Artificial Intelligence (AI) – the capability of a machine or piece of software to display human-like intelligence – permeates our daily lives, often in ways we do not notice. It touches us in myriad ways. Advanced technology operates behind the scenes, powering and optimizing smartphone apps, transportation, healthcare, retail, and more.
In the words of my colleague at EastBanc Technologies, Levon Paradzhanyan, “AI-capable agents have cognitive functions that allow them to observe, learn, take action, and solve a problem or task autonomously. … [AI-powered]] computers can make use of human brain structure to perform natural language processing, image recognition, and much more, in some cases surpassing human expert benchmarks!”
How is this possible? It’s all about math and computing power. The key is to be able to process more complex mathematics faster and with more accuracy. That requires more computer power: vaster processors and more memory to be able to hold more data. And because Moore’s Law, which in 1975 predicted a doubling of transistors on a chip every two years, has largely held up, we continue to have the computational power needed to carry out increasingly complex and ambitious Deep Learning (DL) models and AI algorithms.
Even though machines can solve certain tasks much quicker than humans, today’s AI remains firmly at its first level, which we sometimes refer to as “narrow AI” because it is limited to handling certain tasks in a very clearly defined – narrow – way. Beyond narrow AI are two more levels, “general AI” and “super AI,” which together offer a theoretical view of what AI may eventually become.
AI has long been a favored subject in popular culture where it has obtained an almost mythical status at times. It’s often dystopian. “Blade Runner” – and the short story on which it was based, Philip K. Dick’s “Do Androids Dream of Electric Sheep” – introduced us to android antagonists called “replicants,” which are generally superior to humans, but some of which – or whom? – do not actually know that they are machines. In “The Terminator,” innovative but reckless humans build intelligent machines, which become self-aware – i.e., autonomous and untethered from human control. Naturally, the machines decide to wage war against humankind in a near-successful attempt at wiping us out. In another genuine sci-fi classic, “2001: A Space Odyssey,” HAL, the onboard computer/spaceship operating system, goes rogue and tries to kill the crew in an act of very humanlike self-preservation. Fascination becomes fear as AI is used to exemplify humankind’s hubris.
Real-life AI is more benign. Driverless commuter trains. Semi-autonomous cars built by Tesla and a rapidly growing number of other innovative automakers. Smart home appliances that “talk” to each other (and us). Financial technology solutions that facilitate payments and incorporate advanced algorithms. Computer-driven coordination of public transportation. All examples of supercomputers running algorithms and making calculations far beyond what any human can do – making our lives just a little bit easier.
One of the most famous examples of just how smart a machine can be was IBM’s “Deep Blue” computer defeating world chess champion Garry Kasparov. How did a computer get the better of the world’s best chess player? Very simply put, IBM’s machine had the ability to crunch through every possible action and outcome on the chess board faster and with more accuracy than any human could. And that – carrying out a clearly defined intellectual task – is where current AI excels. The machines cannot think independently, but when it comes to crunching such numbers, they are far ahead of us.
For all its sophistication, current AI remains “narrow.” To illustrate its limitations, let’s look at some examples. “Deep Blue” took down one of chess’ absolute superstars, but could that same machine defeat the world champion of another board game? Or use its “brain” to handle an entirely different type of human-like intellectual task? Not without further programming. The computing power is certainly there, but the machine would need a human to develop the algorithms and then program it before it could get to work. Think about Google Translate, which has proven immensely useful for travelers, students, and other people who need a quick hand with translation. It covers a wide variety of languages and can often crank out acceptable translations. But try feeding it a colloquialism or some tongue-in-cheek phrase that doesn’t work in direct translation – and Google Translate will return gibberish.
The next level of AI – “general AI” – expands the capabilities of narrow AI to allow it to tackle broader, multi-faceted tasks and interpret human intent on a very basic level, thus enabling more complex and human-like behavior. General AI has yet to arrive, but some technology has the potential to eventually reach that level. Google Search, for example, is becoming better and better at figuring out what we are looking for when we type in a search phrase. It will return a list of computationally-curated search results, which often does include the answer we are looking for. And it keeps learning from user behavior, enabling it to continually improve its accuracy. But it cannot yet interpret meaning. It does not understand irony or humor. It cannot calculate which of multiple meanings of a word or phrase the user actually has in mind. It certainly cannot detect and interpret intonation. Yet. Google’s deep learning capabilities are so good at this point that it just might be edging toward the general AI level.
General AI will be “smarter” and much more versatile than narrow AI. But it will not include self-aware, autonomous entities that work independently of humans and surpass – or even threaten – us. That futuristic and far more advanced level is called “super AI.” What it will look like is largely unknown, but super AI is where machines become super-intelligent. Superhuman, even. Super AI machines will solve any intellectual tasks better than any human specialist in any given intellectual field. This is also where we might start seeing machines that really do think for themselves, maybe even develop real human qualities like empathy and joy. And if it mirrors sci-fi, let’s hope it does so in a benign way!
While we build more and more human-like AI, we do not actually know what future incarnations of this hyper-advanced technology will look like or mean for us. Our understanding of how the human brain works constantly improves, but we are not quite ready to hand the reins over to the machines yet. As for super AI: We don’t even know how to get there or if we ever will. So, while we remain at the narrow level, the possibilities of AI are limitless. And building AI solutions – even narrow ones – keeps many bright minds busy at EastBanc Technologies and elsewhere. Strap in.