Back in 1982, Blade Runner predicted a future so advanced that we would have humanoid robots that not only were able to think on their own but resembled human beings so much (down to autonomy and even emotion) that investigators were actually needed to separate them from actual humans with the naked eye. These replicants were supposed to be walking the rain-slicked streets of L.A. by 2019.
Blade Runner 2049 brought a new generation of replicants to the screen, but we’re still nowhere near spawning androids that are more man than machine. Not that artificial intelligence hasn’t advanced beyond all imagination. You were probably either excited or terrified when the computer Deep Blue beat chess mastermind Garry Kasparov at his own game in 1997, and robo-brains have only been getting smarter—but could we possibly be seeing robots that are indistinguishable from humans in this lifetime?
Neural networks could possibly get to that level of brainpower. These are basically computerized neurons that form connections much as your brain does, and while they can’t exactly function like a human brain yet, they have an extraordinary capacity to learn things such as beating video games and even writing the strangest Game of Thrones fan fiction ever. Until George R.R. Martin actually finishes The Winds of Winter, a computer did it for him. Is the AI-generated novel worth reading? Maybe if you’re an insomniac who has nothing better to do at 3 in the morning and want to learn all about a character named Greenbeard.
Even more impressive is the AlphaGo system. This bot beat its human player at go, otherwise known as the most complex strategy game in existence, last year. How did it learn? Practice (sort of). You would have to call observing millions of games and using a duo of neural networks to determine game status and plot its next move “practice.”
Digital brains may be leveling up, but artificial bodies still can’t match up to the biological model when it comes to physical activity and expression. Programmers would have to inject insane amounts of data—with processing capabilities that would be like today’s most advanced AI on steroids—to even dream of replicating a replicant. There are too many stimuli in the world, and neural networks just can’t out-brain us when it comes to reacting like a human. Even 3D-printed “robot muscle” isn’t the answer. Robots are terrible multitaskers.
"You can train a neural network to be reasonably good at simple tasks, but if you try to get them to do a lot of different tasks at once — speaking, recognizing an object, moving limbs — each of those is a really tough problem,” electrical engineering researcher Janelle Shane told Live Science. “It's hard to anticipate what they can encounter and make them adapt to that."
Shane would know, because when she tried to make an artificial mind conjure Dungeons & Dragons spells, it ended up being a total noob. For some reason it thought the magic word was “Dave." So it tried to cast “Charm of the Dave,” “Storm of the Dave,” “Hail to the Dave,” and whatever "Mordenkainen's lucubrabibiboricic angion" means.
For now, watch Blade Runner again to reboot your imagination, and use your own brain to play RPGs.