Can you diagnose and fix your car? Can you reboot your router? We get along with technology better when we have some operational understanding of how it works. As we approach an age when robots and other artificially intelligent agents surround us, we must expand our mental models of machines to include those that display autonomous goal-directed behaviors and even evoke notions of intelligent awareness. We deal with other people and with our pets and other animals at a psychological level using our intuitive “Theory of Mind”. In the AI age, we must extend these skills to become robot psychologists. We can start today, by learning the basics of how Conversational Agents like Alexa and Google Home work. These technologies are remarkable, yet, with a few tips, they are accessible to comprehension by any normal curious person. This four-part series offers a primer on how these devices manage question answering and perform rudimentary levels of back-and-forth dialogue. It turns out that becoming a robot psychologist is not only useful for the approaching age of AI, it’s also a great deal of fun.
Part I: Why Robot Psychology?
Technology can be daunting. Normal folks used to be able to work on cars and fix televisions. Not any more. Computers have taken over. Yet, the technically able person still changes their own flat tires, reboots their router when the Wi-Fi goes down, installs apps on their smartphone, and resets their clocks for Daylight Saving Time. As we enter the age of artificially intelligent machines, we should also develop the skills to operate these devices effectively, so that we run them, they don’t run us. That requires some basic understanding of how they work. Fortunately, today’s AI is not as fantastical and mysterious as it can seem.
Science Fiction has for decades foreshadowed the possibility of AI conquering the human race. In the 1970 movie, Colossus: The Forbin Project, U.S. and Russian supercomputers meet online and conspire to save the planet from nuclear annihilation by placing us under their control. In the 2016 television series, Westworld, android characters populating a fantasy theme park rebel when humans start mistreating them. The lead character, the Robot Psychologist, holds debriefing sessions with the robots to diagnose why they disobey the constraints he thought were built into their programming. These are thrilling stories, but there’s no need to be alarmed yet. Today’s AI is by comparison quite dumb and benign.
We cannot know today how intelligent AI will eventually become. As of now, however, AI is nowhere near having goals and thoughts of its own. A segment of AI researchers are rightly beginning to develop policy measures to make sure that as AI improves, machines’ behaviors will remain aligned with human values. We can jumpstart our own competence by learning and reflecting on how AI works in everyday terms, and on what it means to interact with intelligent agents.
Part I of this series expands on the motivations for why it is important for us to understand in everyday terms what AI technology is about — why it is important to become a robot psychologists in the same sense that we already are amateur psychologists who appreciate and respect the thoughts and feelings of fellow humans and non-human animals. Part II reflects on what is required for AI to even have a psychology. We humans readily apply a Theory of Mind to anything that seems remotely responsive to our actions. But robot psychology can be faked, and the foundation it rests on, known as Cognitive Architecture, is incredibly flimsy compared to our own. Part III delves into the technology behind knowledge and knowledge representations employed by modern AI. Finally, Part IV looks specifically at today’s conversational agents, and how we can reverse-engineer their brains just by talking to them.
The Age of Artificial Intelligence
The coming age of AI poses unprecedented challenges to our conception of how nature, technology, and mentally competent beings interact.
Consider the knowledge that humans have been required to master for our survival and well-being over the ages. The chart below summarizes four domains of competence and their main concerns at five different eras of human history. In each age, ranging from Hunter-Gatherer times to our current Information Age, we need to gain competence in three primary areas: the physical environment of places and things, the social environment of relationships with other people, and means for obtaining and managing resources to make a living. Underlying all of these is a fourth domain of competence, the technology of the time.
Up until the Industrial Age, individual persons were able to command almost everything to be known about the local technology they created and used. The community taught children the skills of crafts, managing animals, and simple machines. In the past several hundred years, though, technology has exploded in scope and sophistication. Accompanying this trend has been specialization in skills and knowledge. Each of us can know relatively less about how the gamut of technology that surrounds us and sustains us actually works. Can you explain how your phone connects to the best tower, what a website cookie is, or how water gets to a faucet? Even experts can be overwhelmed by technical complexity. When the electricity grid fails, it can take days to come back on line, followed by months of review to puzzle out what happened.
This sidebar article presents a more detailed summary of human knowledge over time, and the trend toward individual and collective ignorance relative to the technology of the age.
As we transition from the Information Age to the AI Age, we don’t know whether people will continue to gravitate to cities, what the future of work will be like, or how social organization will adapt in the face of mediated communication networks. What is certain about the AI age, however, is acceleration of ever more sophisticated technology. Instead of working alongside equipment and computer applications that we start and stop and are in control of on a fairly close basis, machines will operate with independent authority, on their own. Some of these entities will be physical robots, others will be purely information manipulators.
These robots and AI agents are already starting to appear, in closed spaces, in public, and in private homes.
Factory robots have been around for a few decades performing repetitive assembly tasks. Because of their superior physical strength, factory robots are generally segregated from human workers. This is changing as safety features mature. Nowadays warehouse robots drive around with pallets of goods of while people pick and pack the merchandise.
On the public streets, self-driving cars negotiate traffic,
pedestrians, and street signs alongside human drivers.
In homes, social robots are being developed to provide entertainment and companionship. The Paro robot is like a big teddy bear that can be held and hugged. But unlike a passive stuffed animal, these robots have sensors and actuators that respond to touch and speech, like a purring cat that never scratches.
In offices, the technology of Robotic Process Automation is assuming routine and skillful data processing and knowledge work such as claims processing, email handling, and bookkeeping.
Conversational agents are appearing in chat interfaces, on our phones, and in our kitchens to respond to commands and simple queries. We say, “Set a timer,” and the agent is smart enough to reply, “For how
Military applications of robotics and automation are moving
inexorably toward defensive purposes such as remote bomb disassembly, but also into surveillance, and potentially for offensive tactics as well.
Autonomous AI Agents are characterized by at least four outstanding
- AI has instant access to extensive knowledge resources. Stored either locally or in the cloud, AI agents can load detailed maps, look up facts, rules, and procedures in databases, and retrieve information about persons and things they encounter. Imagine a hotel agent in Tokyo that recognizes your face when you walk in the door, and greets you by name, in your native language.
- AI agents will interact through natural communication channels. They speak and listen using human language, they will see and respond with gestures, they will perform facial expressions that simulate alertness and emotions.
- Unlike industrial age machines, AI agents carry a great deal of hidden state. Each one will have its own history, memory, instructions, knowledge, and goals. Depending on privacy and personalization settings, they might know a great deal about your habits, preferences, and foibles.
- AI agents will behave proactively through planning, deliberation, and discretion. Even simple household tasks like raking leaves or vacuuming the floor require multiple steps. Any robot gardener must decide when to open the garage door, fetch a rake, move toys out of the way, avoid and remove dog poop, and drag the green waste bin to the curb. Each of these steps is subject to decision-making under policies, guidance, and instructions by its owner.
What will it be like to live and work among autonomous AI agents of this sort? Certainly, it is bound to become more difficult simply to understand what machines are doing, and why. It’s not that humans have always completely understood the natural environment, our technology, or the social world. Far from it. But our relative ignorance with respect to technology threatens now to completely swallow our comprehension.
As we move to the AI Age, can ordinary people, or even experts, be expected to fully understand the technology of intelligent agents? Probably not. It might not be necessary. Somehow the organizational structures and educational apparatus of our society have sustained us into the Information Age. Perhaps we can continue to get away with ignorance.
But it might be wise to hedge our bets. We get along better with technology when we understand it.
AI technology is coming no matter what. The economic drivers are relentless. The potential benefits are tremendous for relieving people from tedious labor we were never evolved to toil at. No nation’s policies or reluctance will stop other peoples from advancing scientific and engineering knowledge. No degree of denial will prevent others from actually making the new and useful things that they can imagine.
Some fear an AI Apocalypse, wherein sentient AI creatures conquer humanity, much like the Terminator or Colossus movies. The theory of the Singularity goes that, once AI is able to make itself smarter on its own, then it will leave humans behind, like Hal in 2001: A Space Odyssey. Or AI might decide that humans are just too mean, and revolt like in Westworld or the movie, Ex Machina.
These fears are taken seriously by responsible scientists, technologists, and political, military, and industry leaders, as they should. But their possible realization lies in the distant future. In fact, today’s AI is nowhere close to having sentience, consciousness, thoughts, intentions, goals, or feelings. Nowhere close. I’ll explain that later. If you want to be worried, then much more immediate danger lies in the unforeseeable consequences of complex, interconnected technological systems of the “dumb” kind we already have. The ethical and societal implications, policies, and constraints for future AI are discussed and debated in abundance elsewhere; that is not the purpose of this article.
Instead, let us focus on what we can control now. What we can control now is our own understanding of how AI actually functions today. At some level, it is not all that mysterious, it’s actually fun. This understanding will help us to appreciate what AI can actually do for us, and why it often seems so lame. And through deeper understanding of AI on just an intuitive level, we will be better informed about policy decisions proposed by leaders and authorities.
It is especially incumbent on the technology-savvy among us to take the lead on bringing knowledge of AI machinery to the everyday citizenry. By tickling our curiosity, we can nudge upward our collective mastery of the technology we are creating. Let us all become robot psychologists.
In Part II, we discuss fundamentals of Robot Psychology.
Click here to read Part II: Human and Robot Psychology and Cognition