Siri debuted in 2011 as one of the first intelligent personal assistants. Since then, personal assistants have become an integral part of the smartphone experience, providing a way to more efficiently interact with the hand-held device. They can perform simple tasks like taking notes and setting reminders. While this doesn’t seem like much in 2016, these features were ground-breaking a mere five years ago.
Today, Facebook uses facial recognition that makes tagging friends easier, while Netflix and Spotify use learning algorithms to suggest your next favorite movie. All of these examples show how various areas of artificial intelligence are already having a deep impact on our lives.
So, why are the major advances in the field of artificial intelligence often portrayed as the coming of an impending apocalypse in popular media? How likely is the threat of a robot uprising?
While it is easy, and sometimes even profitable to reach frightening conclusions, the answer itself is not simple. And to reach it, a few things need to be understood.
The first of which is the definition of artificial intelligence. The Merriam-Webster dictionary defines it as “a branch of computer science that deals with simulation of intelligent behavior in computers.” Prof. Bart Selman, computer science, said he agreed with this definition but also said that it could be thought of better as a sum of parts.
“The ultimate goal [of artificial intelligence] is to really model human thinking, human cognition and maybe even go beyond that. But we generally break it down into sub areas,” he said.
He said that it is these sub areas that have had “quite a few successes.” For example, computer vision, has hugely improved the programming of self-driving cars. These cars are no longer only dependent on GPS to navigate, and now detect other cars and objects in their present environment. Beyond personal assistants on phones and laptops, Google’s introduction of voice editing in Google Docs is an instance of the increased modes of technological interaction beyond typing. Machine learning, another sub area of AI, has been helping suggest your next favorite song.
So, while the definition might suggest that the goal is to develop a single fully intelligent machine, it would be incorrect to assume that it would necessarily happen. Innovations are often born out of necessity, and a fully intelligent machine doesn’t necessarily solve any of the problems that we currently face.
However, combining these sub-areas to help perform various tasks is very much a reality. Selman said that we will see the emergence of partial intelligence first.
“The self-driving car is a good example. It is already working but will be perfected over the next two years,” he said. “So there you have an intelligent task — driving, that will be taken over by a machine.”
Another thing to understand is that people and machines exist in a symbiotic relationship with one another. The Google’s Go player, Alpha Go’s win against the Go grandmaster Lee Sedol was as much a learning opportunity for the players as it was a milestone for AI.
“What Alpha Go did is they combined that learning by literally having it play millions of games, and combine that with a previously developed technique in terms of reasoning and game play,” Selman said, “Alpha Go is seen as combining the recent advances in AI and is taking them to the next level.”
It learned Go by watching millions of human games, and then some more by playing against itself. It may now be our turn to learn from it.
“There is a certain sense in the community that given how dramatically Alpha Go won, that human Go playing is not close to optimal, and that we have developed a style of play over centuries, that is very good, but not as comprehensive as it could be.”
Selman points out that this is currently speculation, but probable since better forms of gameplay have developed for other games too.
The human sense of understanding is very different from the way computers currently process information, though it is not for the lack of trying. A lot of human communication is context-based. Selman pointed out that though we tend to think that in communication, most of the meaning is in the words, we forget how much is much is actually unsaid.
“So, in natural language we communicate very little with the amount of words or text we exchange, because we rely so much on this common sense knowledge,” he said. “It’s so obvious, that it’s not possible to get it from the web.”
This kind of understanding is not yet possible for computers. Selman said that while companies are working on it “like crazy, they [computers] don’t really know what a bookshelf is.”
And so to attain this human-like understanding, companies are working on building rich knowledge bases — not databases as they are not just facts — but knowledge, things that are related to each other.
So what would this understanding mean for consciousness of the machine or the program?
The answer to that question depends on how we define consciousness.
“We don’t quite know what conscious is, but it is reasonable to expect that once machines start to get our level of understanding, they might start to ask questions about themselves, and that they are a machine,” Selman said.
So coming back to the question of a robot uprising, while it is not out of the view of possibility, it seems unlikely.
However, this technology and its massive potential for change does raise important questions about the future of our society. How different would the job prospects be in the future if a lot of the work could be done by machines? How would you measure professional accountability in life threatening scenarios? What would automation mean for the military?
As evidenced by the debate that surrounded the FBI’s demand for Apple to unlock a mass shooter’s iPhone, the 21st century’s clash between technology and policy is not easily resolved.