Eyes on the Prize: Why ‘Automation-or-Bust’ Thinking Gets It Wrong on AI

AI technologists are in fast pursuit of artificial general intelligence (AGI), with some predicting it could arrive as soon as in the next few years. Just this month, Google outlined their latest step towards AGI with a new AI model that interacts with a convincing simulation of the real world.

It’s easy to see why there’s been so much hype around AGI. An AI tool that can think as well as humans, act on its own, and do everything a person can with the processing speed and knowledge of a tireless AI system would completely transform our economy and definition of humanity.

But going all-in on AGI is a misplaced goal, as doing so risks putting too much weight on automation-at-any-cost. We’ve become so enthralled by the idea of AGI that we’ve failed to realize the inevitable shortcomings that may accompany it: we’re thinking it’ll solve all our problems, when they’d be better solved with the collaborative technology that’s already right here in front of us.

AGI and full automation shouldn’t be the main goal in our AI journey, and robots’ lack of robust perception in complex environments plus a still-limited ability to adapt makes achieving this goal a ways off regardless. Instead, we should be focused on seamless human-AI systems in which we combine human judgement and intangibles with the productivity and speed of AI. With AGI as the singular goal for many technology developers, however, is it possible to create advanced AI while keeping humans at the helm?

Nothing makes the current predicament clearer than the world of robotics. While it would be incredible to have a fully automated robot that could run the factory floor by itself, this robot would not be of much use to anybody if its productivity remains a fraction of that of a human worker. I know this issue well, especially as it relates to intelligent robotics.

During my two decades as Chief of the Software, Robotics and Simulation Division at NASA’s Senior Executive Service, I saw up close the balancing act of full-blown automation and the need for a human touch. Often times, we get too caught up in creating the perfect robot; one that can do everything and anything, at any cost. But this often misses the point entirely. In both robotics and AI, technology is a tool, not a craftsman. Furthermore, automating the last 10% of a system often costs as much as the first 90%, so a good business decision results in a mix.

To this end, Human-in-the-Loop (HITL) systems, which combine human intelligence and machine-learning capabilities, are still the best way forward for AI and robotics. They also demonstrate a clear template for how we should engage with this emerging technology and how we can prepare for the onset of AGI.

Take for instance the robot dog ‘Spot’ who helps monitor the buried Roman city of Pompeii. This robo-pup can gather and record data to help manage the area, keep it safe, and notify archaeologists of any parts of the site that have fallen into disrepair or pose potential safety risks. This machine helps archaeologists focus on the tasks only they can do, including carefully digging for ancient artifacts and continuing their tireless research. No matter how advanced AI gets, the final archaeological judgements must be made by a human expert, as only they can assess the inherently human value of the artifacts in question.

A close quadruped relative of Spot is the Swiss inspection ‘cobot’ (collaborative robot), which helps professional inspectors survey the equipment, safety, infrastructure, and air quality of industrial facilities. While a human expert must still analyze the data and make final determinations, this inspector-bot can help gather up-close, high-quality information that would otherwise be difficult or impossible to collect. Using visual and thermal cameras, an ultrasonic microphone, powerful spotlights, and even Lidar sensors, this machine can traverse nearly any terrain on Earth, even resorting to parkour if need be. The inspections are far too important to conduct with robots alone, so human judgement is added once all the data is collected.

These robots are already in the field, simplifying dangerous or previously laborious physical processes to increase productivity many times over, and this type of human-AI collaboration is where we must focus.

We’re seeing these strides beyond fieldwork as well, in situations where human lives are on the line. In the operating room, the Da Vinci Surgical System is helping doctors conduct surgeries with extreme dexterity and consistency. Using this system, surgeons use a viewfinder, a patient cart with robotic arms, and a 3D visualization tool to carry out complex procedures with tiny incisions, meaning less pain and a faster recovery time for the patient. Consider the viral video of a grape undergoing a surgical procedure. That was done completely with the help of the Da Vinci system, combining the judgement of senior surgeons with the manual coordination of the best young hands.

While the technology performed a large degree of the work, it was all overseen by a highly trained medical professional, giving much more peace of mind for whoever’s going under the knife. However intelligent and capable the AI may be, few patients would likely feel comfortable with the final medical decisions being made by an entity than does not, and cannot, know what it means to be human.