AI is coming, and it will be boring
I was asked about my opinion on this topic, and I thought I would have some profound thoughts on this. But I ended up rambling, and this post doesn’t really make any single strong point. tl;dr: Don’t worry about AI killing all humans. It’s not likely to happen.
In an interview with the BBC, Stephen Hawking stated that “the development of full artificial intelligence could spell the end of the human race”. Whereas this is hard to deny, it is rather trivial: any sufficiently powerful tool could potentially spell the end of the human race given a person who knows how to use that tool in order to achieve such a goal. There are far more dangerous developments - for example, global climate change, the arsenal of nuclear weapons, or an economic system that continues to sharpen inequality and social tension?
AI will be a very powerful tool. Like every powerful tool, it will be highly disruptive. Jobs and whole industries will be destroyed, and a few others will be created. Just as electricity, the car, penicillin, or the internet, AI will profoundly change your everyday life, the global economy, and everything in between. If you want to discuss consequences of AI, here are a few that are more realistic than human extermination: what will happen if AI makes many jobs obsolete? How do we ensure that AIs make choices compliant with our ethical understanding? How to define the idea of privacy in a world where your car is observing you? What does it mean to be human if your toaster is more intelligent than you?
The development of AI will be gradual, and so will the changes in our lifes. And as AI keeps developing, things once considered magical will become boring. A watch you could talk to was powered by magic in Disney’s 1991 classic “The Beauty and the Beast”, and 23 years later you can buy one for less than a hundred dollars. A self-driving car was the protagonist of the 80s TV show “Knight Rider”, and thirty years later they are driving on the streets of California. A system that checks if a bird is in a picture was considered a five-year research task in September 2014, and less than two months later Google announces a system that can provide captions for pictures - including birds. And these things will become boring in a few years, if not months. We will have to remind ourselves how awesome it is to have a computer in our pocket that is more powerful than the one that got Apollo to the moon and back. That we can make a video of our children playing and send it instantaneously to our parents on another continent. That we can search for any text in almost any book ever written. Technology is like that. What’s exciting today, will become boring tomorrow. So will AI.
In the next few years, you will have access to systems that will gradually become capable to answer more and more of your questions. That will offer advice and guidance towards helping you navigate your life towards the goal you tell it. That will be able to sift through text and data and start to draw novel conclusions. They will become increasingly intelligent. And there are two major scenarios that people are afraid of at this point:
- That the system will become conscious and develop their own intentions and their own will, and they will want to destroy humanity: the Skynet scenario from the Terminator movies.
- That the system might get a task, and figure out a novel solution for the task which unfortunately wipes out humanity. This is the paperclip scenario— an AI gets the task to create paperclips, and kills all humans by doing so — , which has not yet been turned into a blockbuster.
The Skynet scenario is just mythos. There is no indication that raw intelligence is sufficient to create intrinsic intention or will.
The paperclip scenario is more realistic. And once we get closer to systems with such power, we will need to put the right safeguards in place. The good news is that we will have plenty of AIs at our disposal to help us with that. The bad news is that discussing such scenarios now is premature: we simply don’t know how these systems will look like. That’s like starting a committee a hundred years ago to discuss the danger coming from novel weaponry: no one in 1914 could have predicted nuclear weapons and their risks. It is unlikely that the results of such a committee would have provided much relevant ethical guidance for the Manhattan project three decades later. Why should that be any different today?
In summary: there are plenty of consequences of the development of AI that warrant intensive discussion (economical consequences, ethical decisions made by AIs, etc.), but it is unlikely that they will bring the end of humanity.
- Jaron Lanier on the dangers of the Myth of AI, with answers from Steven Pinker, George Dyson, Stuart Russel, Rodney Brooks and others
- Elon Musk warns about AI (Guardian article)
- Bill Joy on why the future does not need us
- Oren Etzioni on why he is not afraid of AI
- Ethics of AI on Wikipedia
Published originally on Medium on December 14, 2014
Start the website again
Gödel and physics