Algorithms are everywhere nowadays. Facial recognition technology allows us to tag friends in social media photos, chatbots answer all of our questions on vendor sites (well, most of our questions), and navigation apps tell us which roads to avoid when the traffic is bad.
This trend is not going away anytime soon. In fact, Google CEO Sundar Pichai believes that Artificial Intelligence (AI) will have a bigger impact on human development than the discovery of fire – that’s nothing to be sniffed at.
Here are some of the biggest developments in the field of AI in 2021, as well as a look at their potential applications in the real world, and what to expect from AI in the year to come.
Restoring lost art
When the Nazis burned down the Immendorf Castle in 1945, they assumed that nobody would ever see the three Gustav Klimt masterpieces locked inside. Now, with the help of AI, experts have attempted to recreate something very close to the destroyed originals.
The hardest step was trying to colourise the black and white photographs of the paintings in a manner consistent with Klimt’s distinctive palette. The Google Arts & Culture Lab, which led the project, fed an algorithm some 92,000 images of paintings, teaching it to recognise discrete objects in works of art, and to understand the rough colour schemes and textures regularly attributed to different artistic subjects.
Next, the algorithm was trained to imitate Klimt’s own radical colourisation technique. Using 80 different pictures of Klimt’s more vibrant paintings, it learned to emulate the skin tones, flowing garments, and swirling skies which make Klimt’s work so recognisable. After six months of work, the AI technology completed its own colour representations of Klimt’s lost Faculty Paintings. Whether they are perfect representations, nobody will ever know – it’s very unlikely. What is more certain is that computers now have the power to revive lost masterpieces in a way that can evoke some of the same wonder as the originals.
This isn’t the only AI project in the world of art. The Rijksmuseum in Amsterdam used AI to recreate the missing edges of a Rembrandt masterpiece, The Night Watch, which was trimmed down significantly in 1715 to fit a new location following a move.
And with computer scientists now working to develop the ability of AI not just to imitate but to create its own work, the possibilities are endless. AI-generated art is appreciating in value, and it may not be long before computer-created paintings hang alongside the masterpieces of Impressionists and Dutch Masters in the most prestigious galleries.
Writing…pretty much anything
Thanks to a process known as Natural Language Processing, AI can now complete almost any language task, from translating English into computer code to writing poetry. It can even write convincingly in the style of Kafka!
The most advanced interface when it comes to writing in English is a deep-learning neural network called GPT-3, which is owned by Microsoft, but which other users can access too. Tools like Sudowrite are plugged into GPT-3 and allow writers to input a short prompt (the beginning of a story or an essay, for example) and press a button, which tells the computer to generate the next part of the work.
When 80 people were asked to judge if short articles were written by humans or GPT-3, they guessed wrong 48% of the time, which is only slightly better than if they had guessed randomly without looking at the texts.
With around 175 billion “parameters”- variables and datapoints used to process language – GPT-3 is one of the biggest supercomputers in the world. It can be used for a range of applications, from word processing to designing games and imitating correspondences with historical figures.
The next version of this interface, GPT-4, could be created as early as next year. Some estimate that it may contain up to 100 trillion parameters.
All of the above is very exciting… but also slightly sinister. To the overactive imagination, it might conjure up apocalyptic scenes of superintelligent computers training themselves to enslave puny humans. But in other areas, AI itself is an invaluable tool for keeping people safe from the harmful effects of technology.
Smart algorithms have learned to track network traffic and spot patterns suggesting that a particular user might have nefarious intentions. From malware to phishing attacks and hacking, AI technologies can spot a range of online threats across millions of data streams in a matter of seconds. It can even predict some risks before they happen.
This technology still has a way to go before it becomes fully-fledged. Sophisticated attackers can occasionally get past such algorithms by working out which data they use to detect harmful behaviour and avoiding setting off these triggers. Some of the most exciting AI advances of 2022 are likely to be in this area.
Coding for dummies
GitHub’s Copilot tool writes code for you if you give it some short excerpts as a prompt. It will even generate its own functions for you if you give it clear instructions. If you want a tool which can add numbers together, you merely need to tell GitHub to do this (in roughly the correct format), and it will spit out a function for you.
In this way, Copilot helps to make coding more accessible. This trend is set to continue as barriers to coding are lowered and more people come to appreciate the widespread applications of AI. One of the winners of Silicon Valley’s Synopsys Science Fair 2021, for example, was a sixth grader who built an AI system to reduce crashes at stop signs.
Computers on the edge
A recurring theme with AI is that it still needs to be developed further and implemented on a much larger scale to see its full potential. This is the fastest way of working out the virtues and shortcomings of any given interface, and it’s also the best way to guarantee that the infrastructure to sustain more AI systems gets built.
This is particularly urgent when it comes to edge computing. Traditional data client-server computing involves redirecting large amounts of data from the client endpoint (like a user’s computer) through a corporate LAN, and back to the client. But this is not the most efficient way of handling data, and it’s becoming outmoded as the amount of data produced by devices grows, with centralised data centres unable to keep up.
Edge computing means setting up remote servers closer to where the data is being produced, which allows data to be gathered and acted on more quickly. This is where the AI side comes in – edge computers need to be able to differentiate between useful intelligence and unnecessary data, feeding back only the most urgent insights to the principal data centre.
Tech target recently did a survey in which 54 percent of respondents expressed an interest in edge computing, but only 27 percent said that they had already implemented it. This will be one of the trends to look out for in 2022, as companies working with large amounts of data embrace the benefits of quicker and more reliable computing. ABI Research expects that 43 percent of artificial intelligence tasks will occur on edge devices by 2023.
One exciting application of edge computing is in autonomous vehicles, where it is necessary to gather, sort, and respond to data streams extremely quickly. Edge AI is used in the vehicles competing in the world’s first head-to-head high-speed autonomous land race, the Indy Autonomous Challenge.
Ethical challenges remain
Of course, 2022 will also bring its fair share of challenges for artificial intelligence. As concern grows around algorithms inclined to discriminate on the basis of appearance, more and more countries are legislating to contain its growth until we understand it better. The U.S. White House is currently working on an “AI Bill of Rights”, designed to guard against powerful AI impinging on civil liberties in the same way that the original Bill of Rights aimed to limit the power of big government over individuals.
But while there is a consensus that greater scrutiny of AI is sorely needed, it remains unclear what form that scrutiny should take. Some AI sceptics favour mandatory audits of algorithms, much like a company’s financial audits. Others would rather see “impact assessments”, like environmental impact reports. Expect to see more debate on the ethical quandaries of AI in the year to come.
2022 is set to be an exciting year for AI. Look out for more news on the development of the fabled Metaverse in 2022, as well as fully self-driving Teslas and the first Atlantic crossing by an autonomous ship. Above all, expect the unexpected. Brand new mind-boggling AI initiatives will be born next year which we couldn’t even dream up – maybe it won’t be long before computers can dream them up for us.