There's No Fire Alarm for Artificial General Intelligence • Hacker News Board

Re: There's No Fire Alarm for Artificial General Intelligence

#121

Post by pdimitar » Sat, Oct 14, 2017, 3:59 PM UTC

> They will believe Artificial General Intelligence is imminent:

(A) When they personally see how to construct AGI using their current tools. This is what they are always saying is not currently true in order to castigate the folly of those who think AGI might be near.

This struck a nerve. Too often, in many scientific disciplines, and even in informal conversations, the people who always demand 100% clear evidence use this fallacy to shut down discussions. (They very often come off as not impressed with the evidence even if it exists and is presented to them as well.)

HN also has a huge camp of such discussion stoppers, even for topics where you CLEARLY have no way to have 100% clear evidence -- like the secret courts and the demand to spy on your users if you're USA based company; thousands more examples exist. Many discussions are worth having even if you don't have all the facts. We're not gods, damn it.

That was slightly off-topic.

Still, I find myself in full agreement with the article and I like the attack on the modern type of shortsightedness described in there.

Also, this legitimately made me laugh out loud:

> Prestigious heads of major AI research groups will still be writing articles decrying the folly of fretting about the total destruction of all Earthly life and all future value it could have achieved, and saying that we should not let this distract us from real, respectable concerns like loan-approval systems accidentally absorbing human biases.

Re: There's No Fire Alarm for Artificial General Intelligence

#122

Post by ollin » Sat, Oct 14, 2017, 4:30 PM UTC

ianai wrote: post #111

mark_l_watson wrote: post #62

Great read, and I don’t mind at all that the last section was a pitch for donating to MIRI. I have been an AI practitioner since 1982 and have enjoyed almost constant exposure to people with more education and talent than myself so I feel like I have been on a 35 year continual learning process. I think that deep learning is overhyped, even though using Keras and TensorFlow is how I spend much of time everyday at wor…

Got any suggestions for a knowledge source of AI at a 100-1000 foot view? Ie not stuck in the weeds, but enough to know what’s going on nd where.

Maybe skim https://www.getrevue.co/profile/wildml periodically?

Re: There's No Fire Alarm for Artificial General Intelligence

#123

Post by gooseus » Sat, Oct 14, 2017, 4:30 PM UTC

johansch wrote: post #120

astdb wrote: post #28

Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. But the catch is the unreliability of (even expert) predictions on technology futures. My take is that it's worth taking pragmatic steps towards studying AI safety measures (i.e. OpenAI), but not going as far as to talk the likes of 'AI research regulation'.

> Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. (Just venting here, not even primarily at you.) 360k babies are born each day. Clearly it is possible to reproduce intelligent machines. The only way it would be impossible to artificially do the same is literally if life was a magic, non-physical thing. I wish people who state things like this…

Just because we can observe something happening, doesn't mean we can understand the mechanisms of how that thing happens and even if we CAN understand the mechanism, it doesn't mean the mechanism is feasibly reproducible within our resource capability.

An example might use crypto... you observe random information flying through the air, you may recognize it as an encrypted channel and you may see a machine acting in response to this encrypted signal.

With enough observation you may be able to mimic the encrypted signal to get the machine to act in a certain way, but you haven't decrypted the actual signal (and can't ever, if you believe in strong crypto) and can't ever say with any certainty you know the full scope of communication taking place or the capabilities of the machine you've been observing.

At any point you can make your own version of the machine mimicking the language tuning it to be an exact replica of the original and even responding to the original signal. Yet, is it truly a copy of the original?

Re: There's No Fire Alarm for Artificial General Intelligence

#124

Post by amelius » Sat, Oct 14, 2017, 4:37 PM UTC

marcosscriven wrote: post #67

icebraining wrote: post #45

Earlier quoted context omitted.

Elevators have pretty good safety mechanisms. Even cutting the cable wouldn't have killed anyone. I'd probably leave too, but just because I wouldn't want to get stuck inside it if it stopped in the middle of two floors, especially since it was full. But for fear of death? Nah. From Wikipedia: "In fact, prior to the September 11th terrorist attacks, the only known free-fall incident in a modern cable-borne elevator h…

The gym I used to go to had an incident with someone being killed by the lift falling: http://www.dailymail.co.uk/news/article-1278427/Health-club-... Admittedly it’s rare, and personally I wouldn’t be worried about it as a liklihood.

Apparently, in that case it was in fact a good idea to avoid the elevator after it started failing:

> Too much had gone wrong for too long with the lift at Broadgate Health Club in the City of London before it dropped on March 12, 2003, killing Polish-born Katarzyna Woja, Southwark Crown Court in London was told.

Re: There's No Fire Alarm for Artificial General Intelligence

#125

Post by stretchwithme » Sat, Oct 14, 2017, 4:39 PM UTC

JabavuAdams wrote: post #118

deong wrote: post #106

Earlier quoted context omitted.

The article makes a lot of good points, but for me, the critical error is in assuming that if short term prediction is hard, long term prediction must be massively harder. He asked a panel for the least impressive thing they did not believe would be possible within a few years. In other words, pick the point closest to the boundary of that classifier. Obviously my future knowledge is imperfect, and anything close to…

> I don't think 30 years will see a machine that can make dinner by gathering ingredients from a supermarket, driving home, and preparing the meal. Really? Why not? Once or twice, if we cherry-pick its performance, or reliably? This is really surprising to me.

It won't be a single machine. It will be multiple systems. And it's not that far away. Probably only 15 years.

And it will be a great boon. Quality of meals will go up and costs will go down. The restaurant market will shrink but not completely disappear.

But McDonald's will certainly die, as there'll be no need to sacrifice quality and nutrition to get speed and convenience. In fact, a table at McDonald's will be an inconvenient booth.

Re: There's No Fire Alarm for Artificial General Intelligence

#126

Post by amelius » Sat, Oct 14, 2017, 4:45 PM UTC

kryptiskt wrote: post #46

IanCal wrote: post #42

Earlier quoted context omitted.

It's reasonable to think that a randomly selected elevator is safe enough to ride, while also thinking the one you've just seen have some kind of fault is not.

But why would an elevator that had demonstrated that it had a working safety brake like that be more unsafe than any other elevator? I trust in Elisha Otis.

By the way, here is a nice video explaining the mechanism:

https://www.youtube.com/watch?v=sSjJjKcoNRk

Re: There's No Fire Alarm for Artificial General Intelligence

#127

Post by saltcured » Sat, Oct 14, 2017, 4:47 PM UTC

johansch wrote: post #120

astdb wrote: post #28

Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. But the catch is the unreliability of (even expert) predictions on technology futures. My take is that it's worth taking pragmatic steps towards studying AI safety measures (i.e. OpenAI), but not going as far as to talk the likes of 'AI research regulation'.

> Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. (Just venting here, not even primarily at you.) 360k babies are born each day. Clearly it is possible to reproduce intelligent machines. The only way it would be impossible to artificially do the same is literally if life was a magic, non-physical thing. I wish people who state things like this…

Yes, we can find a mate and create a baby. We can't know whether it will be ready to fill a particular functional role after 20 years of training. This works OK for an entire society filling its workforce or army, but seems rather inadequate for a technology company to deliver working products on spec and within a reasonable contract delivery period.

If this is the basis of future AGI, I have to wonder if which flavor of dystopia we'll get to enjoy. Will it be a child-selling dystopia where we all raise a dozen kids hoping that some of them will pay off? Or more like silk-farming, where some capitalized breeder sells kits to all the villagers, and buys back the developed products if and only if the villager was lucky enough to raise them to fruition?

Also, if a human baby is our only basis for assuming AGI, then we ought to think about it like genetic-engineering or human augmentation. We'd better anticipate providing schools, hospitals, psychiatrists, courts, and prisons to deal with the wide variety of behaviors and misbehaviors which will come with these new products which have so little determinism as a baby's lifecycle.

Re: There's No Fire Alarm for Artificial General Intelligence

#128

Post by UncleMeat » Sat, Oct 14, 2017, 5:02 PM UTC

Tossrock wrote: post #77

For me, the "smoke under the door" moment was Karpathy and Li's Deep Visual-Semantic Alignments for Generating Image Descriptions [1]. The almost perfectly grammatical machine-generated captions of photos were unnerving to me in a way that simple categorization was not. It somehow called to mind the image of a blank-eyed person speaking in a monotone while images flashed in front of them. What if they wake up? [1]: h…

Yet Andrej himself thinks that RNNs are not really making meaningful progress towards AGI.

Re: There's No Fire Alarm for Artificial General Intelligence

#129

Post by johansch » Sat, Oct 14, 2017, 5:05 PM UTC

saltcured wrote: post #127

johansch wrote: post #120

Earlier quoted context omitted.

> Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. (Just venting here, not even primarily at you.) 360k babies are born each day. Clearly it is possible to reproduce intelligent machines. The only way it would be impossible to artificially do the same is literally if life was a magic, non-physical thing. I wish people who state things like this…

Yes, we can find a mate and create a baby. We can't know whether it will be ready to fill a particular functional role after 20 years of training. This works OK for an entire society filling its workforce or army, but seems rather inadequate for a technology company to deliver working products on spec and within a reasonable contract delivery period. If this is the basis of future AGI, I have to wonder if which flavo…

This is hilarious. (I mean, it was intended as a joke, right?)

Re: There's No Fire Alarm for Artificial General Intelligence

#130

Post by johansch » Sat, Oct 14, 2017, 5:11 PM UTC

gooseus wrote: post #123

johansch wrote: post #120

Earlier quoted context omitted.

> Having done some work with the state-of-the-art of AI, I personally don't think AGI is near - might not even be possible. (Just venting here, not even primarily at you.) 360k babies are born each day. Clearly it is possible to reproduce intelligent machines. The only way it would be impossible to artificially do the same is literally if life was a magic, non-physical thing. I wish people who state things like this…

Just because we can observe something happening, doesn't mean we can understand the mechanisms of how that thing happens and even if we CAN understand the mechanism, it doesn't mean the mechanism is feasibly reproducible within our resource capability. An example might use crypto... you observe random information flying through the air, you may recognize it as an encrypted channel and you may see a machine acting in…

You're missing the point several times over. The entire point is very simple: if there is no magic involved it's down to physics. If it's possible to create brains without magic, it is possible to do so artificially.