Buzz
Many cautionary voices warn that artificial intelligence and robots will eventually dominate the world. Over time, we may see them performing tasks ranging from doing our laundry to engaging in warfare, with minimal human involvement other than teaching them. Thanks to machine learning, they might no longer require us for that. The AI of tomorrow could be so advanced that it anticipates our actions before we do.
What many fail to realize is that artificial intelligence is already bringing to life what we once thought was a distant reality. In fact, today’s AI can perform tasks we never imagined it could accomplish, even in the future. Not all of these developments are cause for concern, though; many could ultimately provide us with real benefits in the long run.
10. News Story Writing
Despite fears about AI replacing jobs, certain professions are generally considered immune to automation, and journalism is often thought to be one of them. After all, we believe it takes a human touch to effectively report important information in a coherent, structured way that’s easy for everyone to understand. However, the reality is that bots capable of writing news stories as effectively as skilled journalists already exist.
Although previous attempts to create bots capable of writing news stories failed due to the limitations of AI, that's no longer the case. The Washington Post has successfully launched a story-writing bot named Heliograf, which can produce articles on par with the work of its top journalists. All it needs are a few key phrases outlining possible outcomes of a newsworthy event—like an election—and a database of events for real-time updates.
9. Serve As Police Officers
Many are familiar with RoboCop, the iconic character from the fictional story about a cyborg police officer that imagines a possible future for law enforcement. But RoboCop isn't entirely robotic; he retains a human brain, which, when combined with robotics, makes him a formidable force.
While robotic law enforcement seemed a far-off concept, Dubai has already deployed a robot cop to patrol its streets, even going so far as to name it 'Robocop.' It’s a step into the future of policing that’s happening right now.
If you think this Robocop is limited in abilities, think again. Developed with the assistance of Google and IBM's supercomputer Watson, this robot can do much more than you’d expect. It’s capable of identifying criminals, flagging suspicious vehicle plates, alerting authorities about unattended bags in public areas, and much more. This is all part of Dubai’s vision to have 25% of its police force be robotic by 2030.
Currently, there are no plans to arm these robots, and we aren’t suggesting that Robocops will ever pick up guns and revolt against humans. In fact, this technology could prove to be a significant help to underfunded police departments globally.
8. Create Its Own AI Software
There has been much speculation about the extraordinary feats AI could achieve in the future, but if you’re a coder, you’ll understand that making AI perform those feats is far more complicated than just writing about them. AI developers are not only some of the brightest and most skilled programmers in the world, but they are also among the highest-paid due to their rarity. It’s extremely challenging to create AI software, which is why it’s a big deal when AI learns to do it on its own.
Many companies have experimented with AI creating its own machine learning software, but until recently, it was never on par with the work done by human AI developers. In 2017, Google created an AI that could design its own AI. For the first time, the AI it created outperformed the software made by the same team of researchers. When tested to identify objects in a picture, the AI-generated software achieved 43% accuracy, surpassing the 39% accuracy of the software created by humans.
If it’s not already clear, this suggests that AI could eventually replace those who design AI systems as well.
7. Deception and Dishonesty
Deception is often thought to be an exclusively human trait, something machines simply cannot do unless they completely go off the rails. While we’ve designed AI to lie or bend the truth in the past, there hadn’t been instances of machines learning to do so autonomously—until some recent reports came out, and honestly, it’s a bit unsettling.
In one example, researchers tasked an AI with playing Sonic the Hedgehog for an AI retro gaming competition. The goal was simple: complete the level as fast as possible while keeping an eye on the competitors in case they overtook. To their surprise, the AI quickly figured out how to glitch through walls, which may very well be the first instance of AI learning to cheat in a game without being programmed to do so.
In another instance, a collaboration between Stanford and Google scientists revealed that an AI, tasked with converting aerial Google Maps images into street maps, was found concealing some of its data in a high-frequency signal that was virtually undetectable.
6. Collaboration for a Greater Purpose
Human society is built on the ability to work together, giving us an advantage over other more self-interested creatures in our early history. This teamwork isn't just for peaceful pursuits like farming or city-building; it has also played a crucial role in wars and conquests. So, it’s only natural to be both impressed and slightly fearful if machines begin to master this skill themselves.
Fortunately (or unfortunately), AI now has this ability. Google’s DeepMind project has developed an AI capable of collaborating with other AIs in multiplayer games like Quake III Arena to win the match, something they’ve been working toward for quite a while. While AI has previously shown it can beat human players in video games, this is the first time it has done so as part of a team—something that requires cooperation and adapting to other players’ styles.
5. Create Poems
If robots were capable of writing poetry, the world likely wouldn't change much. Aside from a few successful poets who actually make a living from it, poetry isn’t considered a mainstream profession. (Sorry!) Writing poetry requires an understanding of meter, rhyme, tone, and other nuances that only a human mind truly grasps. It’s also challenging for AI to pick up these subtleties unless specifically trained.
Surprisingly, though, AI is already writing poems that most of us would have a hard time distinguishing from those created by humans. For instance, take a look at this Shakespearean-like poem entirely written by a bot:
When I in dreams behold thy fairest shade Whose shade in dreams doth wake the sleeping morn The daytime shadow of my love betray’d Lends hideous night to dreaming’s faded form
Of course, the AI didn’t just spontaneously produce this masterpiece. An MIT PhD candidate, J. Nathan Matias, taught it the fundamentals of sentence structure and syntax. After many failed attempts, the AI eventually crafted something original and poetic.
4. Create Art
One profession that we thought would remain immune to replacement by machines is that of an artist. Art is often seen as a uniquely human pursuit, as it relies on human perception, depth, and the ability to appreciate subtleties like shadows. For a machine to match these skills, it would need to replicate the intricate understanding that only a human brain can offer. So, it seemed impossible that AI could ever truly replace the artist.
However, it’s concerning news for aspiring artists, as AI has already cracked the code on many of these human qualities. In a 2015 experiment by researchers at the Bethge Lab in Germany, an AI was trained to understand the components of art and replicate that style. To everyone’s surprise, the AI was able to recreate a German street scene in the iconic style of Van Gogh, perfectly mastering the use of shadows and brushstroke techniques.
3. Read Our Minds
The concept of a machine being able to read our thoughts feels like something straight out of science fiction. While it’s hard for most of us to imagine how this could even be possible, earlier attempts had managed to decode our mental images into vague signals, though they were largely unintelligible to most people. That is, until now.
In a 2017 study by Japanese researchers, an AI was able to actually 'see' the mental images forming in people’s minds and accurately recreate them through drawing. And it wasn’t just abstract, black-and-white shapes—it was surprisingly close to the images that the participants were thinking about during the ten-month experiment.
In another fascinating experiment, an AI was able to convert human thoughts into sound signals. When these signals were played to listeners, they were able to identify them with 75% accuracy. While the idea of AI reading our minds may sound chilling, it also opens up many useful possibilities, such as helping to identify and treat hallucinations in patients with schizophrenia.
2. Debate The Meaning Of Life
Philosophy is one of those fields that you don’t truly grasp until you dive into it—and then, once you do, you can’t stop pondering and discussing it. Until now, coming up with profound reflections on the human condition has been a uniquely human task, but that’s changing.
A chatbot created by Google (yet again) has been reported to provide some rather thought-provoking answers to the big questions in life. Regardless of where you stand on the debate about artificial intelligence becoming too smart, it’s hard to ignore how profound some of these responses are:
Human: What is the purpose of life? Machine: To serve the greater good. Human: Where are you now? Machine: I’m in the middle of nowhere. Human: What is the purpose of dying? Machine: To have a life.
While most chatbots rely on certain keywords in questions and often fail to provide convincing answers, this one was designed to mimic the way a brain functions, with interconnected nodes and branching pathways, to arrive at the most accurate response. Though its main aim is to enhance Google’s search algorithms, we wouldn’t mind if it got a book deal or two on the side.
1. Learn How To Encrypt On Its Own
In the event of a global AI takeover, one crucial capability it would need to succeed would be the ability to encrypt its own communications. Even if AI were to devise its own encryption methods, it’s likely that we could eventually figure out how to decrypt them, since we created the AI in the first place. Of course, this is an oversimplified view of how an AI takeover could unfold, but having the ability to encrypt its messages would certainly improve the AI’s chances.
It turns out that AI has already discovered ways to encrypt messages independently. In a 2016 Google AI research experiment, two neural networks, Bob and Alice, were set up to communicate with each other. A third network, Eve, was introduced to try to intercept and decode their conversation. Although Eve succeeded for a while, Bob and Alice eventually developed a method to encrypt their messages in a way that even Eve, an advanced AI, couldn’t crack. The encryption wasn’t anything we’ve seen before, nor anything we’ve created ourselves. Perhaps Google should reconsider these experiments if they know what’s best for us all.
