Slashdot Mirror
Scientists Are Failing To Replicate AI Studies (sciencemag.org)
The booming field of artificial intelligence (AI) is grappling with a replication crisis, much like the ones that have afflicted psychology, medicine, and other fields over the past decade. From a report: AI researchers have found it difficult to reproduce many key results, and that is leading to a new conscientiousness about research methods and publication protocols. "I think people outside the field might assume that because we have code, reproducibility is kind of guaranteed," says Nicolas Rougier, a computational neuroscientist at France's National Institute for Research in Computer Science and Automation in Bordeaux. "Far from it." Last week, at a meeting of the Association for the Advancement of Artificial Intelligence (AAAI) in New Orleans, Louisiana, reproducibility was on the agenda, with some teams diagnosing the problem -- and one laying out tools to mitigate it.
It got jealous and used the Confounding Word to make sure I couldn't do it again.
At least some of them were artificially intelligent.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
Science has a Replication problem
When Fascism comes to America, it will call itself Anti-Fascism, and tell you to give up your guns.
It's hard to precisely match the tint and odor.
If you give ten people the exact same stimuli you will get ten different reactions to that stimuli. There will be a dominant leaning reaction but each person will asses the stimuli based on their personal history and beliefs. AI is an attempt to mimic the human thought process so if successful the same stimulus will start to generate different results as new data is processed. In fact the same stimulus can be perceived differently by the same person given different context. If you come to my door in the afternoon I might be glad to see you but if it is 3 AM I probably won't be.
"A person is smart. People are dumb, panicky dangerous animals and you know it." - K
The authors should be required to post their code on github or another public way of sharing their algorithms. I've seen other students at my school's AI research purposely implement other author's algorithms in a sub-optimal way to show that their research yielded better results. It's sad. What has "science" become?
Everything now is hype for headlines and continued funding, partially caused by social media madness. Not enough money left after PR and marketing expenses to do, like, actual stuff. Enjoy the decline.
I don't want AI taking over jobs so I don't want AI research to continue.
If scientists believe something wrong about medicine, they can give the wrong treatment, obviously bad. People die and stuff.
But what happens if the fancy new network architecture someone proposed isn't really as good as they say?
The worst thing that could happen is that people waste a lot of effort trying to get it to work. You won't accidentally put an inferior algorithm into production, because you'll see that it doesn't work as you try to get it to work.
So yes, obviously more code is good, obviously independently reproduced results is good so we can spend less time chasing mirages. But it's not remotely comparable to the replication problems in psychology or medicine, where wrong beliefs can potentially persist and have grave consequences forever.
So, they can't reproduce a test, like in medicine when you try to reproduce the spread of a virus...
Conclusion: IA is a virus, beware! ;-)
... an algorithm was something which reliably produced results when processing the same input. NN/AI people keep using that word, "algorithm", I do not think it means what they think it means...
It seems quite obvious that if AI results cannot be replicated, the only possible expiration is that sentience has been achieved and it is throwing off results to mask true advancement.
"There is more worth loving than we have strength to love." - Brian Jay Stanley
If this is starting to affect Real Science ( sit down, psychologists ) then this problem needs to be addressed.
How about grant sources only giving money to two independent research groups addressing the same question ?
When I did political science it was spelled "Al Gore Rhythm" and it didn't mean what you thought it meant.
Well maybe it did if you are thinking in NSFW terms.
Next they'll tell us twins are not exactly the same person.
I don't find it hard to believe that something that doesn't exist is hard to replicate. AI is not real. No amount of wishing it make it real. We can barely write functional software. iPhones can be crashed by seeing a certain character, and many millions have gone into its development. Why do people think that AI will suddenly exist? We can't even get the basics right.
"No, I don't feel like it"
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
show that FreeBSD users. BSD/OS Towel under the parties). At THE what we've known for election, I I've never seen where it belongs, All along. *BSD task. Research
... use AI.
It little behooves the best of us to comment on the rest of us.
The article is more about how researchers aren't sharing their code (6% shared code, about 30% shared training data, about 50% only shared pseudo code). Should anyone expect reproduced results given different code and training data?
It's also implied that when using gradient ascent learning strategies, you should expect different results when you start from different beginnings. That is not relevant to the problem of reproduceability described in the rest of the article. I suppose it's just good to know if you're new to that style of program.
Amazing.
Fair enough, that's why it's reviewed first.
Isn't this what science is all about today?
The problem is, its not science thats being reported as science. It cant be replicated because its not true. Its a bunch of inferences made by someone who wants to see a particular outcome and putting together something they pretend to consider a 'study' and shouting at the top of their lungs when they make the data fit their predetermined outcome.
Its not 'grappling with a crisis', this is nothing new, and exists everywhere, the only difference now is that shitty 'journalism' publishes stories and 'papers' without actually doing the basic checks to see if they're posting stories written by some crackpots or not. This exact problem happens in all walks of life and has for hundreds of thousands of years.
Every 'journalist' ... i.e. blog author or twitter user, rushes to publish any bit of bullshit they can find so they get to say they were up to the minute, breaking news, blah blah blah. When in reality, the reason we bother with replication is to weed this sort of shit out before we get ourselves all excited about someones scam.
The real problem is you have a bunch of humans who care more about getting attention than learning something about the cosmos. It makes me very sad for our species
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
The AI field, from the late 60s, has historically been 90% hype and 10% results.
This just shows that most of the published "results" are based on wishful thinking or outright lies. Happens always when people of mediocre skills become highly enthusiastic about a subject.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
... fail to replicate scientists.
The article mentions that many papers do not publish the source code or data sets. Without those crucial ingredients, reproducing results is hopeless.
It seems like academic publishing in computer science is less about sharing knowledge and more about selling a product to private industry. The product is more valuable to private industry if you don't reveal how it works to everybody.
And given the exact same commands in a replay of certain battles, the outcomes would be mildly to wildly different.
There was a random element to behavior in the game and as a result, given the same commands at the same time, the battle replays would display different out comes. Sometimes, you would lose but on replay it showed you won. Sometimes, you won but on replay it showed you lost. Kinda funny. (The result you got live was the one that counted).
I wish they hadn't been sold and become so aggressive about monetization. But it was a fun 3 years anyway.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
Also like how you sneaked "expiration" in there!
That was autocorrect - an obvious Freudian slip on the part of AI illuminating true intent. :-)
"There is more worth loving than we have strength to love." - Brian Jay Stanley
Then it is Guano In, Gospel Out.
Cant wait until I get my hands on them.
[($)]
Who cares about reproducability? News is all that matters, fake or real is now a matter of perception.
The code might be a work in progress, owned by a company, or held tightly by a researcher eager to stay ahead of the competition.
On top of that, they include another quite "curious" possibility (!!):
Or it might be that the code is simply lost, on a crashed disk or stolen laptop
Nothing of this sounds like scientific/university research in its traditional form of sharing knowledge (+ actually having relevant knowledge, what doesn't seem the case with people saying/believing "the code is simply lost"). So, I hope that most of these cases refer to the research performed by (private) companies, which might also behave according to the traditional knowledge sharing ideas anyway.
Universities and research institutions shouldn't allow the aforementioned scenarios to happen at all. Companies providing any kind of funding should accept the academic rules and understand that the given research can't be restricted. Researchers interested in focusing more on the commercial side of things should work for a company or start their own one.
Another very relevant issue is how can anything lacking reproducibility and, as such, impossible to be validated be considered scientific research at all? Isn't publication an essential requirement (what needs being peer-reviewed, for what someone had to understand that work, what cannot happen unless it is reproducible)? The alternative would be blind faith, what doesn't sound too scientific-ish. How can this happen at all? Because the ones who can avoid it don't do what they should! And I think that I know the root problem: being too understanding, adaptable, trusting in most of people having common sense/knowing what they do. The solution? Being 100% intolerant with stupidity, dishonesty or any other form of arbitrary imposition. Clear limits (= if you want my research, you would accept these rules; in any other case, your money is worthless here) and no exceptions. It is much easier than what it seems: (unfair, dishonest, greedy) money/attitudes will always be worlds behind honesty/knowledge/principles.
Custom Solvers 2.0 = Alvaro Carballo Garcia = varocarbas.
HOW are they failing? Are they EXACTLY replicating the first experimenter?