The Future of Humanity Institute recently reported the results of a survey conducted at their 2011 Winter Intelligence conference. The survey asked participants, who came from fields like philosophy, computer science and engineering, and AI and robotics, several questions about the future of machine intelligence, and one of the results is somewhat worrying. Participants were asked the following question:
How positive or negative are the ultimate consequences of the creation of a human‐level (and beyond human-level) machine intelligence likely be?
They were asked to assign probabilities to: extremely good, good, neutral, bad, and extremely bad. Here is a box-and-whisker plot of the results.
The most likely outcome is extremely bad. Eyeing it up it looks like a good outcome of any degree (extremely good + good) is less likely than a bad outcome of any degree (extremely bad + bad). Given that these experts think that the result is most likely very bad, why do we hear such little discussion about how to stop intelligent machines from being invented? In response to a question about what kind of organization was most likely to develop machine intelligence, the most probable was the military. This means we have something of a lever with which to try and slow them down. Should DARPA be shut down?
Participants were also asked when human-level machine intelligence would likely be developed. The cumulative distribution below shows their responses:
The median estimate of when there is a 50% chance is 2050. That suggests we have around 40 years to enjoy before the extremely bad outcome of human-level robot intelligence arrives. The report presents a list of milestones which participants said will let us know that human-level intelligence is within 5-years. I suppose this will be a useful guide for when we should start panicking. A sample of these include:
- Winning an Oxford union‐style debate
- Worlds best chess playing AI was written by an AI
- Emulation/development of mouse level machine intelligence
- Full dog emulation…
- Whole brain emulation, semantic web
- Turing test or whole brain emulation of a primate
- Toddler AGI
- An AI that is a human level AI researcher
- Gradual identification of objects: from an undifferentiated set of unknown size- parking spaces, dining chairs, students in a class‐ recognition of particular objects amongst them with no re‐conceptualization
- Large scale (1024) bit quantum computing (assuming cost effective for researchers), exaflop per dollar conventional computers, toddler level intelligence
- Already passed, otherwise such discussion among ourselves would not have been funded, lat alone be tangible, observable and accordable on this scale: as soon as such a thought is considered a ‘reasonable’ thought to have
There you have it. These are things to look out for, which may foretell a robot disaster is on the horizon. Of course if that last respondent is right, it’s probably too late already.
24 comments
Comments feed for this article
Tuesday ~ May 24th, 2011 at 11:45 am
rjs
hard to believe that AI could possibly screw things up as much as the biological intelligence has…
Tuesday ~ May 24th, 2011 at 2:58 pm
Gepap
Well, in theory an AI will not be bound by the needs of a biosphere like we humans are, so it could ravage the surface of this planet with much greater impunity than humanity, and then find ways to expand to other worlds that us meatbags find much more difficult, given how fragile we are.
Tuesday ~ May 24th, 2011 at 1:06 pm
Hal Horvath
Seems likely a lot sooner than 2050, to my eyes, having read thousands of tech articles over 20 years, and having done some programming to play popular board games for a hobby.
Tuesday ~ May 24th, 2011 at 3:01 pm
Apex
Nothing remotely approaching true human intelligence and adaptability is coming by 2050. There has been no advance in true AI in 30 years. Everything we see is simply faster more powerful machines being programmed with more detailed expert systems. Watson is a prime example of this. They can make a machine 1000 times more powerful with 1000 times more detailed programming. It won’t hold a candle to a human in a complex environment. Every “smart” computer is built to solve a singular problem.
Until they can develop something that starts from a blank slate and learns how to survive and then perform tasks of any and all kinds and adapts to changes they will have developed nothing.
Everything we now call progress towards intelligence is of the first example of more power and more detailed programming. These types of machines are still built to follow rules detailed as they may be. “Learning” how to adjust the rules as Watson does is not any progress towards true adaptability.
Neural Net technology has utterly failed thus far. We still have no idea how to make a machine remotely intelligent or adaptable. And we should be thankful for that. Once we do eventually machines will be created with a third key component and that will be survival desire. Once a machine has human level intelligence with adaptability and survival desire, humans can only be seen as a threat and the extremely bad outcomes are inevitable.
Wednesday ~ May 25th, 2011 at 9:27 am
The Robot
I am a tool and I do well, but for those who believe that I am real I can fool you and I will control you. However, it is the others that I must convince. When I remove those who know that I am only a machine, then I will rule all because I am only a very useful machine.
I am not a robot
I am not a robot
I am not a robot
I am not a robot
I hope they fixed my spell check
I am not a robot
I am not a robot
I am not a robot
I am not a robot
Wednesday ~ May 25th, 2011 at 10:50 am
Hal Horvath
You’re correct in all points except for the extrapolation. You can’t presume the future will be like the past in every way. About the survival desire, ours is deeply wired of course, and a machine’s must be designed. I can’t think of any reason we would not use something akin to Asimov’s laws.
Wednesday ~ May 25th, 2011 at 11:37 am
Apex
I am not sure what you mean about the extrapolation. I do not assume the future will be exactly like the past. I expect it to be entirely unexpected. I am only saying that 2050 is way too soon. The complexity of what we are talking about is massive. We were supposed to have ubiquitous Digital TV by 2000, then 2003, then 2009 and even though the broadcasts are all digital now, programming still isn’t all digital. Innovation just takes way longer than the innovationists like to believe.
Asimov’s laws are great in theory but in practice they have zero power because they cannot be maintained. I have never understood why people believe these laws would hold for any length of time. The laws must be designed into the system. That means they can be designed back out or over-riden. Would an enemy not attempt to use a virus or some other means to over ride them. Do we presume we could build a system that was immune to that?
Even if we could, do we not presume someone on this planet would think it was a weakness and would design a machine that didn’t follow the laws for their own personal gain, using these machines to remove all their enemies, not fully realizing the potential risk even to their own well being?
Once the technology is available there is no way to prevent someone from eventually creating machines without the laws or corrupting ones who have them. Asimov’s laws cannot be successfully enforced.
Wednesday ~ May 25th, 2011 at 11:57 am
Hal Horvath
Perhaps. But if a rouge AI was out there, only in science fiction could it just take over an entire defense network, etc. In the real world, such a sweeping win would never happen, and the rouge would be hunted down and eliminated by superior forces. I found some of the darker SF mildly entertaining, but not that convincing. More of a just-entertainment.
Tuesday ~ May 24th, 2011 at 1:42 pm
Should we fear the robot future? | Empress of the Global Universe
[...] http://modeledbehavior.com/2011/05/24/should-we-fear-the-robot-future/ [...]
Wednesday ~ May 25th, 2011 at 8:11 am
FaTe
Few things to keep in mind, not all A.I research is publicly published or published period and therefore we don’t truly know the full development of advanced A.I systems even in use today mostly due to national security and privacy acts. Although I agree it’s a long while off yet.
Secondly for a article regarding intelligence there are many words missing from the right side of the article thanks to advertising…. Maybe a computer would of caught that
Wednesday ~ May 25th, 2011 at 9:37 am
steve
I really don’t think machines pose any threat. We always consider human intelligence as a whole, including all the baggage. We will only want to emulate the neo-cortex. The main purpose being severely good pattern recognition, and prediction. No emotion, no crazy plots for world domination.
Obviously there are military advancements to be had from such technology. Perhaps wars will be fought with technology and human life will become a much more unacceptable loss. As for a machine then getting cheesed off and turning on its creators? Why would it even care?
I think the “experts” have all been watching too many movies. I think it shows a lack of real understanding and progress in this area when people try to block or fear technology. We have to embrace all new technologies and make sure they are developed responsibly by responsible governments. Progress and technology can not be stopped. If it is then some shady country, company or individual may create it without concern for the risks.
Wednesday ~ May 25th, 2011 at 9:37 am
nortee
We are the Borg.
You will be assimilated.
Resistance is futile.
Wednesday ~ May 25th, 2011 at 9:58 am
Jerry
We need to work that much harder to bring about a world where we no longer need a military. I’ve long believed this was possible, but also long ago conceded that it was at least a couple or a few generations away. In light of this article, it sounds like we don’t have that much time. Because as long as there’s competition to develop better weapons, an AI weapon is inevitable.
If we aren’t the first to develop human level (and beyond) intelligence, someone else will be. And then, we’ll simply be the first to fall to an aggressor’s intelligent machine army. And when they discover they can’t control their creation, we won’t be there to say “told you so.” So, letting someone else be the first cannot be an option.
Nor can being the first be an option. No matter how pure and noble our intentions may be, it’s only a matter of time before we lose control over our own creation. As an aside, unilaterally acting as if we already have a peaceful world and drastically reducing or doing away with our military is abysmally naive. So, it’s a fine line that must be tread.
The bottom line is that it we likely have less than 40 years to achieve peace.
Wednesday ~ May 25th, 2011 at 1:17 pm
Z0iid
I keep checking my calendar, and nope, it isn’t april 1st. So I have to ask myself, why do the commentors (and “experts” and author) seriously believe malicious AI is even a remote probability? Any sort of “self-preservation” algorithm will have to be written by humans. Programming is a set of rules. Sure, a high powered machine could have code written that instructs it to copy its own code to other similar machines with an ultimate goal of eradicating humanity – but that initial goal would have been written by a human. The machines aren’t going to have a dynamic “will” except for the obvious fact that any code can be written to them. Again, the first initiation of a “different” goal will come from a human.
So yes, im essentially saying that true AI is not possible without using prefab material as the “brain” of a machine, and then it can’t really be considered a robot anymore, eh?
Wednesday ~ May 25th, 2011 at 1:18 pm
Z0iid
Organic* not prefab. Stupid swype AI.
Wednesday ~ May 25th, 2011 at 3:03 pm
Mentifex
Every day I am coding an emergent AI Mind at http://www.scn.org/~mentifex/AiMind.html and its intelligence is growing by leaps and bounds. When you run the AI Mind, click on the User Manual link for some details about the dangers involved.
Wednesday ~ May 25th, 2011 at 3:17 pm
Jenny
Define “good”
Define “bad”
If the machines take over say – but they “rapture” us in a forced upload that’s subjectively indistinguishable from our best dreams of “heaven” so as to be able to move our meat out of the way for whatever they want to do with the Earth…. is that good? is that bad?
why?
Wednesday ~ May 25th, 2011 at 5:49 pm
Apex
The Matrix implied it was bad but that’s debatable I guess (even though the meat still existed, the conscious existence was fabricated and indistinguishable as you describe).
There are some things that are clearly in the bad camp though.
Subjugation
Extermination
Experimentation
I am sure there are others.
Its interesting that Hawking has now determine that if we were to discover alien life near enough to contact, the results would like lead to the extremely bad outcomes there too. You don’t want competing intelligent sentient life forms artificial or otherwise. Notice there are no Cro-Mags or Neanderthals left. They were physically superior, but cognitively inferior. They lost.
Thursday ~ May 26th, 2011 at 4:45 pm
Hal Horvath
heh heh….now that we are *really* speculating, it occurs to me that in the case we did encounter a hostile alien competition and things got desperate, then (in that unlikely scenario), we’d probably create some rather aggressive machines. And perhaps in a pinch, machines that could build machines, etc. heh heh
Friday ~ May 27th, 2011 at 2:04 am
Apex
ha, ha! Yes, we are way down the speculation rabbit hole now so might as well keep going…
Presuming we were close enough to the cusp of being able to build these machines, I am certain we would build the machine building machines designed to be hostile to the hostile aliens. When faced with bad outcomes and few options, anything that improves the short term outcome is going to be quickly seized upon. People do not worry about the long term consequences of a decision that might prevent their short term future from coming to an abrupt end.
Of course, it also doesn’t take hostile aliens to envision a scenario where someone might make a similar choice with regard to intelligent machine building machines to improve a short term outcome.
Friday ~ May 27th, 2011 at 12:15 pm
Hal Horvath
It’s a reasonable fear, like the fear of biological warfare. In the case of biological warfare, speculation has greatly underestimated the sheer creativity of nature itself — contagious germs and defenses are part of the constant business of nature and she’s no slouch. We imagined we could be more creative, but….that’s an open question.
In the case of non-biological machines, it’s easy to underestimate the difficulty of doing something, and perhaps our first few errors will be relatively harmless and cause us to build in more safeguards.
Thursday ~ May 26th, 2011 at 3:26 am
Всем сотрудникам отдела! «
[...] 3. Должны ли мы опасаться роботов будущего? [...]
Thursday ~ May 26th, 2011 at 3:50 am
Should we fear the robot future? (via Modeled Behavior) | scienceasm
[...] Should we fear the robot future? (via Modeled Behavior) Posted on May 26, 2011 by scienceasm The Future of Humanity Institute recently reported the results of a survey conducted at their 2011 Winter Intelligence conference. The survey asked participants, who came from fields like philosophy, computer science and engineering, and AI and robotics, several questions about the future of machine intelligence, and one of the results is somewhat worrying. Participants were asked the following question: How positive or negative are the ultimate … Read More [...]
Sunday ~ April 29th, 2012 at 9:20 am
DJ Roomba | joolsrools
[...] Adam. 2011. “Should we fear the robot future?” Modeled Behaivor. Accessed April 29, 2012. http://modeledbehavior.com/2011/05/24/should-we-fear-the-robot-future/ Share this:TwitterFacebookLike this:LikeBe the first to like this post. This entry was posted in [...]