A major leap towards AI

[rickdrat]

A machine has finally passed the Turing test.

http://www.theguardian.com/technology/2 ... CMP=twt_gu

[--NightBattery--]

me too
behold my real form.

[darkbutflashy]

Now I hope your supervisor is not Homer Simpson.

[FaceoffFembot]

What is not said in the paper is that this simulation ran during a Call of Duty online match, and that its parser only contained the strings "mate", "noscope", "hacks" and "your mum". Nobody could have guessed.

[D.Olivaw]

Seriously though, I remember many years ago thinking that when this finally happened, we would be awash in usefully realistic AI, and that we would be on the threshold of very realistic android technology.

It doesn't seem now like I thought it would then.

That's because stuff like this is just a glorified chatbot, a further development of the "Eliza" type programs that have been around since the 70's. Such programs, given something like:

I have a dog.
His name is Spot.
He is in my house.

Still have a hard time doing anything other than sputtering gibberish if asked "Where is my dog?" or "What is my dog's name."

I also note that the version of the Turing Test "passed" by the program in question was much, much easier than the one actually proposed by Turing.

That said, there has been progress in AI, you'll just never see it in the newspapers. It's boring and slow, not exciting and scary.

[rickdrat]

This just in:

https://www.techdirt.com/articles/20140 ... tter.shtml

...and that's exactly the kind of debunking you'd expect... from a suddenly self-aware, 13 year old AI in a subtle effort to distract the world from it's penultimate goal of world domination.

I, for one, welcome our new machine overlords....

[dale coba]

[quote="NYT today - Intelligence Too Big for a Single Machine"].
“We’re seeing a rebirth of artificial intelligence driven by the cloud, huge amounts of data and the learning algorithms of software,” said Larry Smarr, founding director of the California Institute for Telecommunications and Information Technology.

The emerging global network, Mr. Smarr said, will be the equivalent of a “planetary computer.” What might that mean, in terms of its practical effect on everyday life?

Mr. Smarr points to the recent movie “Her” as a fairly accurate glimpse of what will be possible in the not-too-distant future. The protagonist, Theodore Twombly (played by Joaquin Phoenix), has clever software on his smartphone that seems to know all about him. It has read his email, his text messages and the books, magazines and everything else he has read. It has seen all the movies he has seen. It knows his buying habits and preferences. It retrieves information and answers at his whim. It communicates with him by talking, conversationally (in the voice of Scarlett Johansson).

“That’s where we’re headed,” Mr. Smarr said. “That kind of hyper-personalized assistance is going to be common in 10 years. It will appear to be on your smartphone or Google Glass, but it will actually be in the cloud.[/quote]
Oculus Rift VR sounds like an even more desirable interface, for private conversations.

Which other celebrities will be the first to have their natures cloned and sold like Scarlett's "Her"?
- Angelina as Lara Croft?
- Jessica Alba?
- Natalie as Padmé?
- Or a package deal, a ménage à ensemble: Charmed, Buffy's squad, or the ladies of Smallville and Metropolis?
You anime-heads will know a few likely candidates I don't.

If Smarr says "common in 10 years", then competition is fiercely on, right now. Remember, I think we don't need anything nearly as sophisticated as a human brain to produce all the artificial "intelligence" we need for a truly satisfactory experience with a simulation.

- Dale Coba

[D.Olivaw]

Yeah, it's generally recognized that you need to pair the Turing test with a functional analysis (peek under the hood of the AI, as it were) for valid results.

[DukeNukem 2417]

Here's an article about discarding the Turing Test as a method for assessing AI:

http://www.wired.com/2014/06/beyond-the-turing-test/

The VICIverse got rid of the Turing test a looooong time ago, ladies and gentlemen.

[Spaz]

The real test should be:
*Can the A.I. make independent decisions?
*Can it act intelligently?
*Can it adapt to changing circumstances?

[D.Olivaw]

The real test should be:
*Can the A.I. make independent decisions?
*Can it act intelligently?
*Can it adapt to changing circumstances?

You must be very, very careful when making such tests that most humans can pass them. That may sound like a joke, but think through the consequences.

[--NightBattery--]

The real test should be:
*Can the A.I. make independent decisions?
*Can it act intelligently?
*Can it adapt to changing circumstances?

You must be very, very careful when making such tests that most humans can pass them. That may sound like a joke, but think through the consequences.

yeah, you don't want another polemical-philosophical science thing that make it looks like some humans have less human factors than other humans.
like the bell curve

[D.Olivaw]

The real test should be:
*Can the A.I. make independent decisions?
*Can it act intelligently?
*Can it adapt to changing circumstances?

You must be very, very careful when making such tests that most humans can pass them. That may sound like a joke, but think through the consequences.

yeah, you don't want another polemical-philosophical science thing that make it looks like some humans have less human factors than other humans.
like the bell curve

The Bell Curve was hardly polemical in nature; it's not controversial that people have a roughly normal (Gaussian) distribution on various traits like IQ, conscientiousness, height, strength, etc. The book was about modern society/technology allowing people to self-segregate on the basis intelligence, whereas in the past high-intelligence individuals more frequently interacted with lower-intelligence people in their daily lives (and, to a lesser extent due to relative population proportions, vice versa). One of the least read but most (unfairly) maligned books ever.

Machiavelli is very similar. Read at least a few of his books all the way through instead of excerpts selected for shock value and it'll blow your mind
Long story short, he's misunderstood for the same reason von Clausewitz is: he was being descriptive rather than normative, but everyone naturally assumes that a book on politics is necessarily normative. Machiavelli's normative goal was the union of Italy into a strong republic (on the Venetian model) so that it could stop being the battlefield for 2/3 of Europe's wars. Not so evil and crazy, eh? Burnham's "The Machiavellians" goes into good detail on that topic if you're interested.

The problem my comment was attempting to highlight was that the criteria given by Spaz are useful measures of performance for an AI (if needing much further specification), but that it is tempting to set them so high that a double-digit percentage of the human race fall underneath one or more of the thresholds. Also, if you set them so low that that doesn't happen (60 IQ, perhaps, given that IQ is highly correlated with Spaz's criteria), then it might not be as useful a test as we'd like. You'd never say a 60 IQ person was not human, or not worthy of existence, so those sorts of tests are missing something important.

Now, we currently have no idea how to make an AI that even comes close on any of those criteria, so it's not a big worry. But we might in the future, so best to carefully plan ahead.

[dale coba]

The Bell Curve was hardly polemical in nature; it's not controversial that people have a roughly normal (Gaussian) distribution on various traits like IQ, conscientiousness, height, strength, etc. The book was about modern society/technology allowing people to self-segregate on the basis intelligence, whereas in the past high-intelligence individuals more frequently interacted with lower-intelligence people in their daily lives (and, to a lesser extent due to relative population proportions, vice versa). One of the least read but most (unfairly) maligned books ever.

From what I recall at the time, Murray and those on the right-wing of the political spectrum who sympathetically promoted him did a very poor job of representing what you have stated above; but I can believe you, that there was real academic value beneath all that mangled media coverage.

What no one appreciated the time, is that while DNA is inheritance, defects in the maintenance of the DNA can also be inherited. That extra-genetic inheritance is the subject of the science of epigenetics. There can be devastating toxic exposures in the womb or in parental gonads, resulting in DNA lacking some of the crucial markers that determine which sections of code should be transcribed at different times over the lifetime of different cell lines.

Toxic exposures are largely determined by class - food, water, occupation, neighborhood - and may be passed not only from parent to zygote, but directly to you from your mother's mother, when your mother was a fetus, when her eggs were forming and being epigenetically encoded.

Teetering on the edge of eugenics is a dangerous and complicated territory, but it gets a lot less complicated when you can tease out the epigenetic factors from what everyone presumes to be genetic inheritance. Many of the diseases for which we "still haven't found the gene" are not caused by a misspelled sequence of DNA, but rather by bad punctuation laid onto the DNA, especially at that crucial stage when the male or female parent must lay down a set of many markers specific to their gender on the DNA of egg or sperm.

Our various ethnicities' genes are not that different, not so much severely less fit from one group to another. The self-segregation that you cite is surely a factor responsible for some of whatever I.Q. divergence has been measured, and affects both social nurture and DNA/nature. However, much of the differences are substantially due to our ancestors' different, historical toxic exposures.

If they were asked for any possible scientific justifications, ignorant racists would say they're discriminating against people on the basis of inferior genetics. What they are really doing is trying to devalue the cumulative epigenetic damage of another people's historical deprivation.

Lost within the data, Murray was almost certainly looking at some epigenetic consequences; but there was no epigenetic science back when he published. He needed all the scientific truth he could've gotten, to protect himself and his scholarly work from his racist, so-called friends.

- Dale Coba

[D.Olivaw]

From what I recall at the time, Murray and those on the right-wing of the political spectrum who sympathetically promoted him did a very poor job of representing what you have stated above; but I can believe you, that there was real academic value beneath all that mangled media coverage.

What no one appreciated the time, is that while DNA is inheritance, defects in the maintenance of the DNA can also be inherited. That extra-genetic inheritance is the subject of the science of epigenetics. There can be devastating toxic exposures in the womb or in parental gonads, resulting in DNA lacking some of the crucial markers that determine which sections of code should be transcribed at different times over the lifetime of different cell lines.

Toxic exposures are largely determined by class - food, water, occupation, neighborhood - and may be passed not only from parent to zygote, but directly to you from your mother's mother, when your mother was a fetus, when her eggs were forming and being epigenetically encoded.

Teetering on the edge of eugenics is a dangerous and complicated territory, but it gets a lot less complicated when you can tease out the epigenetic factors from what everyone presumes to be genetic inheritance. Many of the diseases for which we "still haven't found the gene" are not caused by a misspelled sequence of DNA, but rather by bad punctuation laid onto the DNA, especially at that crucial stage when the male or female parent must lay down a set of many markers specific to their gender on the DNA of egg or sperm.

Our various ethnicities' genes are not that different, not so much severely less fit from one group to another. The self-segregation that you cite is surely a factor responsible for some of whatever I.Q. divergence has been measured, and affects both social nurture and DNA/nature. However, much of the differences are substantially due to our ancestors' different, historical toxic exposures.

If they were asked for any possible scientific justifications, ignorant racists would say they're discriminating against people on the basis of inferior genetics. What they are really doing is trying to devalue the cumulative epigenetic damage of another people's historical deprivation.

Lost within the data, Murray was almost certainly looking at some epigenetic consequences; but there was no epigenetic science back when he published. He needed all the scientific truth he could've gotten, to protect himself and his scholarly work from his racist, so-called friends.

- Dale Coba

This interpretation is not well supported by the evidence. The problem for the epigenetics uber alles viewpoint (the form of which you suggest is not supported by any of the population geneticists I've talked to, or the papers I've read) is that the twin studies show the same correlations for heredity that would be expected by statistical extrapolation of simple Mendelian genetic processes, at least for most traits. They also indicate that all shared environmental effects more or less cancel out by adulthood (usually age 25ish). While the science of epigenetics is very important and poorly understood, through the magic of statistical analysis we can determine what quantity of variation is likely due to it. That quantity is very low compared to the effects of traditionally understood inheritance.

Also, while Murray's statistical analyses did contain a few errors, most of them have been supported by later experiments and analysis, including most of the controversial ones. Have you read the book? It's definitely worth a careful read, including as many of the citations as you can get your hands on (when I did so I had the advantage of a University library with all its journal subscriptions, which I realize may not be the case for most of us )

[D.Olivaw]

Our various ethnicities' genes are not that different, not so much severely less fit from one group to another.

It also occurs to me ( a little late because it's worded differently than usual) that this is an example of the spectrum fallacy. The fact that there is no hue/frequency discontinuity in the color spectrum doesn't imply that red is the same color as blue, for instance. A consistent 1 standard deviation difference, even a 0.2 standard deviation difference between large populations means that though the vast majority of the area under the curves is identical there is an enormous difference at the far ends (3+ sigma). For traits that occur at those far ends, like mathematical prodigy or sociopathy, a very small difference in the overlap of the overall populations can lead to differences of as much as an order of magnitude of the occurrence of the trait in one versus the other.

[--NightBattery--]

[dale coba]

[I think I've stayed on topic and civil, but as you know, sometimes my old college debating snark escapes my notice or control.]

Lost within the data, Murray was almost certainly looking at some epigenetic consequences; but there was no epigenetic science back when he published. He needed all the scientific truth he could've gotten, to protect himself and his scholarly work from his racist, so-called friends.

This interpretation is not well supported by the evidence.

I'm sorry that the word "some" may have been unclear. I meant Murray would have had no way to clean up the noise epigenetics created within his conclusions. Not that he was wrong, I am in no position to say that. Is it of me wrong to suggest he was not in a position to understand the whole picture?

The problem for the epigenetics über alles viewpoint (the form of which you suggest is not supported by any of the population geneticists I've talked to, or the papers I've read) is that the twin studies show the same correlations for heredity that would be expected by statistical extrapolation of simple Mendelian genetic processes, at least for most traits.

Complex "traits" like I.Q. are the product of many different genes. The maths of Mendelian genetics as observed for discrete, simple traits like eye color would seem to me to have absolutely no bearing. Was this premise truly part of their reasoning? If so, OUCH.

Speaking of papers, have you read anything along the lines of the epigenetic explanation for homosexuality? Homosexuality as a Consequence of Epigenetically Canalized Sexual Development
Or Decanalization, brain development and risk of schizophrenia. The explanatory power of epigenetics to answer some of the hardest "genetic" questions we have left really blows my mind. I have to say, after an Australian Coursera I took last year on the subject, I realize I've drunk more than a little bit of the Kool-aid; but I'm also pissed that seven years after the Nova special, epigenetics seems a stranger to so many of the scientists searching for the broken gene that isn't broken, but paved over or misread.

They also indicate that all shared environmental effects more or less cancel out by adulthood (usually age 25ish).

I was talking about populations with historically distinct and uneven toxic exposure, not ones which have historically shared environments.

As for your statistical data sets, I don't think the consequences of toxic exposures could be fully seen within. The epigenetic damage in people alive today is from exposures that occurred a generation or two or more ago. There's a historical timeline, with milestones like the introduction of industrial coal mining, industrial chemistry in the latter half of the 19th century, plastics..., and while new toxins are invented and accumulate with every year, the damage we can see is only an echo of a much less toxic decade.

The numbers for autistic spectrum disorders have skyrocketed far beyond any plausible genetic explanation. You don't have to have frank autism to suffer loss of emotional intelligence, and unlike on simplistic tests, the ability to understand other people matters. So whatever steps it takes to show that effects cancel out by the 25th year of a person's life... that's so far away from causality, it's no wonder his work was never embraced more broadly.

Much of the controversy erupted from Chapters 13 and 14, where the authors write about the enduring differences in race and intelligence and discuss implications of that difference.....

Discuss the implications? Of racial genetics? When they didn't even know about epigenetics? I guess epigenetics must be truly irrelevant, if you think there would be no consequences to such uninformed speculation.

...Citing assertions made by Murray in The Bell Curve, The Southern Poverty Law Center labeled Murray a "white nationalist," charging that he has a long history of advocating discredited ideas that are rooted in eugenics.

In 2000, Murray authored a policy study for AEI on the same subject matter as The Bell Curve in which

Try to imagine a GOP presidential candidate saying in front of the cameras, "One reason that we still have poverty in the United States is that a lot of poor people are born lazy." You cannot imagine it because that kind of thing cannot be said. And yet this unimaginable statement merely implies that when we know the complete genetic story, it will turn out that the population below the poverty line in the United States has a configuration of the relevant genetic makeup that is significantly different from the configuration of the population above the poverty line. This is not unimaginable. It is almost certainly true.

This thing is, I have watched Mark Potok speak on many occasions over more than a decade, and his organization does not throw a label around like that lightly. I'm sure on more than one occasion I've idly speculated that I would take a bullet for the kind of man he is, for the great work that he and his organization have done to protect our democracy from racist radicals.

One way to be a much milder form of racist is to "professionally" speculate rampantly outside of your field, and these guys were political scientists (if one insists there even can be such a thing as a science of politics - an art, surely; but never a science, not so far).

While the science of epigenetics is very important and poorly understood, through the magic of statistical analysis we can determine what quantity of variation is likely due to it. That quantity is very low compared to the effects of traditionally understood inheritance.

I'm afraid your magic will have to remain just that to me, magic. Given that epigenetics was in no way a science at the time Murray was writing, this would have to be analysis conducted many years after Murray's original publications and the media sensationalism that tarred him.

Also, while Murray's statistical analyses did contain a few errors, most of them have been supported by later experiments and analysis, including most of the controversial ones. Have you read the book? It's definitely worth a careful read, including as many of the citations as you can get your hands on (when I did so I had the advantage of a University library with all its journal subscriptions, which I realize may not be the case for most of us )

Strange as it may seem, I have read very few books in the last few years - and Murray's splash was a long time ago, almost too long for me to care, even if he had been ultimately regarded as relevant and important - which does not seem to be the case. I'd be glad to look at a link, if you like, but if it leads to testimonials from some American Heritage Institute-affiliated political scientists' ramblings, I'm certainly not interested

- Dale Coba

[dale coba]

Comfortable there, Battery?
Can I get you anything? Freshen up your drink?

- Dale Coba

[darkbutflashy]

I think that's a nice demonstration of how news coverage totally exaggerates on anything being or claiming to be science. Should that be a follow-up to the original topic, you both totally passed the turing test.