AI beats the game Go

[cosmicalstorm]

Interesting news on the development of AI, he takes a punt at those who think it will take a long time;

Facebook:

Eliezer Yudkowsky on the Go AI that beat the European champion player:

> People occasionally ask me about signs that the remaining timeline [to general AI] might be short. It's *very* easy for nonprofessionals to take too much alarm too easily. Deep Blue beating Kasparov at chess was *not* such a sign. Robotic cars are *not* such a sign.

> This is. [...]

> As the authors observe, this represents a break of at least one decade faster than trend in computer Go.

> This matches something I've previously named in private conversation as a warning sign - sharply above-trend performance at Go from a neural algorithm. What this indicates is not that deep learning in particular is going to be the Game Over algorithm. Rather, the background variables are looking more like "Human neural intelligence is not that complicated and current algorithms are touching on keystone, foundational aspects of it." What's alarming is not this particular breakthrough, but what it implies about the general background settings of the computational universe.

> To try spelling out the details more explicitly, Go is a game that is very computationally difficult for traditional chess-style techniques. Human masters learn to play Go very intuitively, because the human cortical algorithm turns out to generalize well. If deep learning can do something similar, *plus* (a previous real sign) have a single network architecture learn to play loads of different old computer games, that may indicate we're starting to get into the range of "neural algorithms that generalize well, the way that the human cortical algorithm generalizes well".

> This result also supports that "Everything always stays on a smooth exponential trend, you don't get discontinuous competence boosts from new algorithmic insights" is false even for the non-recursive case, but that was already obvious from my perspective. Evidence that's more easily interpreted by a wider set of eyes is always helpful, I guess.

> Next sign up might be, e.g., a similar discontinuous jump in machine programming ability - not to human level, but to doing things previously considered impossibly difficult for AI algorithms.
I hope that everyone in 2005 who tried to eyeball the AI alignment problem, and concluded with their own eyeballs that we had until 2050 to start really worrying about it, enjoyed their use of whatever resources they decided not to devote to the problem at that time.

https://www.facebook.com/yudkowsky?pnref=story

[Adam Reynolds]

Someone like Yudkowsky is very vulnerable to the confirmation bias. The fact that he believes that an AI is a potential threat means that any evidence in its favor is something he will tweak to fit his beliefs while any evidence against is something he will find a reason to ignore. Having a high IQ doesn't actually make people less vulnerable to a bias like this.

Though I would not go as far to say he wrong, I certainly lack the knowlege to really say. I wonder what Starglider's opinion on this would be.

[cosmicalstorm]

Hes far from the only one impressed by this though I quoted him because I like him. This fits a bit too well with predictions made by some smart people, a bit too well for me to write it off as confirmation bias, though I hope this is wrong and AI will actually take a century.

[Lord Revan]

well tbh "someone I like" and "some smart people" aren't exactly what I'd call high standards for information and lets be honest games still have fairly set and logical rules (maybe not simple not still reasonbly logical) so beating a game doesn't mean that a self-aware AI is really that close to becoming a reality.

I'd like to know more as well before I make any real judgements either way.

[Jaepheth]

Self aware-AI? No

AI capable of displacing massive numbers of people from their jobs? Quite possibly.

Go is an abstract enough game that beating top level human players without relying heavily on monte-carlo style computation specialized to that particular game shows, to me, that AI is poised to take over many human-held positions in management, resource allocation, planning, etc.

[cosmicalstorm]

well tbh "someone I like" and "some smart people" aren't exactly what I'd call high standards for information and lets be honest games still have fairly set and logical rules (maybe not simple not still reasonbly logical) so beating a game doesn't mean that a self-aware AI is really that close to becoming a reality.

I'd like to know more as well before I make any real judgements either way.

Well a more demeaning term would be "various transhumanists". Take it for what it's worth. At least they resonate with me, where most run of the mill experts in other subjects strike me as suffering severe myopia as soon as they approach AI.

[Adam Reynolds]

Self aware-AI? No

AI capable of displacing massive numbers of people from their jobs? Quite possibly.

Go is an abstract enough game that beating top level human players without relying heavily on monte-carlo style computation specialized to that particular game shows, to me, that AI is poised to take over many human-held positions in management, resource allocation, planning, etc.

Indeed. It will likely become the case that a CEO is easier to replace than a plumber. That would be gloriously ironic. It would be interesting to see someone like that justify their obscene salary when an AI is doing all of the actual decision making. To some degree, we already see this on Wall Street in which traders take orders from algorithms.

If we can't survive the societal shifts this will create, it really won't matter whether we can deal with an ASI.

[Starglider]

Someone like Yudkowsky is very vulnerable to the confirmation bias. The fact that he believes that an AI is a potential threat means that any evidence in its favor is something he will tweak to fit his beliefs while any evidence against is something he will find a reason to ignore. Having a high IQ doesn't actually make people less vulnerable to a bias like this.

That's not a meaningful criticism without more detail. How has this evidence been 'tweaked'? What counterexamples have been ignored?

My personal opinion, based on the small selection of papers I've read (connectionist approaches haven't been my focus recently), is that deep neural networks (various recent approaches to) are a significant advance but plateau quite quickly in the amount of layering and the scope of the abstraction they can do. This plateau is sufficiently ahead of the previous plateau that Go went from 'don't have a good enough heuristic to narrow the search enough that we can leverage brute force' to 'can restrict the search enough that brute force can be brought to bear as in Chess'. Note that it is still doing billions of positions worth of forward search, unlike a human, but unlike Chess algorithms that use pretty simple and often handwritten heuristics to guide the search, it is using fuzzy pattern processing trained from millions of sample expert moves. It's an impressive application of machine learning, but the problem that the ML solved is 'a reasonably good Go search pruning heuristic'. Most other domains aren't amenable to a hybrid NN search control / brute force search in this way, because it relies on there being a simple obvious world model that can be coded up by humans to efficiently simulate (a huge set of) future actions for. This specific application is actually a byproduct of DeepMind's big data focused research; the NN engine is the main product, connecting it to and using it to improve the efficiency of a conventional Go move-search engine was a bit of side research which grew into a PR event.

I wonder what Starglider's opinion on this would be.

I broadly agree with it, which is not surprising, as I have consistently agreed with EY on general AI issues. That just means we both have a reasonable understanding of the field. The things I disagreed with EY about were detail points of technical design and strategy. Even there, it's been quite a few years since I last did a detailed comparison and both our positions have evolved since then, so I don't even know what details we do and do agree about right now.

That said, obviously I agree with this;

Next sign up might be, e.g., a similar discontinuous jump in machine programming ability - not to human level, but to doing things previously considered impossibly difficult for AI algorithms.

...as I am working on it for a commercial application right now and I plan to resume bluesky experiments in the area this year, possibly with the help of some other UK based AI researchers. There is an obvious synergy with advances in connectionism, in that with automated programming to build and link the models and control layer, and select and tune the fuzzy pattern processing algorithms, the AlphaGo approach does become applicable to many more domains. In fact that is kind of what I am doing except with stastical ML instead of NN.

[Starglider]

It will likely become the case that a CEO is easier to replace than a plumber.

The CEO would be hard to replace because their main roles (at a large company) is investor relations, internal power politics, public relations, employee motivation and holistically imagining what general shape and direction the company should have.

The CFO on the other hand, should totally be replaced with a computer system at the earliest opportunity.

[Starglider]

Go is an abstract enough game that beating top level human players without relying heavily on monte-carlo style computation specialized to that particular game shows, to me, that AI is poised to take over many human-held positions in management, resource allocation, planning, etc.

The problem is that you'd have to create a detailed model of the domain you want it to optimise, probably with lots of custom interconnections to existing systems. Natural language comprehension is a bottleneck for ingesting a lot of the relevant data and we're still not there for structured documents (beyond simple queries and commands). It's worth doing by hand for things like algorithmic trading, where there is a lot of money on the line and correspondingly high budgets, but not so much for typical company tasks. Again, automated programming could change this; I've been saying this for a decade or so but VCs are hard to convince, seemingly if it's not big data analytics they're not interested.

[Adam Reynolds]

I suppose I stand highly corrected. I was saying this in response to the last line of the posted article, about it significantly advancing the timeline. What I meant was that it seems there are too many variables to seriously say anything about long term prospects, even with this sucess.

As for the other point, A CFO would indeed be a more beneficial example, but a CEO over a plumber just sounded more dramatic. I actually had thought about the PR point as well. That likely would still require a figurehead regardless of whether he was seriously involved in making the majority of company decisions to the same degree.

Wouldn't automated programming also put you out of a job over time? Well, maybe not you personally, but it would be ironic that those that are automating others' fields themselves face the problem of their jobs being automated.

[Grumman]

Self aware-AI? No

AI capable of displacing massive numbers of people from their jobs? Quite possibly.

Go is an abstract enough game that beating top level human players without relying heavily on monte-carlo style computation specialized to that particular game shows, to me, that AI is poised to take over many human-held positions in management, resource allocation, planning, etc.

Indeed. It will likely become the case that a CEO is easier to replace than a plumber. That would be gloriously ironic. It would be interesting to see someone like that justify their obscene salary when an AI is doing all of the actual decision making. To some degree, we already see this on Wall Street in which traders take orders from algorithms.

As I said in the Happiness Code thread, I am of the opinion that no AI should ever have decision-making authority. It should serve an advisory role only. If the AI at a toy factory decides it's cheaper to use lead-based paint, there should be a human in the loop who will be held personally accountable if he gives the order to do so. To allow the company management to avoid accountability by writing it off as a glitch that will be fixed in CEOBot v1.1 is insane.

[Starglider]

Wouldn't automated programming also put you out of a job over time? Well, maybe not you personally, but it would be ironic that those that are automating others' fields themselves face the problem of their jobs being automated.

There are certainly hordes of low-to-mediocre skilled programmers who can and should be eliminated by relatively narrow automated software engineering tools. Again, I've previously pitched exactly this to VCs. They didn't go for it because in recent history compilers etc haven't been a particularly profitable or sexy product area. I probably should've stressed platformisation more.

I am of the opinion that no AI should ever have decision-making authority. It should serve an advisory role only.

Well that is just naive. Relative moral reliability of AI systems vs human executives aside, it's not as if execs currently have much personal liability for decisions like that.

[Purple]

We programmers seriously need a guild of some sort that would work to prevent us being automated away and punish those that violate that tenant.

[Ace Pace]

We programmers seriously need a guild of some sort that would work to prevent us being automated away and punish those that violate that tenant.

Why? Looking at it from a self interested manner, I'd love for there to be less programmers doing grunt work. In a sideways manner it'd improve my situation drastically.
Furthermore, unlike other professions, the ability of any single researcher to come up with a ground breaking idea, implement it and publish it are far greater and any such guild would be powerless to prevent it.

[Purple]

Why? Looking at it from a self interested manner, I'd love for there to be less programmers doing grunt work. In a sideways manner it'd improve my situation drastically.

It would however remove a lot of entry level positions and seriously screw over a lot of people because managers would get it into their heads that they don't need anyone since they can just buy a programmer AI. Do we really need more managers thinking they know how programming works? About the only one I can see profiting off this are those doing high end cutting edge science stuff. Everyone else is screwed.

Furthermore, unlike other professions, the ability of any single researcher to come up with a ground breaking idea, implement it and publish it are far greater and any such guild would be powerless to prevent it.

I know. It's not practical. But a man can dream.

[Ace Pace]

Why? Looking at it from a self interested manner, I'd love for there to be less programmers doing grunt work. In a sideways manner it'd improve my situation drastically.

It would however remove a lot of entry level positions and seriously screw over a lot of people because managers would get it into their heads that they don't need anyone since they can just buy a programmer AI. Do we really need more managers thinking they know how programming works? About the only one I can see profiting off this are those doing high end cutting edge science stuff. Everyone else is screwed.

Already many entry level IT positions are being automated. Manual testers are a seriously rare position these days. So are more and more entry level sys admin jobs. Entry level e-commerce developers are being killed off by WiX and other competitors.

Unlike many other professions, people can jump into mid-range positions. Harder? More than you can imagine but it's possible. The entire field isn't exactly static and never was.

[The Grim Squeaker]

Already many entry level IT positions are being automated. Manual testers are a seriously rare position these days. So are more and more entry level sys admin jobs. Entry level e-commerce developers are being killed off by WiX and other competitors.

Unlike many other professions, people can jump into mid-range positions. Harder? More than you can imagine but it's possible. The entire field isn't exactly static and never was.

A recent project I worked on involved, and I quote, the team head mentioning that our AI platform doing a sample task they had would mean effectively eliminating a quarter of their Q&A technical staff. After just the prototype run .
Our platform also automates a lot of the stuff I would normally need to do as a data scientist, and you know what? It's much more fun to work that way. I'm more productive, and lowering the difficulty of tasks isn't a bad thing when demand outstrips supply to a massive degree .

[Purple]

A recent project I worked on involved, and I quote, the team head mentioning that our AI platform doing a sample task they had would mean effectively eliminating a quarter of their Q&A technical staff. After just the prototype run .
Our platform also automates a lot of the stuff I would normally need to do as a data scientist, and you know what? It's much more fun to work that way. I'm more productive, and lowering the difficulty of tasks isn't a bad thing when demand outstrips supply to a massive degree .

I have no issue with you automating away everyone who is not in our industry. Like literally if you managed to destroy every single job but ours I would not mind. But I take issue when you start threatening my potential carrier prospects.

Already many entry level IT positions are being automated. Manual testers are a seriously rare position these days. So are more and more entry level sys admin jobs. Entry level e-commerce developers are being killed off by WiX and other competitors.

Unlike many other professions, people can jump into mid-range positions. Harder? More than you can imagine but it's possible. The entire field isn't exactly static and never was.

And it's something I personally would not like to do even if it was very easy. Fact of the matter is that entry level jobs exist for a reason. They allow beginners to get experience in the field and company without having the ability to screw something important up. I know just how inexperienced I am and would not like my first ever job to be one of great responsibility. But there are plenty of young people, especially the millenials who don't think that way. Do you really want some uppity kid with no experience jumping into a mid level position immediately? How would you react if it was in any other field? Like if say you had a kid fresh out of school and immediately becoming a brain surgeon?

[Ace Pace]

I have no issue with you automating away everyone who is not in our industry. Like literally if you managed to destroy every single job but ours I would not mind. But I take issue when you start threatening my potential carrier prospects.

I think I'll just leave this here.

And it's something I personally would not like to do even if it was very easy. Fact of the matter is that entry level jobs exist for a reason. They allow beginners to get experience in the field and company without having the ability to screw something important up. I know just how inexperienced I am and would not like my first ever job to be one of great responsibility. But there are plenty of young people, especially the millenials who don't think that way. Do you really want some uppity kid with no experience jumping into a mid level position immediately? How would you react if it was in any other field? Like if say you had a kid fresh out of school and immediately becoming a brain surgeon?

Entry level jobs exist in many fields to start learning how the field works, to start padding the resume. Fact is, these entry level jobs are usually an important stepping stone because some things you can't learn outside a real job.
However, in IT, this isn't true. I can and did have a homelab that simulated my workplace IT and I learned at home. You can learn to do really complex SW development at home, whether by working on OSS or developing your own stuff.
There are places that recruit into mid level positions with mentoring, based on potential. Are there enough? No. Are there places that will give you a good entry path to such places? I'll give one example - Miltaries (before the nitpickers come in, no, not everyone can work there), multiple Western militaries will let you jump into interesting IT jobs and teach you.

[Adam Reynolds]

I guess I really wasn't wrong about the irony point.

I am of the opinion that no AI should ever have decision-making authority. It should serve an advisory role only.

Well that is just naive. Relative moral reliability of AI systems vs human executives aside, it's not as if execs currently have much personal liability for decisions like that.

Indeed. Even if not completely malicious, as in the case of something like the financial crisis, I doubt an AI would have failed to see the potential consequences that humans missed. That is the problem, humans really aren't very good decision makers on the level that we need to deal with high level decision making in the modern era.

Having a mixed human/computer decision making team is in many cases worse, especially in stressful situations. The shootdown of IranAir 655 by the USS Vincennes is a classic example of this type of problem. What happened in a nutshell is that the computer changed the target track without the crew realizing it such that when they saw what was actually an airliner, it had the same number as what they had just seen as a military target. Had the computer alone been making the decisions, it would not have had this problem(contrary to the story told by PW Singer in Wired for War). Though he is correct in his general point that the man in the loop is becoming all but impossible due to reaction times, ironically had he been closer to the true story it would actually have been more of a point in his favor.

[Purple]

...

I need not reply to your post in pieces because I only really need to say one thing. Or rather ask one thing. Why do you find it in any way strange, surprising or potentially negative that I would, like any sensible human being despise anything that has the potential to make my life more difficult?

[Jub]

Already many entry level IT positions are being automated. Manual testers are a seriously rare position these days. So are more and more entry level sys admin jobs. Entry level e-commerce developers are being killed off by WiX and other competitors.

Unlike many other professions, people can jump into mid-range positions. Harder? More than you can imagine but it's possible. The entire field isn't exactly static and never was.

At least half of my most recent tech support job could have been resolved by a program that could read the error codes out of our monitoring software, read out the network path from a network diagram, and then create and send and email advising the on-site staff about what their first step should be and how to contact our representatives if they need further instruction. As an add-on, it could also keep perfect track of properties with high-frequency issues so we could have our engineers work up a more permanent fix. It might have taken a graduate level programmer a month or two to make a completely working model with the emails, at first, going to our phone team or senior support person for proofreading before rolling it out completely.

[Simon_Jester]

...

I need not reply to your post in pieces because I only really need to say one thing. Or rather ask one thing. Why do you find it in any way strange, surprising or potentially negative that I would, like any sensible human being despise anything that has the potential to make my life more difficult?

Purple, you are either trolling, or failing the ethics equivalent of the Turing Test.

If you're trolling, you're not worth talking to.

If you're so shortsighted that you cannot grasp the basic problem with trying to keep special rights for yourself, while explicitly not caring if anyone else gets them... you're not worth talking to.

[The Grim Squeaker]

Already many entry level IT positions are being automated. Manual testers are a seriously rare position these days. So are more and more entry level sys admin jobs. Entry level e-commerce developers are being killed off by WiX and other competitors.

Unlike many other professions, people can jump into mid-range positions. Harder? More than you can imagine but it's possible. The entire field isn't exactly static and never was.

At least half of my most recent tech support job could have been resolved by a program that could read the error codes out of our monitoring software, read out the network path from a network diagram, and then create and send and email advising the on-site staff about what their first step should be and how to contact our representatives if they need further instruction. As an add-on, it could also keep perfect track of properties with high-frequency issues so we could have our engineers work up a more permanent fix. It might have taken a graduate level programmer a month or two to make a completely working model with the emails, at first, going to our phone team or senior support person for proofreading before rolling it out completely.

When I said "A quarter" I might have been sugercoating it.
The exact quote by the head of the department was "get rid of most of the QA division" .
What you're describing is jus classifying a bug as belonging to a predefined category + a script with the answer/class.
So - what's your current job ?