Artificial Intelligence (AI) and machine learning has been in the news a lot over the last few years. We are astonished and amazed self driving cars and other achievements by Google’s Alpha Go and appointment tool demo. We are also concerned, rightly so, about the accuracy and safety of autonomous systems which may have been trained with biased or flawed data.
However, claiming AI to be the world’s biggest, most immediate problem seems like a bit of a stretch. Most argue that climate change or various political and social issues are way up there, and even far more important. This article is about trying to change your mind, by making the case that AI, is the world’s most immediate problem, which, if left unchecked, could invite catastrophe.
It’s not all bad news, so let me start with the good news. Take comfort that this is an issue which affects every human on the planet indiscriminately – this is your problem too. If we can all agree on the problem, there is hope that we might change course and do something about it. The good news doesn’t stop there! AI is going to make our lives immeasurably better in ways we haven’t even dreamed up yet. However, beware, the allure of massive short-term benefits may possibly blind us to the longer-term downsides that we are going to cover here.
Starting Simple
To get started, but also frame this whole argument, I want to make one simple, incomplete, economic argument, which will really help if you buy in.
A company which produces the same goods or services at a lower cost and/or a higher quality will generally outperform competitors.
It isn’t perfect and we could certainly add a number of caveats if we had time, but, it is fairly reasonable argument right? It also goes some way to explain the benefits of capitalism like technological innovation and the industrial revolution.
The Industrial Revolution is ancient history that we don’t think about much. The results of massive technological innovation allowed our species to go from 95% to less than 5% of the population producing the food for the nation by replacing horses and humans with mechanical muscles to do much of the laborious work. We even generated vast surpluses we can trade abroad too. As a result, products and services of all kinds came down in price and in turn, humans could explore other creative endeavors.
Humans
Human ingenuity, grit, adaptability and innovation has kept us busy and productive too, as well as improving our well-being throughout history. But alas, our usefulness may be coming to an abrupt end. We are largest expense associated with most businesses, and until recently, a necessary expense. Historically, machines have not been able to do what humans do extremely well, and well, this is where the bad news starts. We humans have annoying needs like sleep, food, paying bills and other demands like ‘working conditions’ and ‘time off’. This is fairly inconvenient for businesses, of which has the job of creating profits. Machines on the other hand, don’t need to sleep and cost only pennies in electricity.
Fast forward to today, the human advantage is diminishing. Cue: the rise of the machines. Everything is about to change…
Machines
Now might be a good time to get you caught up on what machines can already do – right now. I want you to keep them in mind, because I’ll be relying on your imagination to apply these techniques to solve all kinds of problems.
Machine learning is one of the most prevalent AI tools we have today. It emulates one of the most common types of cell in the human brain, the neuron. Of course, the human brain is a complicated biological machine, and we still aren’t anywhere close to understanding everything about it. Every human has around 100 billion of these neurons in your brain and from what we understand, these neurons form connections as we learn, firing the correct synapses when you interpret data from the environment, so hopefully you do something useful. To give you more visceral example: An infant brain isn’t well developed, so many of these connections we take for granted haven’t been made. As a result, an infant will fall down when learning to walk, until they train their neural networks, forming all the relevant connections in their brain.
As I said earlier, there is still a lot we don’t know about the human brain. We don’t yet need to worry about conscious AI, or Artificial General Intelligence (AGI). However even without super-intelligent AI, these techniques, among others, are already extremely effective at solving repetitive routine tasks.
Classification / Pattern Recognition
Classification is pattern recognition and information processing is intelligence. As humans, we are exceptionally good at classifying things. Our brains automatically find patterns in everything, even when they aren’t there (optical illusions). When humans classify new things they haven’t seen before, we are remarkably good at creating models of these ‘new object’, in fact, we may only need to see it once in order to recognize it again in future. It’s just like when you learn a new word, and then you hear it again on TV or in public. Those new neural networks will fire when they recognize something you have learned, leading to the thought to arise in the forefront of consciousness.
Machines are getting better at the task of classification. Though we have had some innovations in ‘single shot detection’ (which we aren’t going to cover today). Generally speaking, machine learning, needs huge numbers of labelled examples of whatever you are looking to classify. Depending on what you are trying to classify, it can be time consuming and tedious to generate large datasets to classify a certain patterns. For some problems, huge datasets already exist, ready and primed for machine learning, such as Mozilla Common Voice project.
While humans are good at classifying some information quickly, and easily, they are absolutely terrible at classifying other types of multivariate information without a great deal of effort and research. Now is the moment to understand the importance of the generic nature of solving classification problems. Classification is the ability to tell people apart. The difference between large and small, cars and bikes, conservative or liberal, gay or straight, happy or sad. A human might be able to recognize a bike or car with ease (and now machines can too), but humans cannot easily determine if you are gay or straight just by looking at you, but machines can. How? Mostly due to gigantic labelled datasets from the data we are all continuously leaking or voluntarily giving away online.
Here is where things are getting a bit scary. Companies like Facebook & Google can easily classify your sexual orientation or politics without you telling them. How? Enough people have told them already through their profiles, likes, interests, affiliation and searches. The data generated by these labelled individuals can be used to infer lots of information about you from the data you generate.
We all like to think we are special and unique and in some ways we are, but in other ways we all fit into overlapping characteristics which can easily be predicted. Today, aforementioned companies use this technology to classify you to send you ‘targeted’ ads. But it is very arguable at this point that this could be considered a form of user manipulation. Is it wrong to knowingly show someone classified as having a gambling habit an advertisement for a casino?
Computer Vision
Seeing is believing, so now we’re now going to demonstrate some of the building blocks of AI that we can piece together to create complex systems. One of these building blocks is computer vision, essentially, another classification problem. The human eye captures light, and our brain interprets this input and if we can all agree on what our brains are hallucinating, we call it reality. It’s safe to say the human brain is amazingly complex and usually gets it right, though not always:
“Cafe Wall Illusion” The lines are straight. Your brain thinks otherwise.
Obviously being able to correctly interpret visual information is an advantage to any organism, but the machines are catching up fast. Services like Amazon Rekognition, and Google AI, as well as the ability to train your own computer vision models, already offer some very useful, although imperfect, capabilities. Machines can quite readily detect objects in images and video or describe a scene.
Object detection in images and video.
Automatic description of a scene.
A lot of these machine learning models are imperfect in certain scenarios, but in some cases they actually surpass human ability. It’s with these sorts of combined capabilities which is where things start to get really interesting. If you can identify and track objects in real time, you also might also be able to perform some action on them (think robotics).
Things are going to get considerably better.
The picture that I’m trying to paint is that AI is a revolutionary change on par with the industrial revolution. There are so many ‘classification’ type problems which we can solve using machine learning, everything from medical breakthroughs for better diagnosis, to detecting sharks with buoys. Nearly every business today has computers systems integrated to operate optimally, and as such, machine learning can benefit every sector and industry, especially those who already have amassed vast amounts of data.
The Warning
This is the part where I try and convince you, given everything we have already discussed, that AI could be a problem for us all. I will start by saying that I am open to being wrong and that part of agenda for putting this out there is to try and identify any compelling perspectives which I may not have considered.
The news cycle tends to be dominated by climate, political and social issues and while these issues are all important, the AI revolution is happening now, and will hit faster than we expect. My base case for massive economic disruption is less than 10 years, with some effects being felt right now and increasingly every year going forward. If only very few of us are employable through no fault of our own, today’s headline issues are going to start to look very trivial and we will wish that we had focused on these issues earlier.
The risks go much further than jobs too. As the last 2016 US election cycle demonstrated, we also have to worry about manipulation of populations across the world. Companies like Google and Facebook understand all too well that our decision making and beliefs are heavily influenced by what we experience. These companies are increasingly tweaking our search results, news, social media results to show us what they think we should see by using algorithms with no transparency.
On Jobs.
The Bank of England warned that over 50% of jobs are vulnerable to automation due to the routine and repetitive nature of many jobs. You may think that AI will create new jobs, but unfortunately that doesn’t seem to be the case either – nearly all of the job types we have today have existed for over 100 years and the newer jobs like software engineering are way down the list of jobs by type.
Companies are investing in AI at an ever-increasing pace and have incentives to automate because the largest expense for almost every business is people. Prior to this AI revolution, developed nations relied heavily on outsourcing their workforce to nations with lower labor costs, or importing low-skilled labor to do the jobs which their citizens just won’t do. These jobs are going to disappear rapidly when machines can do the same work for a fraction of the cost.
I like to use the car-sharing, gig economy example here to demonstrate how wildly unprepared we are. Transportation jobs number roughly 4 million jobs in the US. Both Uber and Lyft have self-driving car initiatives which can only be designed in order to fire their drivers yet still provide the same service. Of course, I’m sure they will deny this, or proclaim that this future is very far away. But very few of the drivers I have spoken to have this on their radar.
Arguably, autonomous driving is probably one of the harder problems to solve, especially when compared to boats, planes and trains which have fewer variables to consider when navigating the real world. But we’re starting to get a sense of the sorts of jobs at risk of automation.
AI doesn’t have to be conscious to be dangerous
AI doesn’t care who uses it or how it’s used. Repressive regimes could use it to target and identify dissidents, ethnic minorities or sexual orientations. Powerful AI weapons can easily be conceived by combining existing technologies today. E.g. A dangerous combination of autonomous flying drones, guns, object detection classification is possible today, if not already in the works by every capable nation. To my mind, I could easily approximate the steps it would take to create such weapons.
Society, capitalism and western civilization has largely depended on individual autonomy and meritocracy. For the longest time, this has been a good thing. But in a world where more and more jobs are going to be automated by machines, imposing an ever increasing strain on those competing in the job market. While it’s true that we have recently hit record unemployment numbers, largely due to massive economic stimulus and low interest rates. Once unemployment peaks – I fear we will never, ever, hit these levels of unemployment ever again.
This is going to happen in a remarkably short time-scale as the rate of innovation is spurred by faster chips and breakthroughs in AI techniques which increase accuracy, reduce training time, or amount of data required to achieve low error rates.
This is not a case against capitalism or western society. It’s also not a case for for any other ‘ism’. To be fair, I don’t have the answers either, but as I do believe this to be the most pressing issue for us all. It is my hope that some people smarter than me can come up with solutions.
Some have suggested a ‘Universal Basic Income’ as a solution with the most traction. Though I do have many concerns with this concept. For those of you who don’t know what it is, it’s basically free money for everyone, every month, unconditionally. My major concern with this is solution is that it is the only a patch to the problem and does nothing to solve the dignity and self-worth we attribute to work.
I’ll conclude my argument. With every passing day, machines are able to complete tasks better than any human on the planet will ever be able to. AI-based algorithms are constantly competing for your attention and attempting anticipating your every need. We have absolutely no oversight or transparency into how these machine-learned models make decisions, nor their impact on society. We have absolutely no plan to mitigate massive job loss on a scale worse than the great depression.
