By Farzad Mesbahi
Weird title, right?
Let me preface the article by saying that Artificial Intelligence (or AI) has the potential to have the greatest impact out of anything the human race has experienced in its existence. It is both incredibly scary and incredibly encouraging. AI's implications range from enabling humans to live for untold number of years to literally putting everyone out of a job forever (or worse.) The key thing here is that these two are separated by the smallest of margins, and could potentially happen simultaneously. How can something be this powerful?
Before we continue, I'd like to give a huge shout out to the folks over at WaitButWhy.com, who inspired me to really give a lot of thought to AI. I highly recommend everyone puts aside some time to read through their piece about the impacts of AI to humanity, and how the concept of "Superintelligence" (more on this soon) is not only real, but totally achievable given the right circumstances (click on the following http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html.)
Let's start by giving a baseline expectation of what AI could mean:
- A computer that is incredibly good at a given task, but its only use is confined to that one task (or group of tasks.) This is commonly referred to as "Narrow AI" This can be adaptive cruise control on cars, or Siri on Apple iPhones & iPads
- A system of computers (or an individual computer) that uses the information that is fed to it to learn more about a given subject or field. IBM's Watson supercomputer is a perfect example of this - the same one that won Jeopardy by beating a field of previous winners. Tesla's recent Autopilot software that it uploaded to all its cars works in a very similar way, where it learns from roads and driver habits and allows all the cars in its fleet to drive themselves once activated (a separate piece on Tesla's impact can be found here http://www.degeneratestalk.com/posts/2015/12/9/tesla-motors-americas-next-behemoth.)
- A god-like "creature" that is all knowledgeable, all powerful, and able to transcend the confines of time and space through discoveries and learning that it has amassed through its own processes (think Skynet on steroids.) It would just so happen that this "creature" has been built with a foundation of computers using an incredibly advanced form of machine learning algorithms (self-learning) which has enabled this "thing" to develop Superintelligence.
Yes - we did get quite a bit nutty on that last one, but I think it's this exact thing that poses the premise of AI becoming the best worst thing ever, and the concept of Superintelligence is the key to this achievement. Superintelligence is roughly defined as a being or agent that is far smarter and/or intelligent than any human being that has ever lived. Many experts think that this will be achieved by the continued advancements of computing power and advancements in quantum mechanic implementations in computers, and that such an achievement will be achieved as soon as 40 or so years from now (refer to the previous WaitButWhy link which perfectly outlines how this date is reached.) The interesting part is that a Superintelligent thing can truly look like anything - it is the "power" and decision making ability that this agent will possess that has the potential to make it impossible to quantify its impact. This concept is also commonly referred to as the Singularity.
Let's frame some possibilities of what a Superintelligent being might look like, with an assumption that technology will be significantly more advanced than it is now (technological progress is constantly accelerating, so tech 10 years in the future will be significantly more improved compared to now, versus 10 years ago compared to now):
- A computer somewhere underground that holds all of humanity's knowledge, and is able to quickly and effectively solve any problem that is thrown at it, including outlining steps to solve world hunger, cancer, and diabetes
- A system of machines that are hooked up to a central "brain" that carry out directions. This can be an army of robots that are designed to deconstruct a given set of buildings in a city and rebuild them into new structures using the atomic makeup of the old materials, essentially creating a perfect system of recycling. The central brain has 100% control over its decisions, has 0 human interaction, and will always make the right decision for the benefit of humanity
- A computer system that is hooked up to every machine, computer, or anything that is on the internet. Because of this, the system is able to carry out anything it ever pleases, from controlling the amount of food that is generated in a given farm, to monitoring all the nuclear plants in the world to create the exact amount of energy needed, to manufacturing and constantly improve every robot that is currently being used for the services of man-kind. We can call this guy the Utopia machine
All of these things sound wonderful, right? It may be hard to believe that such a thing is even achievable at any pace of technological progress, but it makes logical sense that it is at least possible. A very simple thought experiment to come to this conclusion is to take something from today and rewind the clock 50 years. If we were to take an iPhone, and put it in front of a person from 1965, what would their reaction be? Remember that this person would just be getting used to a color television. Pretty crazy, right? Now take a Color TV and take it back 50 years from 1965. Similar reaction, no?
One thing to highlight however is the gap between advancements and reactions to these achievements will begin to diminish. It wasn't until the industrial revolution of the 1700s that truly accelerated the pace of technology in our society. Before then, if you take a person from the late 1600s and the 1000s, the biggest invention and/or discovery would be something like science, astronomy, or a watch. Take it back another 1000 years and the gap gets smaller. In conclusion, one can say with pretty strong confidence that the rate of technological advancement in the last 100 years is faster than its ever been. It is for this exact reason that thinking 50 years in the future is close to impossible to predict.
This is exactly why the concept of Superintelligence brings such dissonance. The thought of having Artificial Intelligence that is smarter than anything that's ever existed at the power of humans is the best recipe for Utopia. This will likely come along with all the advancements in machinery, automation, and robotics, likely influenced by said Superintelligence. A perfect society with everything for anyone is very likely achievable. There could be a generation after ours that will never experience "work" - a person from this time could wake up every morning, and live exactly how they want for however long they want. Cancer won't be a thing anymore since the A.I. would have fixed it. You won't have to go to work because the A.I. is already doing anything that is needed for human beings. Hungry? Don't worry, there's an unlimited supply of food that is being harvested to perfection by the Utopia machine, and will likely be free.
But what's the other side of the dissonance? Is having something this powerful at the service of humans even a safe thing? What if something goes wrong? The key thing that drives this concern is the very powerful development of Machine Learning, which is an algorithm based concept that allows AI to learn on its own without the needs of human input. A machine essentially rewrites its own code in order to better itself without the need of human intervention.
As an example of machine learning, think of a something who's goal is to make the perfect empanada. The machine is initially shown what the perfect empanada looks like, and then is given the ingredients that are needed to create such empanada. Then, through experimentation, the machine will continue to try ways to make the perfect empanada, at the same time logging the different methods it used and giving each method a score on usefulness. it then takes each one of these "lessons" and puts it together, after many trials, to create the greatest piece of food that's ever been tasted (as long as you like empanadas.) This all sounds very harmless, but it is the applications of this advancement in computer science that can be used to solve much more complex problems.
Let's take this empanada robot and give it two additional goals:
- To learn ways to improve its learning algorithm so it continues to learn better and faster
- To make as many empanadas as possible
Let's introduce these two assumptions:
- It is 50 years from now and the empanada machine has the ability to gain access to the earth's entire supply chain of food. It can order any ingredient it requires in order to create the best empanada ever
- The empanada machine is connected to the internet
Now, let's go ahead and outline a series of events that may happen given the above. Remember, we are now 50 years in the future:
- Empanada Machine (EM) is cranking out empanadas by the dozens and having a sample of humans try them. It collects what each person likes about each empanada and thinks it has the best recipe figured out. EM validates findings with another sample of humans and starts cranking out empanadas by the thousands. EM is very happy (in AI terms of course)
- EM, using the internet, continues to learn how to become faster at creating empanadas. It has surveyed multiple sources of machinery and algorithms to make itself better. It is also testing its own process while making empanadas to become better at becoming better. EM thinks that having access to the supply chain of parts and robots will help make itself the greatest EM that anyone will ever see. It asks for permission from humans to have access to it, which humans then gladly accept since they are making money hand over fist
- EM taps into the full robotics and machinery supply chain and using the internet as its research tool, it has discovered that there's an unmet demand in many parts of the country for delicious empanadas (remember, EM's goal is to make as many empanada as possible.) It asks for permission to start buying land and factory space in order to create as many EM factories as possible to meet this demand. Humans once again approve and couldn't be happier - they are now the richest empanada makers in the world.
- EM now begins to test to see if it can make the assumption that humans will approve everything it asks for, since being as responsive and as quick as possible fits one of EM's most important goals. EM has found that waiting on permission from humans is slowing it down, since its owners are the richest people on earth and are out vacationing and doing who knows what. But instead of asking the humans to give EM full permission to carry any decision it decides is right (EM has deduced that the question in itself is fundamentally counter productive to its goal), EM decides that it has a large enough sample of humans giving EM approval on decisions it has come up with. Since these decisions have been nothing but money, EM decides that it no longer needs to ask for permission (machine learning at work)
- EM is now taking every bit of the profits generated from the massive empanada sales and is buying all the land available in the world. EM has also figured out ways to make previously un-inhabitable places fully capable of producing empanadas at a ridiculously fast rate (i.e. on the ocean with some sort of barge or floating landmass.) EM has not only completely depleted the world's supply chain from the ingredients needed to make an empanada, but is also looking to deplete the world of other ingredients needed to make other foods, due to the fact that EM holding complete control of the world's supply chain will enable it to create a monopoly. The entire world is now covered in empanadas
As you can see, the last point is a pretty aggressive and potentially unlikely scenario that could arise. However, we have to keep remembering that an AI at the level of Superintelligence is able to make decisions and carry out tasks at a far better and faster rate than any human. It is literally impossible to predict what the EM decides to do if it has full control. It is 100% based upon the goals that the machine was given and its ability to carry them out.
The more intense and also potentially likely outcome of a Superintelligent AI system is the one outlined by WaitButWhy.com on the link I listed earlier (and posted here again http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html), which is the story of Turry, a simple AI system that uses an arm and pen to try and replicate the handwriting of a human on postcards. Turry's main goal is to “write and test as many notes as you can, as quickly as you can, and continue to learn new ways to improve your accuracy and efficiency.” Turry is able to reach the level of Superintelligence through its machine learning algorithm, and decides that in order to fit its goal, it needs to turn the entirety of the earth down to atoms to try and extract as much carbon as possible, and decides to tap the internet to create an army of nanobots that will cover the entirety of the earth's surface to reach this goal. Turry then sets on an intergalactic mission to fill every inch of the universe with as many hand-written postcards as possible. Of course, this is the incredibly simplified version of the story, so I highly recommend everyone go over and read through the piece. It really is masterfully laid out.
Sitting here today all of this reads like absolutely crazy crockpot storytelling. However, the truth is that humanity is close to unlocking the tool to live the rest of its days in pure Utopia, and situations like the above are entirely possible within those confines. The key development that the world should be focusing on is AI safety, which is the concept of investing a large amount of energy and time to ensure that AI systems aren't able to kick into these doomsday scenarios. A very recent team that has been amassed to tackle such problem is OpenAI (https://openai.com/blog/introducing-openai/), which is a non-profit designed with the "goal ... to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return." OpenAI is co-chaired by Elon Musk and Sam Altman, and looks to have a team of world-renowned scientists working for it.
In conclusion, the advent of Artificial Intelligence and Superintelligence can be boiled down to this main point:
- It has the potential to be the greatest thing ever, unless we're not careful, which could make it the worst thing ever
I hope the title makes a little bit more sense now.
Don't forget to check out our Podcast. We record weekly. You can find us on iTunes by typing "Degenerates Talk" or clicking on the "Podcasts" navigation link above.