Researchers in Japan have tried to build an artificially intelligent system to make people laugh – but, surprise, surprise, the jokes it told were terrible. The “Neural Joking Machine” (NJM) was created by computer scientists from Tokyo Denki University and the National Institute of Advanced Industrial Science and Technology …
If I were them, I would try to learn to walk before attempting to climb the north face of Mount Everest.
Humour requires to understand humans and their social conventions. It's the unexpectedness of the statement which strikes us as funny, and an AI can't possibly know what is considered as "expected" by humans. Today, in 2018, AIs are good at processing data, not at understanding it.
"Can't just have them all intoning "I'll be back"."
Only the violent, gun-loving, Austrian ones do that. They will then forget that anything ever happened. The calm, quiet, civilized, American ones will merely tell you that they're sorry, they can't let you back inside. Less mess that way. (Unless the American ones are from Detroit. Then you'll have 20 seconds to comply.)
Yeah... I'm not so sure. It seems to me that AI should be able to take a reasonable stab at what's "expected" by humans, if someone takes the trouble to train it on a relevant data set.
My phone's keyboard software is pretty good at guessing what I'm going to type. Rudimentary, of course, but for such a tiny and limited system using essentially zero resources, it's impressive.
@veti. Really ? All my devices predictive suggestions manage to be wrong most of the time. I wish the smart devices would just shut up. Perhaps you have a let unpredicatble vocabulary. No offense intended as I use a few specialist acronyms for uncommon activities. DYAC site best describes my attitude. As for AI jokes, given that most supposedly intelligent humans suffer from an inability to discern good humour, I am sure artificial stupidity will be just fine in merkin mass media.
Some forms of computer generated humor are trivial.
For example, most proverbs have two logical components - a premise and a conclusion. Split them and run a random generator to match the parts. We did that in high school in 1984. Some of the products of that experiment are still doing the rounds in the country where that was done.
"If I were them, I would try to learn to walk before attempting to climb the north face of Mount Everest."
While it does seem a herculean task , nobody said AI would be easy . Voice recognition and playing chess just dosent cut it. You're virtually creating life. I'd say the 20 odd % success rate mentioned in the article is a win!
Its a good test of AI is what i'm saying , maybe we're not ready yet as you say.
After all - it was this very concept of humour test that finally convinced the scientist that:
Johhny5 is ALIVE!
"AIs are good at processing data, not at understanding it."
Well thats not an AI then is it?
@10 hrs Prst. V.Jeltz
Is it really AI though or just shuffling enough words into an order that might make sense and those from a specific set of phrases? For instance, if the AI device had never seen a Terminator movie, but there were quite a few phrases from those movie that are funny in the context of the pictures would it be machine learning that understood why those were funny and when to use them?
I'm not saying that these aren't advancements, but I think we're a long way from machine intelligence.
Since we meatsacks can't tell what is funny and what is not, how can we expect a machine to be able to - can we even judge if the machine is funny nor not? Maybe it is but we just don't get it.
Try telling a mother-in-law joke, or almost anything by Frank Carson, Jim Davidson or any other comic from the 70s and 80s and see how long it takes the Politically Correct mob to start baying for your blood...
I reckon it'd have done better if it was compared against captions generated by Register commentators. It can't do much worse than 4 decade old Monty Python routines repeated ad infinitum, or a Cancy McCancerFace joke that the rest of the world realised was mildly amusing on first discovery, then halved in mirth value after every repetition.
I've seen better stuff from a random phrase generator. Come to think of it, that's what most of the AI output has been like. Perhaps what we've really found is a completely different way of implementing random data generation based on an input data set.
I've seen better stuff from a random phrase generator. .... Brian Miller
Is that better stuff in response to nonsensical random phrase generation, BM? And can it also support arrogance and ignorance with the Gifting of IT Opportunities to provide them with the utility and facilities, if not the necessary ability, to reply sublimely and surreally and move markets to new leading positions/stellar heights/monumental depths?
I love the implied "half-life of humour". I think that is why most knock knock jokes are entirely devoid of humour as one hears more of them more times.
I beg to differ. Most knock-knock jokes are as topical now as they ever were, because most of them are not topical.
Knock-knock jokes becoming unfunny as you encounter more of them is more likely down to Information Theory: the information content of a message is related to its surprise value. Once you've heard enough knock-knock jokes you can often predict the punch line.
I find the same thing happens with sitcoms, except they get stale faster for me than for most people. I become bored with them long before everyone else does. When I was young two series was enough for me before they degenerated into (seemingly) nothing but predictable punch-lines. As I aged I became better at predicting content of new sitcoms because the underlying format was so familiar. I watched every episode of Friends by watching one episode.
The same thing can happen with other comedy formats (at this point I predict most of those who upvoted me based on what I wrote so far will change it to a downvote). I watched everything Reeves and Mortimer ever did, and ever will do, in 10 minutes. "You wouldn't let it lie." Frying pan to face. "You wouldn't let it lie." Frying pan to face. Repeat ad infinitum.
As for "AI", artificial intelligence, at this time it remains HypeHypeHype. What we have are advanced expert systems, built upon technology that's been evolving for half a century and still hasn't reached any level of actual intelligence.
What's made expert systems seem sort of intelligent are the improvements in speech recognition and text to speech technology along with faster processors. I will also thank the ongoing development of 'machine learning'. Meanwhile, our programming tools remain primitive and unreliable. Our software security record is abysmal. The ongoing fad of IoT, Internet of Trash, is more a detriment to our collective lives than a benefit specifically due to it having no inherent security with only hints at a security standard on the horizon. We're all becoming victims of marketing SellSellSell of JunkJunkJunk.
Let's get down to some serious work please and grow our way out of the ongoing Dark Age of Computing. Please.
Meanwhile: HypeHypeHype SellSellSell *yawn*
This is incorrect. Adam Sandler is a very talented man, very talented indeed. However, his talent is not for acting, at least not in the way that most people would think. No, his talent is in extracting lots of money while creating utterly unwatchable messes. He approaches Uwe Boll levels of skill for that kind of thing.
> At least they were considered funny about 67.99 per cent of the time, compared to 22.59 per cent for the NJM, and just 9.41 per cent for STAIR.
That would still make either of the two machines funnier than any british sitcom that has been made in the past 10 years.
The emphasized part of your comment highlights one major flaw that the study has, which is that STAIR results were translated from English before the Japanese public rated them against NJM. Of course it did worse - jokes almost never survive translation especially not between cultures as different as European and Asian. Even within Europe, jokes often need supplemental information before the viewer realises it should be humorous. Mistranslations are often far funnier than the mot juste.
Of course they could have done it deliberately to show they were on the right track and raise more funding for their study. But maybe that is just me being a cynic.
Spot on JassMan. I was wondering about that part.
After having to learn several languages while meandering around the globe, I've only ever considered having a decent grasp of the local language, when I could actually not just understand what was being said, but able to understand the jokes, with all the local contexts, and laugh about it.
Japanese humour is by far not the same as US, or British humour in this case.
Also, if anyone could develop an AI chatbot, that would recognise where a person was from, and then switch the conversation to match the cultural limits, and courtesy rules of that person's locale, now that would be impressive. If neigh-on impossible. Even humans seem mostly incapable of it.
That would still make either of the two machines funnier than any british sitcom that has been made in the past 10 years.
I was trying to think of examples to contradict you and I'm struggling to think of any...*
*Please feel free to suggest some I've missed, it's almost the weekend.
The US version of "Shameless" is really quite good, although the first few episodes are a bit weak, being straight off copies of the UK version. Well worth a watch.
While Frank is still pretty much played for laughs, the other alcoholic characters are well written. Painfully so, if you've ever had a someone you care about suffer from it.
> I was trying to think of examples to contradict you and I'm struggling to think of any...*
Quite. Even the examples presented fail, since most are more than 10 years old (Outnumbered started in 2007, IT crowd 2006 - neither still running) And ones that do qualify, such as Bad Education only managed less than 10 hours of telly - 19 eps of ½hour each and has since been canned.
Instead of trying to build an artificially intelligent system to make people laugh, they should build one that makes people cry. Computers are already good at that. Who can forget the old classic that, even to this day, still brings a tear to my eye...
Not ready reading drive A
Abort, Retry, Fail?
'“Laughter is a special, higher-order function that only humans possess,” they wrote in a paper emitted online this week. It’s something that is difficult to quantitatively measure, but they gave it a shot anyway.'
I'm always annoyed at "X is something that only humans do" nonsense. Especially for something that is "difficult to quantitatively measure". Almost all boil down to "we never bothered to check if any other animal can do X, but we state without a doubt that only human animals does X". You can often find research that shows that other animals do indeed do X.
I know from personal experience that Australian Magpies, while not physically able to laugh, are quite capable of a corvid equivalent. - A tiding of about 6 magpies (typical tribe size), sitting in a line of a fence watching a stupid pee-wee trying to do a typically Magpie swoop (on me) and reacting with obvious shared amusement when it failed spectacularly!
"Laughter is a special, higher-order function that only humans possess"
Laughter is common among primates, and there's decent evidence for similar behaviour (ie. enjoyment coupled with specific vocalisations) in plenty of other animals. Look up videos of things like crows giggling to themselves while tobogganing and try to convince yourself it's meaningfully different from human laughter.
Perhaps people attempting to study behaviour should learn at least the basics before diving in and trying to publish meaningful results. It seems particularly odd in a case like this where the goal is essentially to reproduce human-equivalent behaviour or intelligence; failing to understand what "human-equivalent" might actually mean really isn't a good start.
The “Neural Joking Machine” (NJM) was created by computer scientists from Tokyo Denki University and the National Institute of Advanced Industrial Science and Technology to see if humor could be automatically generated and studied academically.
If this doesn't sound like the setup to a really lame joke, I don't know what it is. Word of advice to the researchers, if you have to explain the joke, it's not funny. If you have to explain how to tell the joke, too, it was never going to be funny.
The Vicar at the chuirch I used to attend once went to visit churches in Japan we had links with via a Japanese missionary who had attend our church in th epast. When there he preached at services with the this contact translating into Japanese for the listeners. During one talk he was telling one of the jokes he used but as he told it he realized the humour depended on a combination of the Englsh words used and the meaning in different contexts so he was surprised to get lots of laughs at the end. After the service he raised this with his interpreter and said how surprised he'd been that they got the jjoke - only to get the reply "Japanese people aere very polite and would not want to embarrass a visitor so I told them that you were telling an english joke and could they please laugh at the end"
“Laughter is a special, higher-order function that only humans possess,”
Starting from an incorrect premise is always going to give dodgy results. Laughter is derived from the standard fear response of other primates. There is evidence to suggest human laugher is still an involuntary fear response, since many jokes depend on the unexpected, or the subject being placed in an uncomfortable situation.
I thought research was suggesting that laughter fulfills a social function and is not restricted to primates.
Certain forms of humour do rely on fear – Feydeau said that the difference between farce and tragedy is timing – as exemplified in slapstick: we anticipate the custard pie or the anvil… That is laughter and humour are related but not the same.
It’ll be a long while yet before Alexa, Siri, or Google Home can make us laugh
I don't know; "I'm sorry, Dave, I'm afraid I can't do that" might hit the spot at the appropriate time.
Humour's a fickle thing. I've watched entire episodes of comedy sitcoms without exercising a single face muscle and they are written by teams of writers who have supposedly honed their craft over many years. On the other hand I have near wet myself at a perfectly timed expletive.
Good luck to any AI trying to figure out what's funny and what isn't
TARS: [as Cooper repairs him] Settings. General settings. Security settings.
TARS: Honesty, new setting: ninety-five percent.
TARS: Confirmed. Additional settings.
Cooper: Humor, seventy-five percent.
TARS: Confirmed. Self destruct sequence in T minus 10, 9...
Cooper: Let's make that sixty percent.
TARS: Sixty percent, confirmed. Knock knock.
Cooper: You want fifty-five?
Biting the hand that feeds IT © 1998–2019