The future of work is greatly unknown. We’ve already seen dramatic changes in the role of human-power in the workplace. We’ve seen the emergence of cashier-free supermarkets with Amazon Go, and drones and driverless vehicles threaten to completely overhaul logistics.
Soon millions of jobs, from accountants and stock brokers to estate agents and your local doctor, will be threatened by the emergence of machine intelligence and robotics.
There will be a jobs apocalypse where a huge percentage of the jobs we currently do will be replaced by smart, inexpensive robots, self-driving cars, trucks and planes, and even cameras and Smart TV screens.
At Microwork we’re preparing for the evolution of work by integrating humans with machine intelligence and we imagine a future where, for example, even doctors aren’t confined to one hospital, but could be doing work remotely with hundreds of hospitals and with the support of machine intelligence.
But what exactly is Microwork, and what do we do?
“Microwork creates datasets to train machine intelligences using intellectual services by means of human computation via human machine interfaces.”
To understand this even more, let’s break it down a little.
At Microwork we personally like to make the distinction between Artificial Intelligence and Machine Intelligence. Machine intelligence isn’t an Artificial Intelligence. It’s a real intelligence, but made from different components than those that we’re used to.
If we take an apple and an orange, both are very different. One is citrusy with a sharp taste and broken into little bite-size segments that burst when you put them in your mouth. The other has a firmer body, a milder taste, and a completely different texture. As the old idiom goes, you can’t compare apples to oranges, but we agree that both are fruit.
Machine intelligence will of course be a completely different type of intelligence than human intelligence. We’re not training it to be human, we’re just training it to be intelligent.
At Microwork we use a process called Deep Learning to train our machines. Essentially we train them in the same way children are trained. We point to a dog, and call it a dog. We point to a cat, and call it a cat. And we continue. The thing with data is, the more you’ve got, the more accurate your learning becomes.
If we point at enough iterations of dogs, then you can also start to learn breeds, body positions, coat markings and perhaps even age. The more things we point at, the better the machine’s understanding of dogs get, until ultimately it can even begin to infer new things we’ve not even told it.
In philosophy, David Hume rationalised that humans can combine simple concepts to form brand new, complex ideas. We know what elephants look like, we know what the colour purple looks like, and so it’s not outside of our realm to imagine what a purple elephant could look like too.
It’s very plausible that machine intelligence will also be able to do the same and infer answers that it’s never encountered before, and who knows, perhaps even things humans haven’t imagined before.
So we’re collecting an ever-increasing amount of data, but we also need to ensure this data is organised. If you had a collection of 100 000 random images, we could refer to that as data, but it remains very unusable in this format.
However if we take those 100 000 random images, let’s say 10 000 are of cars. Of these 10 000 cars we can distinguish 100 different makes and models. Suddenly we’ve got much more structured, usable data. Once you gather that data together, we have a dataset, and we use these datasets to train our machine intelligences.
We believe that the future of work relies on machine intelligence and that for machine intelligence to be as accurate and effective as possible we need a lot of data. We also know that structured data is more usable than random data, but substantial amounts of this high-quality data is expensive. So where can we get it from?
This is where Microwork has the last piece of this jigsaw puzzle. On Microwork we post jobs where machines can use human knowledge in order to learn even more. We integrate human processing with machine data collection in the form of jobs.
But how does this look in reality?
Let’s take the example from earlier. If we need to research cars, we can publish a job where people take pictures of cars. Each picture is tested for quality assurance and if the person who took the pictures fulfils the task they are paid for it.
Meanwhile, our data grows and becomes more and more structured and effective, and this can be used as fuel for machine learning.
Imagine if we took this a step further. We envision a world where an x-ray is completed at the hospital, and before the local radiologist even looks at the results, the data is shot through a system and analysed remotely. These remote doctors could use their specialised skills to analyse the x-ray before sending it back to the local hospital.
This extra layer of external knowledge is both additional quality assurance, but also plays an important role in machine learning. We’ll be training machines to detect and understand x-rays, while providing these remote, specialised doctors with work they could essentially do anywhere without being confined to one hospital. With every iteration, a remote doctor gets well-paid work at something they specialise in, the machine becomes smarter, and eventually maybe even better than humans at analysing x-ray results.
At Microwork we feel we’re on the brink of two giant changes in the world we know. Just as millions of us are coming away from location-dependent careers towards location-independent, project-based work, we provide solutions that follow this new trend of human work. Plus we’re training machine intelligence to work alongside, and even inside, humans to make life easier, smarter and more efficient.
We’re building the future.