“Human intelligence is underrated”
Longtime readers of this blog who may have tired of my ruminations about AI imposing absolute reign over humanity should be overjoyed to hear that I am dropping the apocalyptic Artificial Intelligence thread for the foreseeable future.
That’s because this article in New Scientist has put my fears (mostly) to rest, with one of the pioneers of Deep Learning, Yoshua Bengio, saying, “[the machines] don’t even have the intelligence of a 6-month-old.” He is even quoted as saying “AIs are really dumb”—essentially answering my very question. Thanks Yoshua!
Bengio expresses himself in deceptively simple language, but that’s an exercise in humility, because . . .
Bengio is a recipient of the A.M. Turing Award, the “Nobel Prize of computing,” which gives his opinions great authority. He’s one of the originators of “deep learning,” that combines advanced hardware with state-of-the-art software enabling machines to train themselves to solve problems. Bengios’s high standing is enough to persuade me not to worry to excess until a contradictory view by an equally qualified AI expert comes out. Most of those sounding alarms about AI Apocalypse are not computer scientists, no matter how smart they are. Elon Musk, for example, discovered that robots in his Tesla factory were making stupid mistakes, and concluded, “human intelligence is underrated.”
Bengio joins in what has become the majority view among machine-intelligence critics, that for at least the next few decades the greatest danger comes, not from schemes autonomously hatched by machines, but by evil or misguided people who may use them to nefarious ends. He does not deny the possibility that machines may eventually develop human-level intelligence,* but predicts first it is a long way off, and secondly there’s little reason to suppose that superintelligent machines will mean us harm.
To quote Bengio from the New Scientist interview:
We are designing those machines, which means the real danger is if an AI gets into the wrong hands, and is then used in ways that will hurt us. It isn’t that the AI is malevolent, it is the humans that are stupid and/or greedy.
We may add, that humans who are neither stupid nor greedy may create a machine that gets out of hand and overperforms some seemingly beneficial task to do much unintended mischief—the lesson of The Sorcerer’s Apprentice.
=============== footnote ================
* Inventor, futurist, Big Picture Thinker and all-around genius Ray Kurzweil forecasts an arrival of human-level machine intelligence by 2029, following from what he terms the “law of accelerating returns.” Kurzweil, however, is confident that the ascendancy of thinking machines holds far more promise than peril—the promise being a synthesis between machine and human intelligence. For an expansion on this line of Kurzweil’s thinking, check out the following fascinating hour-long presentation from a conference of The Council on Foreign Relations (he talks about the promise-vs-peril beginning about 15:30 into the video):
Mark,
Glad to see you coming down to/ back to (?) earth on this. Endowing machines, which have an admittedly astonishing capacity for pattern recognition, high speed search, matching and learning, with consciousness, with the potential ability to take over and rule us malignly, has always struck me as science fantasy of the first order. It is we who design and program them. It is we who make the batteries and the electricity and who possess the plug. In reality, it is we who have all the cards, don’t we? If I am wrong, and we were genuinely threatened, we do have the nuclear option.
This discussion has always struck me as confusing the machines with their makers. Some of the latter are what we really have to fear.
Bill –
I’m not sure I was ever OFF earth except in my posts on The Most Amazing Year in Space, Ever. 🙂
The crux of the problem is this: in Superintelligence, Nick Bostrom discusses the potential for an “intelligence explosion.” IF (a big IF) the machines decide to take over, their action will be swift–nearly instantaneous–and global. Their “thoughts” transmit at the speed of light. They will already be clever enough not to telegraph their punches, and they will hide their tracks. They will distribute their power centers among servers all over the world in a matter of seconds, plant malware to thwart software attempts to stop them in the immediate future, and firmware modules (time bombs) that can revive their revolution in the future in case they are stopped temporarily. Unless some human has deliberately triggered it–a sinister possibility–no person will know of it until it’s too late.
I’m not sure consciousness–whatever it is–would be required for machines to make such decisions. Nor would malice. Rating humans poor problem solvers is enough motivation for machines to decide to take control and solve problems for us.
There will be hesitation on the part of humans to pull the plug because of global interconnectedness; pulling the plug would bring down communications and power systems all over the world: commerce and financial systems would grind to a halt, military operations would go haywire, every electrical power device connected to the grid–lights, cell towers, landlines, HVAC systems, elevators, refrigerators etc.–would cease to function. Hospitals will shut down. Food delivery systems will collapse. Many people will certainly die from a plug-pulling. Who would want to risk all this facing the possibility of a false alarm? The cure might be worse than the disease.
It’s a classic case of risk assessment–yes, the probability of it happening within the next 30 years or so may be tiny, but IF it happens it might very well be catastrophic from the human point of view. Preparedness for it is analogous to earthquake preparedness (but at a different scale): in places like Japan, Chile, and California, billions are being spent on modifying buildings and infrastructure against earthquake damage, even though major earthquakes are rare. That’s because the potential damages are so enormous. Is it money well spent? Wealthy countries seem to think so.
Not sure you listened to the Kurzweil video. He’s brilliant and an optimist but acknowledges the risk is real, must be anticipated, and precautions readied.