There are three important things you need to understand about artificial intelligence:
- It exists now
- It’s more capable than you know.
- It will replace you (unless you redefine who you are).
I’ll examine each one – it’ll be interesting to see what you think.
It exists now.
Artificial intelligence, as a concept, has been around for a long time. From Hephaestus building the “fighting machines of the gods” to Mary Shelley’s Frankenstein, humans have thought about and created stories around our desire to replace the mighty gods with ourselves for several thousand years.
We see it today with modern franchises like “The Terminator” and “The Matrix”, and less recently with “2001: A Space Odyssey” and “Do Androids Dream of Electric Sheep/Blade Runner”. AI is so popular and accessible as a concept in mainstream media that if you ask people what they think it is, they think they’ll be able to answer.
But here’s the rub – it’ll sound something like “Killer Robots in the Future!”. Rarely do you hear anyone mention the effect of AI on the stock market or the Amazon website. I find this lack of awareness about AI frustrating and frightening because AI exists now, it has existed for decades, and it impacts almost every aspect of our daily lives.
The birth of AI as a research area happened in 1956 at Dartmouth College in New Hampshire as a small but well-funded programme that hoped to create a truly intelligent “thinking machine” within a generation. They failed of course, as creating intelligence isn’t particularly easy. But they laid the foundations. With others following, the acceleration of our understanding and the number of practical uses for AI has increased. And like most technologies, the rate of improvement in AI can be plotted as an S curve.
Technologies tend to have a slow adoption rate early on as a result of the limited capabilities they offer. As the offer increases, so does the adoption rate. Unlike exponential growth, the S curve understands and plots the reduction of the technology’s popularity as we find its maximum potential, or as market forces push funding into new technologies which will ultimately replace it.
A better way to picture the impact of a specific technology on our lives is as a game of chess. At the start of the game, the choices you make are small and of little significance. You can recover from a mistake. But by the time you reach the middle of the board every decision you make will have large significance, and each creates a win or lose situation. Games of chess also follow the S curve.
By the time a technology reaches the lower midpoint of the curve it starts to have a major visible impact, with the velocity of that impact suddenly starting to increase. Artificial intelligence is now at that point. AI is not just here, it’s a punk teenager and pissed off at it’s parents.
It’s more capable than you know.
You interact with AI and one of its children, Machine Learning, every time you use any web-connected device. You do it every time you search, shop online, fill out a form, send a Tweet, or up-vote a comment. It even happens when you buy petrol, turn on the tap, drive your car to work, or buy any newspaper. Every aspect of your life is measured, stored, and used at some point by an algorithm.
Everything you do while living your life is kept as data so that machines can later parse it and use it to identify patterns. The effect is huge.
Roughly 70% of the world’s financial trading is controlled directly by 80 computers that use machine learning to improve their own performance. They can recognise an opportunity and carry out a purchase or sale within 3 milliseconds. The speed at which they operate means that humans are not only incapable of being part of the process, but have been designed out of the system completely to reduce error.
AI is rapidly getting to the point where it is better at diagnosing medical conditions than teams of doctors. Every patient report, every update to a patient’s condition, and every case history is available as digital data to be parsed, analysed and scored in real time to diagnose conditions that require a breadth of knowledge no single person has. In one case from Japan, AI was used to solve (in 10 minutes) a cancer diagnosis that oncologists had failed to detect (the human doctors had spent months trying).
Statistically, computers are better drivers than people are. In the 1.4 million miles Google’s fleet of self driving cars have covered on public roads, “not one of the 17 small accidents they’ve been involved in was the fault of the AI”. There’s the Google car driving into a bus that happened recently, but deep analysis of the incident showed that the bus actually drove into the car. A study by Virginia Tech showed that Google’s autonomous systems are 2.5 times less likely to have a car crash than a human. Given some of the behaviour I’ve experienced on the roads, I think this is a pessimistic number. AI is also being used to fly planes, with pilots of the Boeing 777 on average spending “just seven minutes manually piloting their planes in a typical flight” (anonymous Boeing 777 pilot). The United States and British governments have had fully autonomous drones flying for well over a decade.
Computers are now writing articles, poems, and even screenplays. Netflix’s now famously complicated taxonomy may have been put in place by people, but it’s machines that use it to work out what the next hit TV show will be. Associated Press uses AI to deliver over 3000 reports per year, while Forbes’ earning reports are all machine generated. These aren’t lists of numbers – this is long-form copy. Many sports reports are now written using AI, and they are published instantly as soon as the game ends. Before the team has left the field, the articles are being read. A study by Christer Clerwall showed that when asked to tell the difference between machine or human-written stories, people couldn’t. I mean, can you tell which parts of this blog were written by a machine?
Computers are better at designing their own components than people are. In the 1990’s Dr Adrian Thompson of Sussex University wanted a test on what would happen if evolutionary theory was put to use by machines in building an electrical circuit. The circuit was simple – all it had to do was recognise a 1KHz tone and drop the output voltage to 0, or for a 10KHz tone an output of 5 volts. An algorithm iterated over 4000 times before finding the best possible circuit. The circuit was tested, and it worked perfectly. The surprising thing though was that nobody could explain how the circuit worked, or manage to produce a better one. This experiment has been repeated many times, with more and more complexity introduced, and each time the machines make parts for themselves better than people can.
Computers are creating art, helping to cure the sick, improving themselves, and taking care of complex or monotonous tasks. We let them drive us, shop for us, fly us, and treat us. We let them form opinions for us, and let them entertain us. Where do people fit in?
It will replace you (unless you redefine who you are).
In 2013 it was estimated that 47% of job roles in the United States were at risk of being replaced by automated systems (See Rise of the Robots by Martin Ford).
And a lot has happened in 3 years.
While your interactions with AI can make your life easier and far more pleasant, they are designed to achieve something more – every time you do something that can be logged and compared, you are training the AI in human behaviour. But we can’t stop living our lives, so how do we stop the machines taking over?
If we change our point of view, AI can become our best ally and our most powerful tool. To explain properly, let’s take a look at this from the point of view of a web agency like Deeson – take a seat, because this is going to sound a bit like a Hollywood film script.
Each of the steps we take to create a site (discovery, design, testing, build, more testing, etc) follows a series of repeating steps, and as we’ve already learned, AI is very good at doing anything that people see as repetitive. If you can define each step as a model, with a standardised input that gives rise to a required output, you can automate the process.
A real-world example would be the first stage in graphic design, where we put a hierarchy around available content so users can find what they need. In the time it takes for a UX Designer to lay out a wireframe and get client feedback, an AI system could design and iterate millions of layouts, as well as test them against virtual users. Yes, you can potentially see the first 1,999,998 layouts being broken, but with each iteration the solution becomes better. This happens now with a tool called multi-variant testing – AI just increases the scale and the speed.
With content hierarchy taken care of, we can now start graphic design. Designers prepare what are called “Style Tiles”, which are the basics we use to ensure the client’s brand is digitally ready, such as colours, fonts and tone of voice. These are the building blocks of the site. If we were using AI they would, along with the content hierarchy created above, be some of the properties in our input model. With these in place we can repeat the same process as above, iterating as many times as is required until we have a visual design which is on brand (defined by the tiles), while maintaining ease of use (defined as a metric).
With layout, colour, and typography resolved, content is next. This is the easiest problem to solve. AI is already writing content – it’s already able to understand the content requirements of people through the analysis of what we do, where we go on a site, what we say on social media, and through the on-going testing of current site content. If the system knows we are building a site with a focus on coffee, and it knows what people on the web are talking about, it can generate stories and publish them within the design framework created during the previous step. Natural language systems do this already, all the time. The parts of creating a website that humans find the most difficult have already been solved by AI researchers.
Once the content is in place, housed in an on-brand and functionally useful design, which itself sits upon a semantically rich content hierarchy, what’s left? For the majority of websites, the answer is very little. AI systems (which thegrid.io claims to incorporate) are changing the way we think of the web, and web agencies need to take note.
Using AI to design and build a website from nothing but basic guiding principles seems like science fiction, but I know from personal experience that we can do it. My experiment involved the first version of the Science Museum’s Information Age microsite, where algorithms chose the content, designed the layouts, and published the site.
The meaning for the content editor’s job role had changed. Instead of deciding upon a narrative, finding the right objects from the collection, building the associated content to support the narrative, and then writing the final stories, editors become parents to the system. They took on a coaching role. They kept the system fed with objects and content which the algorithms would test live through a series of experiments. Where engagement was high the content remained, where engagement was low it did not. The editors had to reimagine their roles, but they weren’t replaced – rather augmented. They became capable of, in collaboration with machines, writing as many stories as they were capable of dreaming up.
The most important aspect of AI is that we must understand. If we embrace and learn how to use it, we advance ourselves. If we ignore it or bury our heads in the sand, it will overtake us. The future of AI in our industry will be defined by those of us who adopt it, and I sincerely hope you do too.
Credit where it’s due, I stole the headline image from: https://www.linkedin.com/pulse/attack-so-killer-robots-business-hiring-bots-katie-bazemore