AI: Artificial Intelligence. There are countless examples of movies, shows, and literature that feature a galaxy-brained computer in the mix somewhere. However, It might come as a surprise that the idea of a constructed being capable of human thought isn’t a new idea resulting from the technological revolution. From the Greek myth of Talos to the Golem in Jewish folklore, there has long been a persistent curiosity with an Other that can match us on intellectual grounds.
But these are all just stories, right? What does AI look like in reality? How do the people who make AI think of them? Here, we’re going to broadly go over what AI is and what it can be.
What is AI?
AI is an umbrella term that encompasses all types of inorganic, constructed intelligence that are analogous to human intelligence. AI is also a category, a science like chemistry or physics, rather than just a “thing”. As a discipline, it started in 1956 and its goal then as it is now is to create machines and programs that can do tasks that are considered “uniquely human”.
AI as a concept can be applied to any program or machine that resembles human intelligence in a way. AI performs tasks using algorithms: a set of rules that it uses as instruction on how to “adapt” to changes in the environment that it is performing its task in. In the broadest sense, these algorithms determine what it can do and how it can go about doing it.
Categories of AI
Like we stated before: AI is an umbrella term that includes many different ways of approaching and making thinking computers. Many things you might think of as separate from AI (for example, Machine Learning and Neural Networks) are included as small subsets of the broader category. However, in computer science we have a few terms used to describe how much “I” is in a given AI. Keep in mind, these are not hard and fast rules, but general ways of conceiving where AI is at.
Weak AI (AKA Narrow AI) is any machine or program designed to perform a specific task or set of tasks to a level of proficiency that it seems “intelligent”. These AI are made to be able to perform tasks (and even adapt to changes in their task environment to a reasonable degree), but that’s it: Weak AI does not have the ability to perform any task other than the few it was designed to do.
Not only that, but it cannot learn to do other tasks, nor can it truly “comprehend” what it’s doing, and therefore only seems analogous to human intelligence because of its output. With this in mind, arguably all AI up to now has been weak AI.
Weak AI is also a sort of philosophical approach for some people in computer sciences. The argument being that there are so many complex, interlocking aspects to the sorts of intelligence humans use to complete tasks that an AI can never truly have this sort of intelligence. Therefore AI can only ever be a simulacrum of human intelligence at best.
Whether you know it or not, you’re probably familiar with examples of weak AI. For instance, the Virtual assistants on your phone (Siri, Cortana, Google Assistant), or in your homes (Alexa, Google Home) are examples of Weak AI.
Strong AI (aka General AI) is everything Weak AI isn’t: an artificial intelligence that is truly human-like in its ability to learn, comprehend, and perform tasks. Hypothetically, A Strong AI would be logically and emotionally intelligent, able to make its own decisions, and be
Strong AI as a philosophy differs in interesting ways from Weak AI: in that it doesn’t provide a counterpoint to the claim that it is impossible to put the immeasurable spark of consciousness into a computer. Instead, the philosophy of strong AI posits that the human mind is analogous to a chemical computer, and that with enough resources it would be possible to perfectly simulate it with computer hardware and software. Such a simulation could be functionally indistinguishable from a flesh and blood human.
One of the simplest methods of evaluating if an AI is strong is the Turing test. Proposed by Alan Turing, the test has a human interrogator question and evaluates the responses of 2 hidden figures: one human, one machine. If the Interrogator cannot distinguish which one is the real human, then the AI passes the test and would be categorised as a Strong AI. One problem with this as an evaluating tool is its simplicity and scope, since a Weak AI who’s only task is to perform human-like conversation would be falsely categorised as Strong AI. There is a real world example of this: Google Duplex. Duplex’s job is to make phone appointments on behalf of people, and is so successful that you likely wouldn’t know you’re talking to some software. But, when push comes to shove, this still isn’t the first case of a Strong AI
This is likely the type of AI you’re most familiar with as a result of them playing the bad guy in some famous stories. Skynet, HAL 9000, and Roko’s Basilisk are 3 notable instances of a fictional computer so powerfully intelligent that leaves humans in the dust.
Superintelligent AI can be broadly defined as a Strong AI who’s intelligence greatly exceeds that of a human in every practical domain. The mechanisms by which it would attain this intelligence are by its very nature unknown to us, but many believe it would likely improve itself (and by extension improve the methods by which it improves itself) at an exponential rate.
While strictly hypothetical, a Superintelligence would pose an epistemological threat to our species’ relationship to the nature of truth and objectivity, and the entire realm of thought. Most of the time though, concerns are around much more pragmatic and literal threats. It could easily go the other way, and a Superintelligent AI could come about with goals that align with our own, but it’s likely we’ll never know either way.
It goes without saying though that we aren’t really interested in making that last sort of AI at GistLens. We’re focused on using the cutting edge of computer science as a tool to solve problems for businesses big and small. AI is such a large area of computer science that we’ll likely touch on different aspects of it in the future, but for now we hope you’ve gotten the gist of what AI is.