It would be a world in which we would use a variety of specialized robots for manual labor and basic services (like healthcare). Such “smart” robots will soon play a larger role in our daily lives. And in the coming decades, robots will gradually move out of the industrial and scientific worlds and into everyday activities–from domestic work to factory labor. So the question arises: How is it that we will be able to make smart robots, well, “smart”? And will a GENERALIZED intelligence ever be possible when the substrate is silicon instead of brain-matter?

Just as physical robotic design is a useful tool for understanding animal and human anatomy, AI research is useful for understanding how natural intelligence works. Such insight is also crucial for designing robots.

But there’s a catch-22 here: The seemingly insurmountable challenge for AI-developers is understanding how NATURAL intelligence works–something we still haven’t been able to do. Developing AI isn’t like building an artificial heart. For scientists don’t have a simple “real world” model from which to work. After all, the brain is beguilingly complex; as it contains billions and billions of neurons…all of which are PLASTIC, not static. We think and learn by establishing electrical connections at different synapses; but we don’t know exactly how all of these connections form, and RE-form, thereby yielding higher reasoning. We are even unsure about the nature of many low-level cognitive operations.

Such complex–and perpetually DYNAMIC–circuitry is confounding even to neuroscientists. Suffice to say: We do not have a simple model for designing AI, which gives us no blue-print on which to base a design of higher functions.

Felicitously, AI has arguably become the most coveted field in robotics. AI would ultimately be a re-creation of the human thought process–effectively: a man-made machine with our intellectual abilities. A postulated “strong AI”, capable of “deep learning”, would boast the ability to learn just about anything. For it would include the ability to engage in deductive reasoning, the ability to use language, and the ability to formulate novel ideas (presumably, by using some simulacrum of creativity).

But we’re not there yet. Currently, AI can replicate a few isolated elements of cognitive processes. Most of us know this already, as we are all familiar with the fact that computers can solve certain kinds of problems, and operate in limited realms (as with, say, a chess-playing program). The fact of the matter is, though, that current computers can only solve the sort of problems that they’ve been programmed to solve…which is to say: They do not have generalized analytical ability. And GENUINE creativity is still beyond the horizon of technical feasibility.

Be that as it may, some robots now have the ability to learn in a rather circumscribed way. “Smart” robots (that is: robots equipped with machine-learning capabilities) can recognize if a certain action has achieved a desired outcome. A robot’s computer stores the information (about positive and negative results), and attempts the successful action the next time it encounters the same situation. Thus the AI gathers facts about a situation through sensors and/or human input, compares this information to stored data, and then decides which course the information signifies. It runs through various possible scenarios, and decides which action will be most successful. All this is based on the collected information, not on original thinking.

So the capacity for modern computers to perform such tasks is very limited. This entails that the array of situations robots can currently handle is quite narrow. Smart robots cannot absorb the sort of “outside the box” information that a human can, which means they are incapable of improvising…or being creative in any productive way.

As mentioned, some of today’s “smart” robots are able to learn by mimicking human actions, and remembering what works. That is a great starting point for further research.

A perk of AI is its amazing ability to deal with massive amounts of data; thereby augmenting humans’ ability to make informed decisions. This offers many benefits across several disciplines, from meteorology to pharmacology.

Climate researchers have had the information needed to elevate their understanding of weather systems for a long time now. They just haven’t had the tools to collate it and cull useful conclusions from it. AI has finally provided meteorologists with the computing power to crunch the data that they have been accumulating for decades. This vast collection of data can serve as a training set for machine-learning. AI can make sense of this massive stock of data, and enhance the performance of climate modeling.

One of the frontiers of AI is the simulation of social interaction. A robot at MIT’s AI lab named “Kismet” recognizes voice inflection and human body language; and responds appropriately. This is based solely on tone (rather than lexical apprehension) and visual cues (picked up by motion sensors). This low-level interaction could be the foundation for a human-like learning system. Going forward, the idea will be to more accurately model natural intelligence…taking cues from human behavior. In theory, “Kismet” can do this by processing information at multiple levels of “consciousness”, just as human minds do.

This arena of research holds tremendous promise. Scientists are now creating neural networks that can pilot driverless cars and autonomous robotics systems. They also write algorithms to help doctors identify disease-carrying genetic mutations. The possibilities seem endless; and that exhilarating fact is what seems to be motivating those with the vision to get involved in this amazing new field, the possibilities of which are limited only by our imagination.