Andrew Ng: Why 'Deep Learning' Is a Mandate for Humans, Not Just Machines
wired.com
wired.com
>The reason I say that I don’t worry about AI turning evil is the same reason I don’t worry about overpopulation on Mars.
I think a technological singularity is a rational proposition and even consider it possible within this century. But I have no fear of a Terminator/Matrix/etc outcome. He seems to think the possibility is just too far away to even consider it. But I agree for a different reason. I don't worry because I don't think I (or any human) can accurately assess what a superintelligence would do. And to assume absolute destruction of all humans seems ridiculous.
> In universe two you have one organization, maybe Carnegie Mellon or Google, that invents a self-driving car and bam! You have self-driving cars. It wasn’t available Tuesday but it’s on sale on Wednesday.
>I’m in universe one.
I am strongly in universe one. I have no faith in the basement dwelling loner making significant contributions to AI.
We're much closer than we were a year ago but not as close as many think we are w.r.t self-driving cars, killer robots or even reliable speech recognition.
Safe-guards are probably a good idea before unleashing an AGI onto the Internet but those working on it are not going to be surprised one morning that their AI project suddenly became sentient or dangerous. It will be a deliberate, massive project with lots of funding and lots of smart people working on it.
The novel of '2001' goes into deeper detail than the film regarding the context in which HAL was programmed. To summarize, his mission was to deliberately withhold certain information about the mission from the crew due to the likelihood of xenophobic sentiment developing amongst them if they knew they were traveling to the site of extra terrestrial intelligence. In a sense, he was fulfilling the parameters of his mission in a way which was contrary to the purpose of it. I always interpreted this story as showing us that the more complex a system, the less accurate our predictions of its failure modes are going to be.
I do not think absolute destruction of humanity is likely either, but I also don't think our first steps with AI on this level will be without the loss of life either (of both human and machine). Keep in mind, the same was and is true of space exploration and is essentially an inherent danger of working with the unknown and on the frontier of our knowledge.
Any person on the street could kill me for no reason whatsoever. But I don't live my life like that is a possibility.
I also don't think about future superintelligences in that way either. I don't worry that they will be holocaust machines teetering on the brink of genocide.
E.g. if the AI decides to use the mass of the Earth to build a dyson sphere around the sun. Or tries to build the biggest computer possible and turns the Earth into a giant computer. Or it values reproduction and tries to replicate as much as possible and outcompetes us. Or it values self preservation and kills us because we could make other AIs which could threaten it.
Hmm, I'm not sure I understand this line of thinking. Intelligence is complex, soft, and can even conflict with itself. Why do you envision artificial intelligence as if it were a pure optimization algorithm with high-level problem solving abilities? It seems to me that a fuzzy metaheuristic with lots of competing goals would be a better comparison and this does not lend itself to the "paperclip factory" quite as readily.
The irony about these conversations is that humans are already far more of a threat to humanity now than an AI might be, possibly, at some undetermined time in the future.
AIs (that don't exist yet) are going to be in charge of controlling electronic systems (examples given). They will behave with a time-constant that is unrelated to human reaction times. They will be able to do harm, and compound that harm, at rates mere humans will be powerless to react/
1) The main approach to creating weak AI that actually works is to define some objective function and train some sufficiently complex mathematical model to maximize that objective. Hence it seems likely that if strong AI is created it will follow this paradigm and therefore such a strong AI will care about nothing other than how to maximize its objective. Such a strong AI would be nothing like a human mind so anthropomorphised thinking is of little use in reasoning about it.
2) Given a sufficiently powerful optimization algorithm (say, for example, if P == NP and an efficient algorithm for an NP hard problem were discovered) one can in fact imagine the above paradigm producing an incredibly powerful intelligence, one that could discover many sub-goals for its fundamental optimization objective which the creator of the fundamental objective could not foresee. Further it could be exceedingly effective at achieving its sub-goals, much like modern chess programs are exceedingly effective at winning at chess, but in a far more general domain.
There are however reasons to doubt such a danger.
1) If P != NP then although it should in theory be possible to achieve roughly human equivalent AI, the sort of truly superhuman AI that some fear may not in fact be possible. In other words it may be that if P != NP there is some limit to how powerful an intelligence can become due to the combinatorial explosion that occurs when one tries to expand the range of possibilities that are considered.
2) Even if no such limits exist there are probably ways to maintain control over even a superintelligent AI. A powerful optimizer would be exceedingly good at finding solutions within its own model of the world but that is only part of what is needed to construct a super AI that actually interacts with the world. You would also need a process for creating and updating the world model based on empirical data. This type of architecture would probably give you several powerful levers for controlling the AI and you could always let the optimizer run on the world model to check on its behavior before allowing it to act on the real world.
Regardless I don't know why you think a messy AI is safe. Having a really complicated utility function doesn't mean it is going to result in anything desirable for humans.
Once you're done enumerating as many possibilities as you can for what a superintelligence might have as its goals, simple and complicated or even nonexistent, and what the various outcomes of a superintelligent agent with those goals broadly look like, find out how many of them are positive, negative, or neutral to the future of humanity, especially if these things the superintelligence does requires resources humans also use. I did this once, it convinced me that the likely case is something negative for humanity, even if it's unlikely to be a Terminator/Matrix/Hollywood scenario or really any specific enumeration. What further convinced me that the likely outcome is still negative for humanity, in the event of people making an honest but uninformed attempt at after having created an AGI making sure it won't be net-negative for humanity, is understanding the complexity of human value and how almost right is still very wrong. (http://wiki.lesswrong.com/wiki/Complexity_of_value)
I actually don't quite agree with the spaces on this chart because the scientific method should set us far apart from mice and chimps so that they would almost merge in a single point.
Anyway, there are a handful of reasons to assume that an artificial general intelligence would likely turn out to be far superior to us: (1) Hardware is a much less noisy computation environment compared to the badly insulated neurons in our wetware (though this might be a feature rather than a bug), (2) our neurons operate at about 200 Hz, microchips at 10 million as much, (3) they would have extremely fast access to physical simulations and other computational resources, knowledge bases, other AIs, (4) they would never get tired, no body to maintain, no social obligations, and (5) they could replicate within seconds and easily modify their own sourcecode.
And how did you assign weights to these options?
I assume you didn't make the sophomore mistake of assuming a uniform distribution over all possibilities.
In another sense being a genius can make you feel alone even when you are collaborating.
Why not? There's a strong history of individuals outside of academia making large contributions to science.
The few who make novel contributions are not representative.
People working in groups building on the success and knowledge of others make the vast majority of meaningful contributions in all fields.
But most basement dwellers don't go to that much effort.
No, not really. "We" academics get cliquey because of the intellectual prerequisites to our work, not because we're trying to form a closed conspiracy for domination of the academic world. If someone can demonstrate findings but has unusual institutional affiliations, we'll give them exactly our normal treatment (ie: skepticism in proportion to how thoroughly they're proving our hypotheses wrong, mixed with enthusiasm for having something interesting to talk about).
I have no faith that Strong A.I/AGI will come from the above efforts as they are anything but General. The whole is greater than the sum of the parts and the current crop of weak A.I algorithms are a small part of a bigger whole.
As far as fear of AGI/Strong A.I. Listen, you can't stop something whose time has come. When it comes and it will, no million dollar consortium of businesses who have leveraged weak A.I to line their pockets will be able to stop it. We can postulate to the ends of the earth about what it will be and how we must prepare for it or we can focus on creating it.
Those who are most likely to create it aren't worrying to the ends of the earth about its dangers. They aren't funding million dollar initiatives to make 'tests' for something that isn't even understood yet. They are out there thinking deep and far and working through what it takes to create it... Likely : Thinking Different than the current crop of people centered on capitalizing (increasing profit margins) on a very productive discovery : Weak A.I. The fear that arises from this group is in knowing that AGI will trump Weak A.I (their cash cow). Thus, the fear propaganda.
So, where do I feel AGI will likely crop up from? The basement dwelling loner asking the deep and general questions about life, matter, energy, information without a particular profit maximizing goal in mind. Their goal being to more deeply understand the very nature/fabric of life, intelligence, and this universe...
Otherwise, hey.. I could sit back and believe that wisdom just emerges when you've dumped enough money in the laps of A.I experts.
Think different. You're not going to achieve anything new and ground breaking thinking and looking at the problem the same way everyone else is. Sure, there will be many failures but therein lies the risk/reward of going beyond the herd.
I had the same feeling while studying computational neuroscience.
There is no way in hell the current approaches is going to lead to AGI anytime soon ! The models we have for even the most simplest operations of the brain is not well though out. Even using the term "Artifical Nueral Nets" is an insult to the real machinary that is a neuron.
However, I do not think someone in some basement is going to figure out the solution. I think the people who are working to solve the problem are the army of nameless minimum wage graduate and PhD students across the industrialized world who are studying all the various topics involving hardware, software, biology, etc. All of which are required to create a a true AGI system.
We are no where near creating Strong A.I. but its going to come incrementally.
There are many contributors to science who go unmentioned. There are many back-stories which, due to not conforming with dominant ideologies, go untold.
http://nautil.us/issue/21/information/the-man-who-tried-to-r...
Best of luck to everyone in pursuit of AGI. The core of creation is elusive for a reason. Many will miss it not due to their education but their mindset.
>As far as fear of AGI/Strong A.I. Listen, you can't stop something whose time has come.
Huh? All of the people advocating AI safety are saying to invest in making AI safe, not to ban AI research.
However that's not even correct. The development of general purpose computers was delayed for nearly a century because it didn't get enough funding. Human cloning and many other areas of technology/research have successfully been banned. I'm not saying that it's a good idea, but you can't argue that it's impossible to prevent technologies.
>The fear that arises from this group is in knowing that AGI will trump Weak A.I (their cash cow). Thus, the fear propaganda.
Are you seriously suggesting some kind of conspiracy by "weak AI" people to prevent the development of AGI by manipulating the media?
"Huh? All of the people advocating AI safety are saying to invest in making AI safe, not to ban AI research."
> Huh? the issue is : no one even knows what true 'Strong/General' A.I is. Are human beings safe? Look at all of the idiotic things we do. Look at what some of the most monied of us do. Furthermore, who restricted them on building their empires of other people's data and weak A.I analytics therein? Nobody. They made up the rules as they went, many times after-the-fact... That's 'business' right? and the same is going to be so for the progression to AGI..
We still don't understand how memory is stored in the human brain. So, how can you restrict research into it? No one even understands it. Thus, you can't. Everyone is free to do as they please in way of research and there's nothing anyone can do about it.
* As for human cloning, can you say as a definite that no one in a dark lab isn't experimenting on it? No. You can't. As I recall, stem cell research was banned and it ended up in Singapore. Now, it's everywhere. Life, as it were, goes on. No silly amount of money or consortium can stop it.
Are you seriously suggesting some kind of conspiracy by "weak AI" people to prevent the development of AGI by manipulating the media?
> I'm seriously highlighting the ignorance of going on about something you don't understand and trying to restrict it. Who restricted social media? Who restricted mass data collection? Analytic? Who studied the negative societal effects they have? They just did it and went w/ it.. But, all of a sudden, it's time to put restrictions on A.I? Poppycock. When AGI is broken into, the [run] button will be hit and that will be that. What, is Elon Musk going to pull the plug? Provide me some definite way you can restrict the coming of AGI and I'll resign my view. I frankly don't care what drives people to 'fear' it. Give me detail on how the least of those who understand it can stop it.
>Are human beings safe?
They have empathy and don't have unlimited power, so yes for the most part. But that's not really relevant.
>So, how can you restrict research into it? No one even understands it. Thus, you can't. Everyone is free to do as they please in way of research and there's nothing anyone can do about it.
Again I'm not advocating for it, but if the government wanted to ban AI research, they could. Sure rebels could study in secret, just like they can with human cloning or bioweapons, nukes, etc. But for the most part lone rebels working on their own in secret never accomplish anything and real work requires funding and peer review.
>But, all of a sudden, it's time to put restrictions on A.I? Poppycock. When AGI is broken into, the [run] button will be hit and that will be that. What, is Elon Musk going to pull the plug? Provide me some definite way you can restrict the coming of AGI and I'll resign my view.
Again, we aren't trying to stop AI, that was something you brought up. We are trying to create "Friendly AI", i.e. AI that shares human values and won't simply turn the universe into paperclips. And it's very important that we solve this problem before we invent working AI.
[edit]: grammar mistake
1. unless we jump into a hidden block somewhere, we're definitely on our way to creating machines in the next 30 years that can out think humans in every way
2. given point 1, we tend to place an anthropomorphic role on the "destructive" nature of future "robots", forgetting that:
A) machine sentience is not required for machine-induced destruction
B) a few humans with very powerful machines can be quite destructive
It's been a long time since humans posed a challenge to computers in chess, and yet freestyle chess competitions aren't dominated by supercomputers, but by (relatively novice) chess players assisted by computers. Why should we settle for a false dichotomy of human versus machine when we can do so much better with cooperation and augmentation?
Is he trying to push the boundaries of neural networks or trying to create a new pattern recognition algorithm or machine learning algorithm like a Naive Bayes classifier?
I just wish he had went into more technical detail about his work.