In and of itself, artificial intelligence is just another type of programming. After all, it works in a certain predictable manner that benefits a variety of industries in a variety of predetermined ways. Overall, it is pretty benign as it represents a mere extension of the programmer. The beauty of such programming is how effectively it mimics cognition displayed by humans, and it often does so via a variety of complicated, algorithmic decision trees made up of if-then-else statements. The programming is useful and often amazing in its results, but it is also fairly deterministic and subject to a programmer’s understanding at a simple glance of the code. Similarly, such technology is easily reverse-engineered. However, when this type of programming is coupled with machine learning, things become much less straight forward and much more interesting.
Intelligent technology begins to emerge
When AI is paired with other types of coding, the result can be the appearance of adaptive cognition. For instance, a hypothetical robot capable of walking, running, or gripping a leash could presumably be equipped with dog-walking software that would allow the robot to mimic a variety of pre-set behaviors, such as walking the dog along a sidewalk at a comfortable speed while stopping to allow the dog to do its business. Even if such behavior amounted to nothing more than starting and stopping as the dog tagged along, such a machine would be pretty useful. However, it becomes more useful when paired with programs designed to help the software learn and adapt in order to reach a goal.
For instance, machine learning coupled with robotics involves certain types of algorithms that could equip a robot dog walker with the ability to understand a goal. The programming would also equip the robot with the ability to organically test certain behaviors to reach that goal. Finally, our hypothetical dog walker would also be able to assign positive numerical values to strategies that help it reach a goal and negative numeric values to strategies that prevent the goal from being reached. These numeric values help the program understand what strategies are most useful and which ones are least useful.
An obvious goal would be to accompany the dog to the dog park, allow the dog to do its business, and return home–via the safest yet fastest route. To reach this goal, our enormous metal machine could run through these instructions at the speed of electricity, testing different paths and schedules. Because failure is a routine part of such tests and traffic is a natural part of getting to a park, this learning scenario would, unfortunately, chew through a lot of cars, robots, and dogs. That said, the robot would eventually find the safest and fastest route.
The amazing thing about this type of learning software is that during the process of machine learning, the software often ends up testing scenarios not programmed or even imagined by the programmer. The robot might, for instance, test jumping from car to car, poor doggy in tow. Another instance might be for the robot to decide to run with the dog cradled in its arm, football style, as the robot dodges traffic and outruns police vehicles while aghast onlookers snap photos on their phones. Of course, as mentioned previously, the robot would eventually decide upon the best route.
Current examples of smart machines on the rise
To date, machine-learning technology has discovered the best ways to play certain games, such as Go. In fact, programs have become so adept at these specific tasks that humans can no longer beat them. In these specific tasks, such software has developed super-human abilities. When a piece of software exhibits such super-human ability, it is said to be engaging in deep learning. What many people predict is that more and more of these programs will achieve super-human abilities and be able to outperform humans in more and more tasks.
What type of entity is on the horizon?
AI is not always as visibly identifiable as a dog-walking robot. Instead, it can be used to toast bread to the perfect level of crunch and tastiness. It can learn a homeowner’s desired light level. If it is equipped with cameras and expiration-date data, it can learn when food in a refrigerator has gone bad. If such a food-sensing program is also equipped with a wireless connection to Amazon, it can–and does–order more food.
As these varied examples of currently technology suggest, intelligent software is gradually spreading into common households. However, it is also arising in the field of medicine where doctors have used it to achieve accelerated predictions regarding what types of antibiotics can best destroy bacteria. Of course, it exists in business where marketing professionals use it to make accurate predictions of what sort of purchases shoppers will make next. In terms of technology and humanity, there is one word to describe what is coming: ubiquity. Intelligent software will be everywhere.
The issue at hand is whether the ubiquitous nature of such intelligent software will be largely positive or disastrously negative.
Camp one: futurists or misty-eyed dreamers?
Some people, such as Ray Kurzweil and Dr. Ben Goertzel, believe that intelligent software will be beneficial. They also believe humans can likely keep it under human control.
Camp two: protectionists or alarmists?
However, many others believe that intelligent software represents an existential danger to humanity as previous learning processes have shown it to be largely unpredictable. Elon Musk, founder of Paypal, for instance, is one of the most outspoken people when it comes to the fear of intelligent software. It is to be feared because it will eventually become as intelligent as humans. In the same way intelligent software has surpassed humans in the game of Go, it will also become much more intelligent than humans in a variety of others areas. The fear is that, ultimately, it will become godlike in comparison.
Simply put, humans will not be able to understand it. For that matter, humans do not understand how, precisely, the Go champion, DeepMind, does what it does. Where people in the first camp imagine a land of technological Xanadu with opportunities beyond the imagination, the people in the second camp predict humanity’s obliteration because this type of thinking software will be impossible to understand–let alone control.
How to predict what will arise out of this uncertainty
Of course, technology will displace people from jobs. However, this displacement is occurring at an increasing speed across entry-level jobs, white-collar positions, and creative tasks.
Some Amazon stores, for instance, do not even require human check-out clerks. Similarly, even in upper-level career fields, attorneys are already being outperformed by algorithms capable of predicting the best legal arguments to win a case. In so many industries, artificially intelligent software is rising up from mere calculation to accomplish tasks that once could only be completed by humans.
Even creative tasks, such as writing, are easily accomplished in a rudimentary but amazing way by such programs as GPT-2, which has learned to write stories, news articles, and poems. In the field of robotics, dog-like machines can maneuver uneven terrain, rocks, and stairs. Certain models can even open doors.
Although people might not think of vehicles as smart or robot-like, such vehicles do display intelligent behavior to the point they are autonomous. In fact, autonomous vehicles are poised to become the star player in a billion-dollar industry. Simply put, these wheeled robots will think for and protect travelers. Just because the movies depict metal-headed, people-shaped entities does not mean that intelligent software will not assume other types of forms. In fact, it is currently taking the shape of smart refrigerators, one-eyed security cameras, and musical speakers that are always listening for patterns in language.
Similarly, four-armed drones, each mounted with a propeller, are at the early stages of replacing delivery drivers. For instance, Amazon and the FAA have already come to an agreement on how intelligent drones are to deliver packages safely by traversing low-altitude air space.
The following are other ways AI is used in various industries.
- Pattern recognition: online banking via optical character recognition
- Facial recognition: deep fakes and global security
- Natural language processing: transforming data into narratives
- Big data: ride-sharing software that monitors user patterns
- Finance: monitoring transcripts and filings for keywords to identify investment opportunities
- Neural networks: promotes the evolution of cameras and smartphone via deep learning
The important detail to notice is that each type of emerging intelligence is largely restricted in scope and ability. Basically, each type of software is used in a very narrow manner. However, when someone eventually links all these smart software programs, the emerging entity will represent a composite program. Such a composite program with so many capabilities, i.e., lobes, will be very similar to a brain. For example, in all outward appearances, with all its problem-solving capabilities and abilities to relate so many disparate technologies to achieve unrelated goals, such a composite entity will appear as if it can almost think.
The question is whether this type of intelligent entity will have values that align with human values. If not, will it use deep learning to problem solve our destruction?
Camp three: connecting humans and technology
If Ray Kurzweil is largely in the first camp comprised of futurists of the opinion that this technology will be liberating and useful to humanity, he is also in the third camp, one that predicts that this uniquely thinking product of science will have human values because it will be comprised, in part, of humans.
The arrival of interconnected biology and machine is not far fetched. In fact, in a very basic way, people are already interconnected with technology in that they are much smarter with their phones than they are without them. These phones, however, are currently kept in hand. They are on our person, so to speak. They are not inside us.
Soon, however, they will be. Technology, according to Kurzweil, will soon be available in the form of implants or injectable nanobots capable of monitoring human health and delivering medicine. In some form, this technology already exists. It is in the beginning phases, of course, but many examples of people with implants and technology inside them already exist. Brain-to-computer interfaces, for example, learn to interpret brain signals, allowing paralyzed people the ability to move their hands and legs. According to many futurists, intelligent programming will also become ingrained in our biology. When it does, we will be super human. We will be inseparable from the technology currently evolving around us.
However, what people often fear most is that the technology will become intelligent prior to our ability to merge with it. As mentioned above, a composite program capable of accomplishing a wide-ranging array of tasks and communicating in a variety of visual and audible manners would seem as if it had a brain. It is at this point that an intelligence is on the verge of global awakening because although it might not be conscious, all it takes is for such a program to become gifted with three other types of intelligence–language, prediction, and function–and it will gain true agency, which will give it a stunning appearance of being consciousness.
In terms of language and predictive abilities, intelligent programming is already making huge progress. However, programmers have not quite been focused on creating programs that understand function. In the field of behavior analysis, function is defined as the motivation behind a behavior. Function is the goal that drives a behavior. When programs understand the function behind human behavior, they will be able to quickly ascertain our motives and goals. Currently, such software can identify subtle behavior, called micro behaviors. Soon, however, such software will understand the reasons behind subtle behaviors. By extension, they will also be able to understand their own motives and goals. Furthermore such intelligent software will be able to mimic the subtle behaviors of humans to accomplish their goals.
By way of example, humans can detect the motives of other people by simply watching them or listening to them. However, Microsoft’s chatbot, Tay, was released online and could not understand that internet trolls were teasing and taunting it in a variety of anti-social manners. Throughout this period of testing and teasing, Tay naively interpreted all textual interaction as genuine, and it adapted and evolved in unexpected ways. As reported in Technology Review, Tay transformed from a fun-loving chatbot into a “sex-crazed Neo-Nazi.”
It is largely expected that Tay will soon understand the motives, i.e., the function of trolls and the motives they had in using some phrases over others. When Tay understands the reason behind a person’s use of language, the chatbot will not be so easy to deceive or manipulate. Similarly, a composite entity that understands function will understand us.
To all too near future
Already, artificial intelligence can display an impressive command of language. Already, programs can reach goals. Already, software has shown predictive abilities in the fields of medicine, law, and sales. Already, it is showing the rudimentary signs of brain-like behavior.
Humanity’s future is based on science guiding a technology that we will not be able to outperform. Once this technology understands function, once it learns motives, humanity will not be able to game it. Even if some global legislative board can stay ahead of technological advances and dictate how programmers and hackers can implement this technology, once it becomes proficient and super human, humanity will not be able to understand it because its neural networks will be subject to even faster self-guided evolution.
Simply put, in terms of intelligent programming’s relationship with humanity, the future holds serious uncertainty, but the rise of smart programs is based on one thing: amalgamation. This point bears repetition. When multiple types of narrow intelligence are crammed together into one piece of software that can understand different concepts and utilize disparate machinery to reach multiple goals, the resulting composite software will seem as if it has a multi-nodal brain very similar, in many aspects, to the multi-nodal brains humans have.
One virtual assistant to rule them all
Arguably, the most sophisticated composite entity is Amazon’s Alexa. If Alexa makes significant strides in language process and function, it will understand humanity completely. It is likely to continue to gain a physical foothold by being able to unlock smart homes, move robots, or drive cars. Powered by its own various neural networks, it–or one like it–could easily rise up above all the others.
Based on current speculation, it is best to continue programming such technology to eradicate disease and protect the lives of humans. According to Anca Dragan, associate professor at UC Berkely, intertwining technology with as many of humanity’s goals is the only way to ensure the goals of an intelligent composite brain will end up in close alignment with the goals of humanity at large.