Science fiction has been announcing the coming of artificial intelligence and robotics for longer than most people have been alive. Our 21st century culture has been immersed in these visions for so long that we take them as manifest destiny.

Over the past week, however, several AI industry announcements sent shivers down our collective backbones, suggesting some of the creepier Hollywood scenarios that have crept into our collective awareness. Anthropomorphism has always been AI’s heart, but much of what’s emerging now from the AI and robotics industries is beginning to cross over into the “uncanny valley” where humanlike entities resemble us so closely that they’re starting to elicit waves of apprehension if not outright revulsion.

If providers of next-generation AI solutions don’t address this issue head-on, they could inadvertently trigger a consumer backlash and corresponding regulatory overreach. In terms of recent stories that might feed this anti-AI backlash, I’ll single out the “jogging humanoid video” that, according to this Washington Post article, is currently “terrifying” people across the internet. This is the latest such video from robotics maker Boston Dynamics, all of which show its AI-equipped robots climbing stairs, doing back flips, opening doors (pictured) and otherwise moving around and manipulating physical objects as well as or better than flesh-and-blood humans.

If Elon Musk needed fresh content to underline his increasingly alarmist anti-AI screeds, the Boston Dynamics video shown below would definitely fit the bill. In it, a robot named “Atlas” achieves “whole-body mobile manipulation” without losing balance, slowing down or appearing to tire of its activities. It can navigate its environment dynamically, relying on stereo vision and a wide range of onboard sensors for situational awareness. Its limbs can manipulate objects in its environment in order to clear away obstacles and otherwise meet its objectives.

All of that is amazingly impressive. However, it’s also unnervingly reminiscent of the android played by Arnold Schwarzenegger in the “Terminator” movies. All you need to do is equip “Atlas” with a head, sunglasses, Austrian accent and rapid-fire assault weapons to complete the dystopian scenario. Compared with the feat of getting Boston Dynamics’ android to walk and exercise situational awareness like a flesh-and-blood homo sapiens, these would all be trivial engineering change requests.

Speaking of Android, Google last week demonstrated its astonishing Duplex technology. When introduced into the company’s mobile operating system, Duplex will enable Google Assistant to carry on natural-language phone conversations without betraying the fact that it’s a bot. Amid all the oohing and aahing over the technology’s sophistication at its Google I/O demo, the announcement triggered waves of concern about its potential to transform robocalling into a tool for mass deception on an unprecedented scale. By relegating synthesized digital-assistant voices to history’s dustbin, Duplex has leaped over that uncanny valley clear to the other side.

What these announcements reveal is that the Turing test is mincemeat and that before long machines will be able move, manipulate and otherwise interact with the world as well as any human, and perhaps any sentient life form of any sort. A new generation of robotically embodied superintelligence is emerging, and society will need to fashion a comprehensive response in order to keep these bots — physical and virtual — out of mischief.

However, AI and robotics developers aren’t satisfied with only imitating our own species. Researchers have built robotic replicants that can imitate any actual organic creature, and many advances in AI are equipping them with the intelligence suited to any and all of the following animal modalities:

• Locomotion: The robotics revolution has spawned a cyber-Cambrian explosion of intelligent robots that can walk, pounce, flap, flutter, hover, trot, creep, gallop, swim, burrow, crawl, slither and otherwise move around as an actual organism might.
• Sensation: To ensure that their creations are in tune with whatever environment they’ve been built for, robotics researchers are leveraging a growing pool of AI research into every perception modality, including vision, hearing, smell and touch.
• Manipulation: To give robots the symbolic-manipulation prowess generally associated with intelligence, researchers continue to push the envelope in cognitive computing, as well as into algorithmic emulation of such associated faculties as emotion, imagination and communication. Just as important, the neuromuscular manipulation associated with physical grasping and grappling are a huge focus of today’s AI robotics community. The neurological substrate that makes all of this possible in actual organisms is one of the most vibrant research areas of all, under the heading of “neurorobotics.” Other fields of AI research — such as drones, swarm intelligence, reinforcement learning, embodied cognition, multiobjective decision making, evolutionary computing and master learning algorithms — are giving robots the ability to grapple with a wider range of scenarios, environments, outcomes and tradeoffs.

In other words, anthropomorphism is giving way to “zoomorphism” as the larger paradigm within which AI and robotics technologies are advancing. Though that may not sound threatening on the surface, we have to remember that the predatory behavior is fundamental to many, if not most, animals’ survival strategy (including our own).

When you consider this perspective, you can’t help noticing the latent threat potential of seemingly innocuous advances. For example, “Ultra Low Power Deep-Learning-powered Autonomous Nano Drones” suggests that some day we may be plagued by intelligent, winged roboinsects. Adversarial threats, such as smart speakers being hijacked to obey white-noise-masked “dog whistles,” should remind us that there are plenty of animal modalities that an unscrupulous human can emulate through souped-up AI wizardry. Combining fine-grained decoupling of AI microservices with distributed mesh networks suggests intelligent swarms of simulated bees, ants and other militarized nanocritters.

Researchers will continue to develop more sophisticated technologies for protecting robots from their own clumsiness and from deliberate attempts to bring them down. Likewise, AI researchers will continue to improve defenses against adversarial attempts to fool their algorithmic brains.

But who will improve humans’ organic faculties for distinguishing AI-powered robots from our own kind? And how will we protect our mortal selves from the fearsome weaponry that robots will have at their disposal? Is “Atlas” friend or foe?

At this point in the technology’s development, how can we know?

Image: Boston Dynamics