Slashdot Mirror
AI Can Predict When Patients Will Die From Heart Failure 'With 80% Accuracy' (ibtimes.co.uk)
New submitter drunkdrone quotes a report from International Business Times: Scientists say they have developed an artificial intelligence (AI) program that is capable of predicting when patients with a serious heart disorder will die with an 80% accuracy rate. Researchers from the MRC London Institute of Medical Sciences (LMS) believe the software will allow doctors to better treat patients with pulmonary hypertension by determining how aggressive their treatment needs to be. The researchers' program assessed the outlook of 250 patients based on blood test results and MRI scans of their hearts. It then used the data to create a virtual 3D heart of each patient which, combined with the health records of "hundreds" of previous patients, allowed it to learn which characteristics indicated fatal heart failure within five years. The LMS scientists claim that the software was able to accurately predict patients who would still be alive after a year around 80% of the time. The computer was able to analyze patients "in seconds," promising to dramatically reduce the time it takes doctors to identify the most at-risk individuals and ensure they "give the right treatment to the right patients, at the right time." Dr Declan O'Regan, one the lead researchers from LMS, said: "This is the first time computers have interpreted heart scans to accurately predict how long patients will live. It could transform the way doctors treat heart patients. The researchers now hope to field-test the technology in hospitals in London in order to verify the data obtained from their trials, which have been published in the medical journal Radiology.
This is pretty meaningless if the doctors are 90% accurate. Also what would the accuracy of just saying very patient is going to be alive in a year. Maybe 85%?
Exciting AI can interpret data and give a result, wake me up when it can actually do something more useful like predict when I want a coffee with 80% accuracy and make it for me.
A /. editor who does understand what paragraphs are for....
if (!mt_rand(0, 8))
{
kill_patient();
lie_about_detecting_it();
}
No. It's just a fucking program. All of these "AI" claims are just programs. AI doesn't exist. Just stop with the AI already, the word has lost all meaning.
1) Eventually.
2) Eventually.
3) Eventually.
4) Never.
5) Eventually.
WEASEL: So the doctor says: "Well the bad news is, you don't have that much time to live." He says, "How long do I have?" The doctor says, "Five." The guy says, "Five what?" The doctor says, "Four, three, two..."
You know the joke...
"10"
"10 months?"
"...9...8...7..."
The same algorithm can predict when a patient is already dead with 80% accuracy.
week of September, and while I wish I was dead, I pretty sure I'm still alive. Well, for certain definitions of still alive. They also said my liver cancer would kill me before the end of November in 1998. I hate my life and have nothing to live for, but again they were wrong. The only thing I'm looking forward to is seeing the end of The Big Bang Theory since it just keeps getting worse and worse, and I can't not be curious about how bad that show is going to get. Sara Gilbert was in the last episode I watched which was nice, but the rest of the episode was Dr Cooper's birthday which made me insanely jealous. I was happy then that made me sad.
Don't let the GOP have this any flagged will be black listed.
It can occasionally be off by a few seconds, what with free will and all.
#DeleteChrome
Professor Farnsworth invested the death clock, which is capable of predicting one's death (from heart disease or whatever) with an accuracy of a few seconds, all the way back in the 30th century.
the software was able to accurately predict patients who would still be alive after a year around 80% of the time
"First they came for the slanderers and i said nothing."
Start a better database on doctors and all their medical procedures.
Find good pathologists to report on every case. Link all past complex work with a pathologist's reports over years.
Who ordered what procedures? What did a specialist do? What did an average doctor on duty do?
Over time the really good professionals who have the skills to save lives will get listed and the average doctors who do things wrong will really stand out.
Work out who your best specialist are, support them and get them working with the next generation of experts.
No new AI needed. Just track all your doctors and have pathologist's report on every interesting case.
Why some doctors just cant get the same average results as a specialist on duty is a question that can be discovered by looking at results.
If a hospital wants good results, stop funding average doctors and having average doctors try to deal with the same very sick people every shift.
Move the average doctors to other areas of medicine and ensure only the best staff get supported.
How to stop the flow of very average doctors into the wider medial profession? Stop accepting very average students into universities to study medicine. Make sure every doctor sits the same national exams and passes well.
Domestic spying is now "Benign Information Gathering"
false positives or false negatives or what mix of those?
"Win treats sysadmins better than users. Mac treats users better than sysadmins. Linux treats everyone like sysadmins."
and wants us to die so hard, it doesn't take an AI to determine that we are going to die. We are all going to die. I'm most especially distraught since I've never been with a girl. With a girl.
This is pretty meaningless unless the full Confusion matrix (with false positive and false negative %) is provided
Researchers from the MRC London Institute of Medical Sciences (LMS) believe the software will allow doctors to better treat patients with pulmonary hypertension by determining how aggressive their treatment needs to be. ... The LMS scientists claim that the software was able to accurately predict patients who would still be alive after a year around 80% of the time.
Hmm... Software predicts patient probably won't be alive after a year, doctors don't treat as aggressively, patient dies shortly thereafter. Sounds like researchers have discovered the self-fulfilling prophecy.
Of course, anyone could make those accurate predictions by simply killing off patients -- of course, not 100% accurate as that would be suspicious.
It must have been something you assimilated. . . .
does not have anywhere near the same meaning as
This is predictive modeling as done at least the last 20 years. For example SPSS Modeler (formerly Clementine) was released 1994, and similar stuff was done with custom solutions long before that..
From TFA:
The LMS scientists claim that the software was able to accurately predict patients who would still be alive after a year around 80% of the time.
That's quite different from claiming that the software is able to correctly predict when 80 percent of the heart patients are going to die.
One thing that never changes... journalists fuck up statistics, and people pushing studies or products are willing to help them fuck up so that the public can be properly deceived.
First, this is not AI, it is statistical classification. No intelligence involved at all. Second, at 80% accuracy (and the thing it actually predicts....), it sucks badly. You will probably have to search a while to find an experienced expert that does this badly on this classification task.
So while this may have some value as triage, that is about it. And even as triage, it may kill people because of the 20% where it is dead wrong. So maybe this is useful and maybe not. It will of course get misused as the AI fans are lacking in natural (i.e. actual) intelligence.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
Yep! It sounds like a statistics program - not really AI.
Can I haz ur toyz?
Sheesh, evil *and* a jerk. -- Jade
In the US with our health care System it will have to be AI to work
The only reason the term doesn't have any meaning is because everyone's definition of "intelligence" is different in the first place. If you can define an unambiguous metric for intelligence, then it becomes pretty obvious what AI has to be: intelligence that is artificial, rather than natural.
File under 'M' for 'Manic ranting'
That's how it was back to the 60s, eventually they will find another buzzword and move on.
The health insurance and organ harvester monsters are like "Predict away my precious..."
What this usually means is that the 'fucking program' uses neural nets which
emulate aspects of how the brain works, and trained these neural nets on the
problem data at hand. The neural net thus gains the information and can apply
this to predict some situation.
Anything in the 'artificial intelligence' department will be a 'fucking program'.
AI has not lost its meaning, rather - it has been rather poorly defined
to begin with. Some think it means sentient, other think it means sapient
other think it must be human-like self-aware cognition.
In reality intelligence means 'the ability to apply knowledge and skills'
This program has a very limited knowledge / skill set indeed, and this
information has indeed been acquired in a very limited way. This limitation
seems enough for some to discount the term 'intelligent'. Indeed such a
definition would make almost any program 'intelligent' in some way. And
it is this flaw in the definition that would justify that 'AI does not exist', or
alternatively 'AI is trivial, but perhaps not what you were hoping for'.
You mean like back in the day when it was called with less sexy Multivarate Analysis? Ahhh, 1958 had such unimaginative people,
https://en.wikipedia.org/wiki/...
I can remember when /. attracted some of the best informed and intelligent people you could find anywhere on the Internet. Now, all we have are members who inform us that we do not need computers, and their ability to find correlations in huge data sets. All we need is the intuition of smart people.
Look back at the early proposed tests for artificial intelligence. When supervised deep learning systems can use the immense processing capabilities of modern computers, to not only match, but to exceed the capabilities of humans in a wide range of problem spaces, it is appropriate to describe the result as "artificial intelligence". We do not mean literally that we have an intelligent bunch of integrated circuits and harddrives. But, the overall system can produce results that we would until recently have considered only achievable by human experts. Indeed, our AIs, in many situations, exceed the capabilities of the best human minds.
I am used to the idea of the general public feeling threatened by the capabilities of modern technology. I just wish sites supposedly intended for intelligent, scientifically-informed individuals could be exempted from such lack of reason.
This has always weirded me out as an argument. Intelligence is simply information processing in physical systems. What humans do when they make intelligent decisions is take in data from the environment and process it to make predictions. So when a doctor looks at the same data as the machine and makes a conclusion about treatment, he is making an intelligent decision, but when the machine does the same it's not intelligence it's 'just a program'? When a human being operates a motor vehicle and adapts his/her behavior to react to oncoming traffic he/she is using intelligence but when a machine does it it's 'just a program'? What?
This whole approach to me reeks to substance dualism; the human brain is a computer, a very advanced one at that, but it's just a computer. there is no 'soul' that somehow makes the human brain the only thing that's capable of intelligent operations. Right now brains are still able to cross-reference data better than computers, making humans as a whole more intelligent than computers but we're already at a point in which computer programs are able to take in data and adapt their behavior to meet a goal, whether it's driving a car or anything else, with better results than your standard humans, but somehow the platform that the program is being run on determines which of these 2 actions is categorized as 'intelligent'. This is nonsensical.
Intelligence is a scale, not a binary thing. The confusion about AI these days is people read 'AI' and they immediately equate that to either 'human level intelligence' and/or 'consciousness', neither of which are required for a system to be intelligent. A dog is more intelligent than a rat, a monkey is more intelligent than a dog, a human is more intelligent than a monkey and so far overall humans are also more intelligent than computers, but it doesn't mean that computers don't have a level of intelligence already, even though you cannot (yet) have a discussion/debate with the computer.
Imagine a few decades into the future wherein these systems are able to recognize speech so that a physician is able to consult with it during an operation. Or when they get to the point that the computers themselves are able to perform autonomous surgeries on people and react to complications on the fly. This is the direction we're headed to and getting there does not require the computers to become self-aware.
You can call it 'just a program' all you want, but using that definition the years of training and practice going on in the surgeon's head as he's trying to figure out the best way to cut the tumor out without causing a hemorrhage is also 'just a program'. The platform on which the program is being run may be wetware or hardware but it does not affect the intelligence of the program.
"It is the business of the future to be dangerous" -Alfred North Whitehead
Funny thing - she didn't bother telling me what I might accomplish if I start eating right, or exercising more, or even if I quit smoking - in fact, she seemed rather dubious that it would have any real effect at all (except the smoking part, for which she was happy to suggest several types of help if I wanted it). She didn't even tell me what my odds would be if I did start spending money on these drugs. I'm sure that insurance will pick up almost all of the cost - and I'm also sure that some pharmaceutical company somewhere would make a fair chunk of change off me for the rest of my unnatural life, sort of an annuity for big pharma. Problem is, I couldn't be sure I'd always be able to afford the drugs, and I'm told "once you start, you can't stop".
Yeah. I think I'd rather die living my life than clutching for more days.
I'd sure like to meet him. Good old Al.
"Nine times out of ten, starting a fire is not the best way to solve the problem." - my wife
This is called the AI effect.
Quoting the article, "as soon as AI successfully solves a problem, the problem is no longer a part of AI."
I predict 100% of patients with heart failure will die within less than a hundred years. Every last one.
Claims of 80% accuracy mean nothing without a lot of context, and little with.
Ahh yes, to be a myopic boomer watching as the world passes by, thinking time and progress has frozen. If only those dumb kids could see it!
I'm pretty sure a good cardiologist with a table of all aggregated research on heart conditions and life expectancy can do pretty much the same.
Health insurances have been doing this sort of thing for decades.
No big deal really.
We suffer more in our imagination than in reality. - Seneca
Closer examination revealed that the main parameters for decision making is the answer to "Patient is insured" and "Patient is a donor".
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Calling it a machine learning program is probably accurate. These are self taught predictors but there is still a lot of human involvement to calibrate the learning
My personal feelings is AI should refer to something that learns independently or in a black box manner. I think some neural networks can do this at a simple level but I don't know of any complex independent learning done by any computer programs.
For some reason, in reporting, they prefer to use the term "AI" rather than "fucking program".
Unless you are planning sexual relationships, "AI" is actually the more accurate term.
http://dilbert.com/strip/2015-...
http://dilbert.com/strip/2015-...
Curiosity was framed; ignorance killed the cat. -- Author unknown
Looks like AI is the buzzword of 2017 like "Big Data" back in the days.
sudo rm -r -f --no-preserve-root /
AI can predict who will die with 100% accuracy!
But it looks like an opportunity for the return of Microsoft Bob in Windows 10.
"Hi! Looks like there is a 77.2% chance that you'll die from a heart attack within the next five minutes. Shall I call an ambulance for you?"
This would probably make me obsess over my own demise for the remainder of my 80% accurate lifespan. Not something I'd want to know with such accuracy.
http://github.com/gbook/nidb
This whole approach to me reeks to substance dualism; the human brain is a computer, a very advanced one at that, but it's just a computer. there is no 'soul' that somehow makes the human brain the only thing that's capable of intelligent operations.
Start here...https://aeon.co/essays/your-brain-does-not-process-information-and-it-is-not-a-computer
Tat Tvam Asi
In the 1980's it was "expert systems". Sit down with an expert and walk through the decisions he or she made.
Then it was "fuzzy logic" because binary decisions didn't always come to the right conclusion, and different factors had different weightings.
Then it was "data mining" to handle vast amounts of data without any human bias. Now it is 'machine learning".
In addition, it's such an ignorant devaluation of AI's incredible achievements. The field of AI, a term everyone except Slashdot naysayers have agreed upon, has performed incredible feats of image and language analysis (among other things) that the average software engineer stringing together API calls could never have come up with.
Honestly, if you told someone to just "write a program" to recognize whether an image has a cat in it or not, what do you think they would come up with? Probably 20,000 lines of messy code that only handles a handful of cases. The insights of learning parameters from data, optimization techniques, and convolutional neural networks were found by groping in the dark for 50 years. For 50 years people couldn't just "write a program" to do that simple task. For a long time people couldn't even write programs that recognized the 10 handwritten numerical digits with competitive accuracy to a human.
Oh my gosh, why does everyone make a big deal about "making cars?" It's called "engineering a solution," and we've been doing it for millenia! Can we just call "cars" what they really are, "engineered solutions?"
The human brain is not a computer because computing is a very small part of thinking and intelligence.
And all of those have been branches of the AI field. Since the field of AI was created, arguably during the 1956 Dartmouth workshop, it has included topics such as expert systems and statistical methods being used to make decisions based on both simple and complex input. Whether you want to accept it or not, even those Pacman ghosts' behavior can be accurately referred to as AI.
If you want a term which only refers to human level intelligence, perhaps you should use either Artificial Consciousness, Machine Consciousness, or Synthetic Consciousness. Those probably match your personal definition of AI better than the broad definition used by the scientific community.
-- All that is necessary for the triumph of evil is that good men do nothing. -- Edmund Burke
We lost the fight with quadcopter v drone. I predict you have lost the AI fight as well.
Far more interesting than how good it is at spotting the average patient, is how close it was on its wrong results. How many people did it give a life expectancy of 30 years that croaked the next week? BMI is a decent measurement, when used on an average human, but is laughably retarded when used on anyone outside of normal. Using a AI, to decide when and how to do treatment might seem like a good idea when you look at the averages, but could be pretty bad for individuals.
Take for example the BMI again. I could program a AI, put it in a black box, and have it decide which patients need a stomach stapling. It would be right 99% of the time, but would send Arnold Schwarzenegger and pretty much every professional football player to get their stomach stapled along with honey boo boo.
Troll is not a replacement for I disagree.
The problem is that a human is able to step out of their algorithm and ask themselves, "Is this really correct?". You may say that this is a design flaw, in which human error can crept in, but this is what gives us our creative process.
they can decide to withhold care from people that they think will be dead anyway. Thus ensuring they are right and saving money
Firstly, even if one agrees that this is true, the human brain is still a computer, even if it's not entirely a computer. We're able to perform calculations and predictions based on learned experience, which amounts to computing. You can argue that on top of this the human brain has some 'extra' abilities that make it 'more than just a computer' but I'm not so sure of that.
The brain is in charge of not only conscious thinking but maintaining the life of the organism. It does this by receiving information from the nervous system and controlling the operation of the organs. It obviously does not (as others have pointed out) receive or process this information in the same way as a human built electronic computer does, but it does so nonetheless, so my way is to view the brain as just an incredibly advanced form of a computer that's way beyond anything we can currently even conceive of manufacturing ourselves.
"It is the business of the future to be dangerous" -Alfred North Whitehead
So is a computer. Error-checking and adapting is something that computers are still not very good at, but the whole point of machine learning and adaptive algorithms is to achieve this same ability.
Humans are able to error-check their own decisions and calculations only so far as we've learned to do it. If I've performed some task 10 000 times I know what kind of errors can occur in the process and can account for those. If I'm presented with a problem that I've never seen or solved before, my capabilities to do error-checking are nonexistent unless I first study the problem to find out which sort of errors can come up.
The same is true for machines: they can do error-checking in so far as they have data and experience to check against. The more data they have in their use, the better the error checking capabilities. A learning machine does not perform the same calculations time and time again, it looks at the result and attempts to change the calculations until the desired outcome is achieved.
"It is the business of the future to be dangerous" -Alfred North Whitehead
Al can predict? Al who? Alvin chipmunk? Al Bundy? Al Smith?
After twenty-five years in philosophy and having worked in the philosophy of mind for many years, this must be about the weakest argument against hard AI that I've ever heard.
No thank you, but trusting the job-destroying AI is not exactly the best of ideas.
Twitter supports and protects racists - by smearing their critics with the "Hate Speech" label.
And all of those have been branches of the AI field. Since the field of AI was created, arguably during the 1956 Dartmouth workshop, it has included topics such as expert systems and statistical methods being used to make decisions based on both simple and complex input. Whether you want to accept it or not, even those Pacman ghosts' behavior can be accurately referred to as AI.
If you want a term which only refers to human level intelligence, perhaps you should use either Artificial Consciousness, Machine Consciousness, or Synthetic Consciousness. Those probably match your personal definition of AI better than the broad definition used by the scientific community.
People have this illusion that AI is a continuum to Computer Sciences. It is not. AI is not digital. AI is not programmable. Computer Scientists and Programmers won't be part of the achievement. It will most certainly come from some physics lab [also involving mathematicians and neuroscientists] when there is a complete Theory of the Mind, which is still beyond the horizon.
This whole approach to me reeks to substance dualism; the human brain is a computer, a very advanced one at that, but it's just a computer. there is no 'soul' that somehow makes the human brain the only thing that's capable of intelligent operations.
I'd suggest that it has less to do with dualism and more to do with general intelligence vs. domain intelligence. Most of these AIs have decent domain intelligence, but very poor or nonexistent general intelligence. It's hard for most people to think of an autonomous car as being "intelligent" when it has no means at its disposal for answering trivial questions such as, "Can an alligator run the hundred-meter hurdle?".
When people today dismiss AIs as being "mere programs", what they're really doing is dismissing AIs on account of their lack of general intelligence. They're calling attention to the fact that AIs are only as intelligent as they've been programmed to be, rather than making an argument about dualism. In fact, most people I've talked to have no problem calling AIs "intelligent", so long as you add the caveat that the AI's intelligence is limited to a very narrow domain.
That said, the day that we have a decent, general domain artificial intelligence is the day that I think a lot of these "it's a program, so it can't be intelligent" arguments will start to become about dualism (or possibly will be about the distinction between weak and strong intelligence). Right now, those arguments are merely standing in as proxies for "it's only as intelligent as it's been programmed to be", i.e. it only understands some things, but one day that may not be the case. I believe you're just thinking ahead a bit, since we're not there yet.
Show me a robot that can cook a meal, drive a car, compose a song, read a poem, mend a fence, and paint a picture and I'll call it AI.
Also, no one said anything about 'soul'. Surely /. is still capable of rational discussion without bringing religion into it?
Hi, my name is Zoltar. Please insert the probe into your butt and put on the wired cap. Insert $5 bill and press the start button.... Thank you.... You will die on Oct 11, 2018."
Although I agree that "intelligence" is a scale rather than a switch and totally agree on the confusion point about people. Really people's fear isn't AI, but automation. But I disagree that your examples should be on that scale. Maybe on a "program complexity" scale but not intelligence. Maybe these two meet at one point or blur together a little.
But "intelligence" should be more than applying past algorithms to current parameters. Intelligence may start at "learning" new knowledge which is a bit more than just reading books. Its imagining the valid and invalid possibilities of the application of such knowledge. Its that "creation" of the internal knowledge that makes something "intelligent". Intelligence moves along the scale in terms of the number of potential possibilities of that application that can be processed. Basically the ability to digest and internalize a new piece of data across the population set.
Computers are no where close to the kernel of intelligence. WE humans are STILL having to painstakingly through almost a trial&error process, translate, apply, and incorporate the basic definition of what a car is. In the hopes that the definition can be used by the computer to recognize a vehicle in front and to stop before hitting it. There are programs that actually "learn" and create knowledge but so far they have been able to basically build mock bridges after using teraflops of computational cycles.
To further use the autonomous car analogy, we are struggling with it on standard roads... 300 years ago, we didn't have roads, just terrain to navigate. And even the simplest "intelligence" could figure out a good enough route to get from point A to point B without dying or using up too much energy. And it could do it even in a place it has never been to before.
10 input "What is your name";A$
20 print "Hello ";a$
30 goto 20
Artificial intelligence
The only reason the term doesn't have any meaning is because everyone's definition of "intelligence" is different in the first place.
I disagree that that is the only reason. Another reason is that we don't agree on the term "artificial." In most cases I would say that all the intelligence, regardless of how that part is defined, came from the programmer and not the software.
We actually can't agree on the meaning of any part of the term!
Calling it an AI is misleading. The paper itself reports it as machine learning "Machine Learning of Threedimensional Right Ventricular Motion Enables Outcome Prediction in Pulmonary Hypertension"
Decode your health
I for one am not being dismissive of the useful of the algorithms at all when I say they're not more or less artificial nor more or less intelligent than other programs.
That's the problem, it is just regular software. Like calling software an "app" doesn't change anything, it is just a label. The problem with the AI label it is that none of the words are accurate descriptions of how it differs from other software. The problem isn't that programs are "merely" programs, but that software programs are not trees, ice cream, stars, or the feeling of spring. If you called apps "ice cream," people would complain, and it would have nothing to do with them looking down on ice cream.
It would be better if they just named the specific algorithm that it uses instead of the useless catch-all of "AI." It makes sense for academia to arbitrarily classify which things go into which classes and departments, but it doesn't make sense for programmers generally, or program users, to slavishly follow those categorizations where there are more descriptive and more accurate ones at hand.
Something might be an expert system, or be an application based on a neural network. But still, unless it is buggy all the intelligence in the system was engineered by the programmers; a "self-learning" algorithm only learns what it was engineered to learn, and what it learned accidentally due to bugs. And the engineer isn't even artificial!
This robot will cook your dinner!
NAO robot drives autonomously it's own car
Are Robot Composers The Future of Music? Nope, they're here now!
Book Reading Robot by Web-cam System
Robot building a brick wall
autoportrait; robot painting a portrait
Oh, did you mean one robot that does all these things? Just cram them all into one chassis.
Mod parent +1: relevant
Sent from my ASR33 using ASCII
I think many get hung up on the word computer, so let's try an alternative: the brain is a causal physical system. So output is a defined function of input and physical state, and physical state is a function of past inputs going way back to the moment of conception. Of course any causal system can be arbitrarily closely approximated by a computer (thanks Godel, Turing et al), so from a functional point of view a human is isomorphic to a computer.
The big question here is subjective experience: even if you can make a computer pass the most advanced Turing test does it have consciousness / subjective experience / insert other word meaning more or less the same thing? But that's a moot point as (a) I don't even know if you have this, (b) simple awareness has no effect on function and (c) subjectivity is by definition not definable in objective terms, so we will by definition never know.
You must be newbie...
It doesn't say in the article what algorithm was used.
The definition of artificial could be obvious... it is anything that is not natural. which is to say that it is not made or caused by mankind. Artificial intelligence would therefore be intelligence that *has* been made or caused by mankind. However, since "intelligence" is so poorly defined, the expression still cannot mean anything, despite half of the expression having an unambiguous meaning.
File under 'M' for 'Manic ranting'
We're perfectly capable of making systems we can't really understand, or which develop uses we hadn't thought of earlier. Lots of software does things that the developers didn't have in mind when they wrote it. We can't understand the internal values in a complex artificial neural net; all we know is how we arrived at them and what they appear to do.
Artificial neural nets are limited by their base complexity and by their inputs. Make one complex enough and give it enough input and you might well get general intelligence, although I doubt it's that simple.
"When you have eliminated the unacceptable, whatever is left, however improbable, must be the truthiness" - Holmes
... but all it responded was "Good Luck".
WTF is that supposed to mean?
Rule extraction from neural networks has traditionally been difficult, so whilst you can show the ability to classify, reasoning about why is not necessarily possible. Arguably one characteristic of consciousness is the ability to introspect but whether you could consider it a distinction between machine learning or AI is another matter. Maybe it is not even a useful distinction
Calling it *just* a program is not very accurate either because it is *just* a program that you can *train* to do many things.
However, "AI" is much shorter then "Machine Learning Program" so we are kinda stuck with it.
Now if they were calling it "artificial consciousnesses" I would be agreeing with you.
The only thing the 1950s needed to obtain recent results in convolutional neural networks, was the planar process of 1959 and a suitably accelerated coefficient of Moore's law. We can get there by applying the inverse Hackermann function.
Dividing 18 by 2 and shifting to a lower unit gives us a doubling time of nine weeks. Probably we're recognizing cats by 1967. Before the modern API was half fleshed out.
Seriously, have you looked at the sophistication of mathematics in the 1950s?
Ramanujan surprises again
The book is not even closed yet on the mathematics of the 1920s.
Nor can a baseball player understand all the complex quantum effects that are required for a baseball to bounce off the bat instead of just mixing together with it! That doesn't keep them from knowing how to hit the ball, though.
Waving your hands at complexity doesn't keep intent from being simple; an engineer intends to create a complex system, and didn't even intend it to be one that required calculation of all the details. It doesn't really mean they didn't understand what they were doing, it just means that humans have limits and the concept of knowledge can be taken in a way that exceeds human capability. But that is not any more or less true when building a neural net than when hitting a baseball; both are rather simple things to the person doing them, and they only need to understands that steps taken, not the whole Universe.