psychologist and Nobel laureate Daniel Kahneman calls the human mind “a machine for jumping to conclusions”.
Machines using deep-learning algorithms, in contrast, must be trained with thousands of photographs to recognise kittens— and even then, they will have formed no conceptual understanding of cats. Small children can easily learn what a kitten is from just a few examples. Not so machines. They don’t think like humans and they can only apply their ‘thinking’ to narrow fields. They cannot, therefore, associate pictures of cats with stories about cats.
But, counterintuitively perhaps, the tasks humans find hard, machines often find easy. Cognitive scientist Alison Gopnik summarises what is known as Moravec’s Paradox: “At first, we thought that the quintessential preoccupations of the officially smart few, like playing chess or proving theorems — the corridas of nerd machismo — would prove to be hardest for computers.”
But as we have discovered, this isn’t true. Computers do these terrifically complicated things effortlessly. As Gopniks says: “It turns out to be much easier to simulate the reasoning of a highly trained adult expert than to mimic the ordinary learning of every baby.”
We are bound to learn more as we work out how our strengths can work in concert. Humans, for example, will employ inspiration, judgments, sense and empathy; computers will bring brawn, repetition, rules adherence, data recall and analysis.
The psychologist and computer scientist JCR “Lick” Licklider, mentor to John McCarthy who coined the term Artificial Intelligence in 1955, predicted this future harmony in 1960. Rather than speculate on computers achieving human-style intelligence, Licklider argued with remarkable prescience that humans and computers would develop a symbiotic relationship, the strengths of one would counterbalance the limitations of the other.
“Men will set the goals, formulate the hypotheses, determine the criteria, and perform the evaluations. Computing machines will do the routinisable work that must be done to prepare the way for insights and decisions in technical and scientific thinking. … the resulting partnership will think as no human brain has ever thought and process data in a way not approached by the information-handling machines we know today.”
It’s a far more productive guide to the future than alarmist preditions of ‘super-intelligent’ AI. The advent of AI is simply the latest in many phases of automation, each of which has begun with fear and ended with more jobs, economic growth and prosperity. Indeed, the history of automation since the industrial revolution shows us that the introduction of machines reshapes jobs, rather than replacing them. They take on the mundane tasks, as humans move on to more complex – often more highly valued and meaningful – work. New technology tend to eliminate tasks, not work.
The goal now is subtly shifting from building machines that think like humans, to designing machines that help humans think and perform better. Most work, after all, is comprised of a mix of tasks: some of which are better suited to humans and some of which could one day be done better by machines.
As the capabilities of machines grow, managers will redesign work to take advantage of the strengths of both their human workers and their machines. Last month, Nestlé announced a new automated plant in the Midlands in which German logistics giant DHL will be installing collaborative robots, or “co-bots”, that will work alongside humans to pack goods.
While on one level, AI is but the latest phase of a centuries-old process of automating human labour, on another, this new breed of machines can sense, learn and adapt in ways that far surpass previous technologies. And as a result, it will ask new questions about human exceptionalism and how work is best done.
Let’s not forget that the heavily automated production of the Model 3 Tesla ran quickly into delays prompting Elon Musk to make an interesting volte-face. “Excessive automation at Tesla was a mistake,”he wrote in April this year. “To be precise, my mistake. Humans are underrated.”
Five ways humans are collaborating with intelligent machines
1. Assigned work
Certain tasks in a human workflow are outsourced to a machine. The machine completes the task unaided, with varying levels of instruction.
- Industrial robots welding and spray painting car parts on a production line, while human workers perform other tasks like fitting the IP panels and custom parts. Industrial robots labouring 24/7 in ‘dark factories’
- Autonomous vehicles in fully autonomous mode on a motorway
Decision making processes are automated, but under a human eye. This mode requires the machine to be aware of and communicate risks and unknowns to human users.
- Airline flights in which the pilot intervenes only in certain circumstances
- Autonomous vehicles, which require ‘driver’ oversight.
We will increasingly live and work alongside intelligent machines, sharing the same spaces, but focusing on separate task-flows. Machines in these scenarios must be able to effectively negotiate shared space and anticipate human intent.
- Pedestrians sharing paths with delivery robots
- Warehouse staff working in parallel with and alongside robots
Machines that will help us perform tasks faster and better. They support particular tasks in human workflows, and will excel in on discerning human goals and learning their preferences.
- Writing assistants that suggest words and how to improve text
- Exoskeletons assisting factory workers in strenuous work
This emerging mode of collaboration is a highly interactive and reciprocal. People input strategic hypotheses and the machine suggests tactical options.
- Chess teams comprised of humans and computers, which beat the best computer-only systems
- Adobe Sensei which automates tedious design processes, and generates options, giving the designer more time for creative tasks