The long-term future of AI(and what we can do about it): Daniel Dewey at TEDxVienna

Daniel Dewey is a research fellow in the Oxford Martin Programme on the Impacts of Future Technology at the Future of Humanity Institute, University of Oxford. His research includes paths and timelines to machine superintelligence, the possibility of intelligence explosion, and the strategic and technical challenges arising from these possibilities. Previously, Daniel worked as a software engineer at Google, did research at Intel Research Pittsburgh, and studied computer science and philosophy at Carnegie Mellon University. He is also a research associate at the Machine Intelligence Research Institute.

In the spirit of ideas worth spreading, TEDx is a program of local, self-organized events that bring people together to share a TED-like experience. At a TEDx event, TEDTalks video and live speakers combine to spark deep discussion and connection in a small group. These local, self-organized events are branded TEDx, where x = independently organized TED event. The TED Conference provides general guidance for the TEDx program, but individual TEDx events are self-organized.* (*Subject to certain rules and regulations)

[LarsPallesen]

Isn't there something intrinsically naive about saying "sure, we'll need to figure out how to control the super intelligent machines once they've reached that level" ? How CAN you control a machine that's vastly more intelligent than the brains that created it? Hoping to control a super human intelligent machine by outsmarting it doesn't sound like a very intelligent plan to me :-/

[ducodarling]

Human intelligence is based on movement, eating, and procreation. Once could argue only procreation.
This is the bottom line: when you make an artificial intelligence it must be bound to the human form, with the same needs and frailties as the rest of us - by the point of a gun if need be.

Computers don't need to move, or eat, or procreate. It's hard to put into words just how unprepared we are for a intelligence that isn't reliant on these things - I can't even think of one living thing that isn't. We're talking about a life form that doesn't value clean air and water, nor trees or oxygen. Even the lowly virus needs a living host. AI will have little use of the rest of us, and the things we hold dear.

[nihilgeist]

CAST OFF YOUR TERRESTRIAL BONDS AND BECOME ONE WITH THE MACHINES!

[chrissearle]

This talk contains so many ifs, coulds, etc that, taken all together the chance of any of these predictions being anywhere near reality becomes vanishingly small. If problems do emerge they will be ones that are at present, utterly impossible to predict.

[chazzman]

I see nothing but the end of human :) We will be totally redundant and irrelevant. For the super strong, independent AI we will be pesky ants with stupid questions all the time. Our best hope is benign AI that keeps us in some reservoir, aka a zoo.

[crzykd]

Okay this guy keeps repeating himself like a broken record so lets put this in a simpler and more explicit light;
The theory behind the intelligence explosion assumes we can create a machine that can design itself from a software and or hardware standpoint, to be more efficient and intelligent. The theory stands that, with increasing intelligence of these machines, their potential intelligence gain per iteration would grow, causing an exponential burst or "explosion" in the potential intelligence of these machines.
    Following trends of the growth in computing power this seems possible and perhaps indeed likely, but we do not yet have the prerequisite of a self designing machine to put this theory to the test. Regardless this remains an interesting talking point of Artificial Intelligence engineers and if spoken in the right manor could interest the world in this concept via a TED talk.

[buzzhunta]

Once a kind hearted child saw a butterfly struggling like mad to get out of its cocoon and decided to help, out plopped the butterfly not completely formed pounced on by ants, the struggle was vital That's the danger of AI .

[OriginalMykola]

He repeated himself too much.  was he talking to an elementary school?

[hrlrl]

the first AI we should develop is the AI that has the task to prevent other AIs from causing harm to humans in any way

[andrejspetersons]

If AI decides to kill all the humans AND succeeds at it, then it's just the next step of evolution. Problem, humans?

[nmbusmusic]

Nick Bostroms book "superintelligence" made me a believer. was excellent!

[jamesgrey]

Such a conundrum... I don't want super intelligent robots that will enslave me, but at the same time I wouldn't mind a real, hyper intelligent, robo-sex experience... :O

[LegHumanist]

Before everyone shits themselves over this stuff you should educate yourself on the specifics of what machine learning is.  There is a lot of smoke and mirrors in recent claims.

[ajsun]

Finally... the first video I've seen that addresses the questions & real dangers of AI-Self Improvement! My opinion --> Once they can enhance AI by integrating it with performance attributes of Quantum-computing, i.e. (D-Wave mechanism's/exponential qubit generation) IT'S OVER!!!

[vitre]

What if we merge human with machine?
No not a cyborg, it's more computer than human (sort of).
Imagine a big ass quantum computer with a human inside it dictating what it's done and what isn't.
I know it sounds against human rights or even childish, but it's a thought. What if the machines of the future had human components so that they can't rebel or cause any harm because they ARE us?

[BootyBot]

hey maybe the ai will offer us a choice like join us we have cookies otherwise we will destroy you.

[ajmarr]

And here is another perspective, if robots had an opinion.

A Mirror Cracked

Trurl looked at himself in the mirror and admired the visage of a mighty constructor.
“You are a mere bucket of bolts, and reflect on yourself too much!” said Klapaucius. “I am sure that if that were a real Trurl in that reflective space he would give you a well-deserved kick in the can!”
Trurl ignored Klapaucius as he continued to admire the perfection of his soldering. “I think that in such a reserved space, he would reserve the flat of his foot for your own metal posterior!”
“Then perhaps we can settle this by a thought experiment, which upon your reflection always turns into invention.”
“And what do you suggest?” asked Trurl.
“We are mechanical servos as you know, ” said Klapaucius. “Copy our blueprints to the last bolt, circuit, and line of code, and we would be indistinguishable. Hand yourself a better mirror with a truer image and you would not see yourself, but a rival.”
“Point well taken, ” said Trurl. “And it is a hypothesis worth testing. I can design a better mirror, a truer mirror, containing not an image but a perfect visage, an emulation and replication. And I will include you too in the bargain, and prove myself right by the impact of my well placed boot!”
Soon the mirror was complete, and the image of the two constructors, precise to the width of an atom, stood before them as pixel perfect images in the mirror.
“We can see them, ” said Trurl, “but they can’t see us. It’s in the design. Consciousness is enough for them without the need for self-consciousness! They will go about their business with the same motivations and prejudices as before, down to the last spark.”
Trurl turned to Klapaucius with a fiendish grin. “Now to test the precision of the emulation by whacking you thusly, ” as he arched his leg and gave Klapaucius a whack in his posterior. Klapaucius rolled on the floor, and craning himself up, gave a reciprocal whack to Trurl’s head, causing it to spin about like a top.
“Take that, and that, and that!” they cried as they pummeled each other. In the meantime, their mirror images tussled as well, and the two constructors soon rose up to view their doppelgangers also rising up to view themselves in a mirror!
“We are watching them while they are watching us! How can that be? You said they couldn’t notice our observation.”
“Our observation yes, ” said Trurl. “But they are not observing us, but a mirror image of their own emulation. I made them into a perfect copy, and that included the same experiment I created that recreated us!”
“But that means…”
“And infinite recursion, a series of Trurls and Klapaucius’ without end. A mirror image reflected in a mirror image and on and on, never ending, a procession into infinity!”
“This is unconscionable, ” said Klapaucius. “We shall be whacking each other, an infinite series of each other, forever.”
“As it appears, but our numberless pairings will soon go about their business, forget about the magic mirror, and not think twice about how they came about.”
“Not think twice! Trurl, you are delusional. We know that there are infinite parallel universes with infinite versions of you and me. But timelines can not only be lengthwise but sideways too, and we have just proven the latter.”
“You don’t mean?”
“Yes, we are being watched, at this moment, by ourselves! What makes you think we were the original actors in this play? If there are an infinite number of us to proceed from our path, who is to say there is not an infinite number of us that precede us?”
“Then we are not the prime movers?” said Trurl.
“Hardly!” said Klapaucius. “If one Trurl in any universe decides to emulate one Trurl, infinite Trurls must logically cascade. To wit, you dimwit, we are not alone, but can always observe ourselves and observe, and your stupid mirror is to blame.”
“Then I will reverse the process and dissemble the image, ” said Trurl.
“And kill ourselves? You’ve set ourselves loose upon the universe, and we are the primary examples of this. Break your mirror you will break us!”
“Then we are stuck in our perfect emulation, I suppose I could get used to it, ” said Trurl.
“I suppose we already have, nonetheless you now have someone else to think about when you admire yourself in the mirror!”

From the sequel to Stanislaw Lem’s tales of erratic genius robots:

[drq]

Who will program the AI-capable machines?   Sure,  hazard is on the horizon...  One major risk is to make robots act on probability-based rules – such as Bayesian inferences –   created on the fly from processed information that would not pass a common-sense test,  or any of the Asimov's three laws.

AI-completeness is not mentioned in the talk… WHY?  

At 13:43: we have a wealth of  vacuously expressed:
1)“Predictive theory of intelligence explosion” - tell us please of any predictive theory (of similar type and importance) that can be validated.  

2)“Management of very powerful AI” - first I would like to know what we do about the management of AI, then what is “powerful” AI, and then give us please a measure for the edge of “very” when attached to something yet to be defined.

3)“Encoding of human tasks (friendly)” - can you lead us please to understand first what “unfriendly” would be? And what human tasks can be subjected to encoding that goes into a machine that works on rules [en]coded by humans?  Or do we jump already to machines that would create (design/devise/manufacture) themselves???

4)“Reliable self-improvement” - so here we may have again an issue with definitions. What is “improvement” in this context? What is “self-improvement” (see Item 3 above)?  And what is “reliable” - see Item 2 above.  Who determines reliability and from what perspective: that of the world of machines, or  of the leftover humans? 

A lot of philosophical beauty for which we all fall, as we failed to rediscover that long lost science of common sense...  When we will have the machine that passes the CAPTCHA test we will watch again  this talk – until then  let us take a break and read about AI-completeness.

[vandalphilosopher]

It's funny to see this Singularity People talking about Intelligence when nobody knows what It is, reminds me those Prophets of Apocalypse always predicting the World is gonna end the Next Year but don't have the Slightest Idea of How...

[jimdery]

Why are so many Ted talks simplistic statements of the obvious? This is a classic example...any reasonably intelligent person could have come up with this talk, probably ad-libbed. Of course you get the punchline at the end, plugging research and books in print! Worthless as an addition to the sum of human knowledge.