Will the Singularity happen? I’m currently reading an international spy techno-thriller pot-boiler whose premise centers around the creation of Artificial General Intelligence (AGI), just asTerminator: Dark Fate is raking in big bucks in theaters. Scary futures are big entertainment business. It’s a perfect time for a provocative question like the one Fandango asks today:
Do you think the singularity will occur? If so, what time frame do you think it will happen in and how will it impact humanity? Alternatively, do you think or care at all about the potential for reaching singularity?
The short answer is: World-threatening technology is perpetually created by humans. Humans then create an alternative to pull civilization back from the brink. Lather. Rinse. Repeat.
Why We Should Fear the Terminator, aka Alexa
There are three camps on the future of AGI: the pessimists, the optimists, and the skeptics. Each argument has merit.
Here’s an example of why we might want to be scared. In a recent study by researchers in Japan and at the University of Michigan, lasers were able to take over the security systems of Google Home, Amazon’s Alexa, and Apple’s Siri. From football-field distances away, experimenters opened garage doors, turned light switches on and off, and made online purchases. Other researchers have discovered ways to send recorded conversations, collected by these voice-assisted devices, back to unauthorized third parties. Meanwhile, my voice-operated TV set still doesn’t recognize my commands half the time, even when spoken in exactly the same way, but it knows everything I’ve watched. This is how technology can sneak up on us, with seemingly-innocuous devices designed to be fun or make our lives easier.
The pessimist view encompasses several strong points:
- Technology can be hacked (Alexa lasers)
- Humans don’t agree on what’s a good idea (just think of presidential approval polls)
- Not all technology is good (mustard gas)
- Programmers can be biased (Amazon hiring algorithm)
- Humans always invent weapons ( machine guns, mustard gas, atom bombs, drone strikes, etcetera etcetera etcetera)
One rising concern with algorithms is their reflection of the inherent bias of the coder. For example, Amazon programmers were asked to design a recruiting program to sift through resumes. Within a few years, Amazon had to shut down the program when they discovered it had an inherent bias against choosing women’s resumes over men. The algorithm was opening them up to a gender discrimination lawsuit. The program itself used smart, machine-learning (ML) logic, but it learned based on the input of “model” resumes it was given, which were from the existing predominantly-male engineers. Humans, to be fair, exhibit such bias all the time. We gravitate toward people like us; we just don’t usually turn it into explicit code.
The code problem, for me, isn’t that the machines will bootstrap themselves into thinking that humans aren’t necessary or are harmful. The problem is that machines are constructed from programming designed by humans who are short-sighted, use poor judgement, and can’t predict the unintended consequences from their choices. Not to mention humans who are fanatical, vengeful, and/or cruel.
But the very essence of these risks carry the solution as well: the spark of humanity.
What a Piece of Work Comes from (Hu)Man
In 1978, when the first baby was born from in-vitro fertilization, the hospital was besieged by bomb threats and paparazzi, and even the Vatican expressed concerns for the future. Forty years later, reproductive technology has been a boon to many people (to a degree, such technology led to the creation of my brilliant handsome children, so I’m for it). We don’t seem to be overrun yet with a clone army of Boba Fetts. Technology, from small pox vaccines to gene-therapy for cancer, has the capacity for immense value to our species.
AI programming is in use that helps detect fraud on your bank account, filter spam out of your email, and show you the fastest route home. ML programming is the way rideshare apps connect you with a driver (not to mention multiple driver options with different prices and car sizes). We’re surrounded by AI and ML programming already, so to decide that we’ll just stop using it is to think we can put the genie back in the bottle. Even if you think you’ll just go “live off the grid,” how will you power your tractor? Can you find un-genetically-modified feed for the un-hormone-filled oxen that will you want to pull your plow?
The optimist argument has its own case to make:
- Not all technology is bad.
- You can’t undo technology just by avoiding it.
- Even if machines learn, and learning mimics creativity, the creativity rests on a previous model; there is no spark.
- No matter how complex the algorithm, I’m not an algorithm.
One of the current concerns mentioned in pessimist arguments is about how AIs can create art. Not only is there a burgeoning art community that deploys AI by existing artists (example here), but there are also cases of researchers programming AI to create artwork that experts can’t distinguish from human artists. In a thought-provoking analysis on Artificial Intelligence: What’s to Fear? by Ronald Dworkin, he raises the spectre of machine-created art, but his solution is over-simplified. He quotes Leo Tolstoy that art is defined as “a human activity” that conveys feelings; no human, therefore no art. He says what the machines create isn’t art, by definition. That’s cheating. My argument would be simpler. Even if a machine can effectively mimic past art to design newly-pleasing art, it can’t go outside the programming to do something new. In other words, it can’t be an Edward Hopper.
Hopper was a painter in the 1920s, during the rise of abstract expressionism, which followed the impressionist movement. While his style was realistic, he had a unique ability to paint people in quiet moments, which emphasized isolation in a way no one had done before. He wasn’t particularly a “product” of current or recent artists. You would be hard-pressed to feed a computer enough recent paintings for the algorithm to create this new Hopper style. The AI could mimic Mondrian or Monet, but never create a Hopper.
Resistance is Futile. Redirection is Imperative.
Because machines can’t completely replicate human creativity and consciousness, there will always be the capacity for humans to surpass machines. This is where the optimist and the agnostic can join forces to create a solution to ensure that potential harmful effects from an AI singularity don’t occur or are minimized. It’s not as simple as suggesting that we stop using technology.
For example, one of the loudest current anti-AI arguments focus on concerns that machines will replace humans at specific jobs. The machine will put people out of work, like truck drivers. Yet we don’t spend much time today wondering where the buggy-whip manufacturers went or what telephone operators are now doing. What did people in the Geek Squad or producing podcasts used to do? My first job involved three days standing at a xerox copying someone’s entire Rolodex. No one, including me, is crying over the technology that has replaced it. Simply resisting technology isn’t sufficient.
A smarter approach is to keep informed about what technology is doing and maintain a healthy skepticism. First, we can ask questions. What is Amazon doing with the data collected by Alexa, and can they be held to account? What are the safeguards against developing gene-based weapons, and who’s doing the monitoring? Secondly, we can ban certain uses of technology, and while it may not prevent their invention or even use by the worst actors, we can at least agree on some with obvious harm. The use of mustard gas is now prohibited by international law. That’s a start.
Several scientists are also focusing on development of a positive AGI to combat a harmful AGI, if and when one comes into existence. The idea is that if programmers can create an AGI that values and protects humans, it could counteract an AGI that arises with malevolent intent or decides to eliminate humans. In simple words, humans make bad machine? Humans make good machine, too.
Singularity: a hypothetical future point in time when technological growth becomes uncontrollable and irreversible, resulting in unfathomable changes to human civilization.Fandango, Provocative Question of 11/6
Technology growth has always increased in ways to threaten humanity with dire consequences. The invention of the machine gun and the airplane changed the nature of war; the splitting of the atom either ended a war or perpetuated a new one, depending on your point of view.
Somewhere in the first few centuries A.D., the first stirrup appeared. Where and when is subject to some debate, but probably China around 200 A.D. But it was the Mongol army, able to stand up and fire arrows in retreat, that made particularly effective use of the new technology. Scholars suggest that their widespread success in conquering a great deal of Asia and making inroads into Europe was due to this terrible weapon. As medieval civilizations eventually fought back the Mongols, they took the stirrup and put it into use for themselves,. This allowed heavily protected knights the ability to move on horseback and led to the domination of cavalry-based warfare (the Norman French) against others. William Then Conquered.
The stirrup was its own kind of singularity in medieval technology. It changed the nature of war, communication, art (chivalric love poetry), and some even think genetics. The solution for the non-stirrup societies was to make their own in order to return to a place of power equilibrium. Civilization was not the same afterward.
Self-aware, machine-learning AIs are already here. We need to learn to co-exist with this technology rather than live in denial or fear. Civilization will not be the same afterward, in any event. If nothing else, there should be plenty of future jobs for people to monitor the AIs.