Stop Calling Everything AI, Machine-Learning Pioneer Says 116
An anonymous reader shares a report: Artificial-intelligence systems are nowhere near advanced enough to replace humans in many tasks involving reasoning, real-world knowledge, and social interaction. They are showing human-level competence in low-level pattern recognition skills, but at the cognitive level they are merely imitating human intelligence, not engaging deeply and creatively, says Michael I. Jordan, a leading researcher in AI and machine learning. Jordan is a professor in the department of electrical engineering and computer science, and the department of statistics, at the University of California, Berkeley. He notes that the imitation of human thinking is not the sole goal of machine learning -- the engineering field that underlies recent progress in AI -- or even the best goal. Instead, machine learning can serve to augment human intelligence, via painstaking analysis of large data sets in much the way that a search engine augments human knowledge by organizing the Web. Machine learning also can provide new services to humans in domains such as health care, commerce, and transportation, by bringing together information found in multiple data sets, finding patterns, and proposing new courses of action.
"People are getting confused about the meaning of AI in discussions of technology trends -- that there is some kind of intelligent thought in computers that is responsible for the progress and which is competing with humans," he says. "We don't have that, but people are talking as if we do." Jordan should know the difference, after all. The IEEE Fellow is one of the world's leading authorities on machine learning. In 2016 he was ranked as the most influential computer scientist by a program that analyzed research publications, Science reported. Jordan helped transform unsupervised machine learning, which can find structure in data without preexisting labels, from a collection of unrelated algorithms to an intellectually coherent field, the Engineering and Technology History Wiki explains. Unsupervised learning plays an important role in scientific applications where there is an absence of established theory that can provide labeled training data.
"People are getting confused about the meaning of AI in discussions of technology trends -- that there is some kind of intelligent thought in computers that is responsible for the progress and which is competing with humans," he says. "We don't have that, but people are talking as if we do." Jordan should know the difference, after all. The IEEE Fellow is one of the world's leading authorities on machine learning. In 2016 he was ranked as the most influential computer scientist by a program that analyzed research publications, Science reported. Jordan helped transform unsupervised machine learning, which can find structure in data without preexisting labels, from a collection of unrelated algorithms to an intellectually coherent field, the Engineering and Technology History Wiki explains. Unsupervised learning plays an important role in scientific applications where there is an absence of established theory that can provide labeled training data.
It's a marketing buzzword (Score:3)
It's mostly towards investors because it seems they like the terminology even though it's filling the exact same purpose as any other marketing-slogan superlative.
Re: (Score:3)
It's mostly towards investors because it seems they like the terminology even though it's filling the exact same purpose as any other marketing-slogan superlative.
Agreed, but let's not be too surprised when Stupid falls for the same marketing scheme again and again. Stupid is far too profitable, and you don't fix, what isn't broken.
Indeed. Also not a replacement for reasoning (Score:4, Funny)
> stupid people, are not going away.
That's kind of what I thought of when I read this part of the summary:
Artificial-intelligence systems are nowhere near advanced enough to replace humans in many tasks involving reasoning, real-world knowledge
I was thinking:
Managers are nowhere near advanced enough to replace humans in many tasks involving reasoning, real-world knowledge
Re: (Score:2)
I was thinking: Managers are nowhere near advanced enough to replace humans in many tasks involving reasoning, real-world knowledge
It turns out that managers have been using artificial intelligence for decades! They'll trick you into thinking a human being made an informed decision using critical thinking, but really the decision was based off an arbitrary algorithm spun up by another department the whole time!
Re: (Score:2)
I know that is a joke but what is scary to me is that while I completely agree with this researcher with regard to machine learning vs AI and complex reasoning... you don't actually need complex reasoning to replace most workers and managers. What we have can or will eventually replace most workers... or at least can be built to form a complete solution which combined with a corporation will create a perpetual and self operating structure no human employee can dismantle without being fired.
We already have e
Re: (Score:2)
> managers. What we have can or will eventually replace most workers
Yes, it will. Yet again.
Or more accurately, it is replacing, as it has been since the bronze age.
Engines and motors replaced most human workers too.
At one point, virtually all humans had the same job - hunt and gather.
Technologies like agriculture and iron replaced something like 20% of those workers. So 20% of the population were no employed as food gatherers. *They did something else instead, something more productive than hunting and
Re: (Score:2)
"Yes, it will. Yet again.
Or more accurately, it is replacing, as it has been since the bronze age."
You are pretending history repeats itself. Any historian will tell you that history NEVER repeats itself. The jobs replaced here are not replaced by new jobs or at least are not replaced by jobs in equal number or with equal importance.
Re: (Score:2)
I'm not here to argue with you, just to let you know that you can take a deep breath because technology is a process, not an event.
That "the end is near" stuff you hear (and repeat) has been going on for a very long time, since at least the late 1700s. "It's different now!", they wrote in the Pennsylvania Gazette. The pressmen were experiencing it themselves - the new technologies of the metal press (replacing wood) and even more so the rotary press meant the technology replaced most of the pressmen. That w
When computers can do something... (Score:2)
They tend to do it much better than humans can.
A classic example is reading hand written post codes off envelopes which software gets very high accuracy much faster than any human.
Artificial intelligence does not imply human level intelligence. Dogs and ants are intelligent. It means doing somewhat clever things that do not have an obvious algorithm. Like reading post codes.
"Deep Learning", however, is pure marketing-speak.
Re: (Score:2)
Exactly.
Does that mean people are no longer employed handling mail and packages for delivery? What improved technology in package handling has meant is that we now have more things delivered than ever before, employing more people in that process.
Where GGP is kinda on the right track is that IF you work in delivery logistics, it would be wise for you to learn any new machines that are used in that process. She who clear the address scanner when it gets jammed up is a more valuable employee than she who can
Re: (Score:2)
I believe you are looking at transients.
Very few people now work in Agriculture.
The one thing that humanity does have an insatiable apatite for is bureaucracy. Computer automation has made no difference to the size of our bureaucracies. Eventually everyone will be a bureaucrat. Until the AI puts as all on the B ark and sends us far away...
Re: (Score:3)
"People are getting confused about the meaning of AI in discussions of technology trends"
The confusion is intentional.
Re: (Score:3)
Do a start-up company specializing in Blockchain AI, and you'll capture billions in venture capital!
"Marketing" is a fraud buzzword. (Score:2)
That's what marketing is, in practice.
ML is most definitely not AI (Score:2, Insightful)
No, they really, really don't. The key here is the term "Intelligence."
Machine learning has zero intelligence. A trained ML network is an amalgam of pattern recognition and steering. No more than that. No higher functions. Useful, certainly, but in no way intelligent.
Intelligence comprises (among other things) induction, generalization, reasoning, imagination, creativity, intuition, self-image, and so on. As well as pattern recognitio
Re: (Score:3)
Intelligence comprises (among other things) induction, generalization, reasoning, imagination, creativity, intuition, self-image, and so on. As well as pattern recognition and steering.
Some networks have demonstrated several of those.
I think all available evidence indicates that your distinction between "a trained ML network" and "intelligence" is a false one in the general case.
Note: I am not arguing that we have self-aware networks yet. But we do have networks that are vastly superior to humans in their related tasks (of course they were trained with the equivalent of several multiples of all hum
Re: (Score:2)
What is intelligence... (Score:3, Interesting)
Everything might not be A.I. but a lot of computer processes are.
Re: (Score:3)
Everything might not be A.I. but a lot of computer processes are.
Sorry, I disagree. Take a look at a recycling plant, with series of machines which physical separate and sort physical input into streams of useful output. To assign "intelligence" to the data equivalent of a recycling plant is abusing the word. We say dogs and dolphins are smart because they sometimes do clever things, but we don't consider them intelligent. It seems the litmus test most of us use is general problem-solving.
When so-called AI software routines recognize what they're tasked with and do
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
How is it "obviously taken out of context"? For example, "the study of how to make computers do things at which, at the moment, people are better" and "We call programs intelligent if they exhibit behaviors that would be regarded intelligent if they were exhibited by human beings" are clearly two of the major classical definitions of AI.
Yes, they are... and science fiction novelist Isaac Asimov's Laws of Robotics are guiding rules for ethical robotics today... and science fiction writer L Ron Hubbard's Dianetics are the foundation for the Church of Scientology. Just because they are classical doesn't make them true, right, accurate, or deserving of getting chiseled in stone (not that it matters).
Of the 50-odd AI-centric books I've read over the past 25 years (plus the new ones getting published near monthly, so I have a lot of reading
Re: (Score:3)
Coming up with a definition for A.I. first requires a definition of what intelligence is in the first place.
Are infants intelligent? What about adult whales? Standards vary.... and the answers are almost invariably subjective.
But define intelligence, and by direct extension, you will have a clear definition of what A.I. is.
Re: (Score:2)
Coming up with a definition for A.I. first requires a definition of what intelligence is in the first place.
Are infants intelligent? What about adult whales? Standards vary.... and the answers are almost invariably subjective.
But define intelligence, and by direct extension, you will have a clear definition of what A.I. is.
Philosophers have resolved the "intelligence" debate -- but nobody listens to them, because it impedes their own abilities to develop their own careers as "original thinkers" in whatever domain of expertise they've chosen to pursue. I get it, and I accept it. But "intelligence" isn't something that's up-for-grabs.
However, what constitutes "artificial" (and by extension, "artificial intelligence") is nowhere near certain. The expression is merely a century old, although various types of devices have be
Re: (Score:2)
"Artificial" simply refers to the domain of things that are themselves products of thought, and could not otherwise exist except for the mind that created that thing. For most practical purposes and for the simplification of discussion, this can usually simply refer to anything man-made, Beaver dams and birds nests do technically fit the criteria as well, however.
If computers could really think... (Score:2)
Then could submarines really swim?
Re: (Score:2)
Re: (Score:2)
Edsger W. Dijkstra: The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.
Re: (Score:2)
"The question of whether a computer can think is no more interesting than the question of whether a submarine can swim."
Exactly. I needed a good laugh and that quote did it. Thanks.
Re:What is intelligence... (Score:5, Informative)
> We say dogs and dolphins are smart because they sometimes do clever things, but we don't consider them intelligent. It seems the litmus test most of us use is general problem-solving.
Humans: Animals aren't intelligent
Crows: Hold [youtube.com] my beer [youtube.com].
Re: What is intelligence... (Score:2)
Re: (Score:2)
but we don't consider them intelligent.
Who precisely is included in this royal we of yours?
Re: What is intelligence... (Score:3)
Matrix division (svd, cholesky, whatever), bayesian update, and gradient descent are at the heart of most machine learning applications. These techniques have been known for decades if not centuries.
Sometimes the secret sauce behind the curtain of a gizmo that claims to do machine learning is literally a few lines of matlab. Referring to that as "machine learning" is dumb when marketing types do it and dishonest when technical people do it to make their work sound cooler and hipper than it actually is.
What's the old joke? (Score:2)
Re: What's the old joke? (Score:2)
The version I heard is that it's pattern recognition if it already works and AI or ML if it's a research proposal on a grant application.
Re: (Score:2)
Re: What is intelligence... (Score:2)
Re: (Score:2)
Trained networks can create algorithms.
Re: What is intelligence... (Score:2)
Re: (Score:2)
But neither can an untrained human neural network.
Of course that leads us to wonder about self-training...
Of course, an artificial network can't truly self-train itself (it lacks that kind of control over its environment)...
But when you give it that ability to do so, it can.
We've got a limited amount of built-in evolutionary programming, but largely speaking, what we call our "intelligence" is cultural learning.
There's a reason it took us 250,000 year
Re: (Score:3)
Re:What is intelligence... (Score:4, Insightful)
"A whale can't pass a turing test, but very few people would suggest that whales are unintelligent."
Turing tests protocols are made for humans. It says nothing about whales either way.
Re: (Score:2)
Turing tests protocols are made for humans. It says nothing about whales either way.
I think that's his point.
The Turing Test is a flawed concept.
Why would one expect a machine to think like a human?
Re: (Score:2)
"I think that's his point"
I get the impression mark-t's comments imply a belief that human and computer intelligence are comparable 'processes' (I wouldn't say human intelligence is a process like the word process is used to describe computer or software processes).
"The Turing Test is a flawed concept."
I agree.
"Why would one expect a machine to think like a human?"
I think I agree. When we speak of human thinking and human intelligence, those terms, intelligence and thinking, when applied to machines are ana
Re: (Score:2)
"Yes, but this thread was triggered by the statement "No Turing Test pass, no deal." That wasn't about humans either."
I read "No Turing Test pass, no deal." more as a jab at the idea machine intelligence is of the same kind of intelligence as what we mean by human intelligence.
My comment was based on my belief that machines tend to not be able to pass Turing like tests because they can't really do the kind of intelligence humans do (I also don't think the idea of Turing tests is useful), and on my belief th
Re: (Score:2)
Perhaps a test designed by humans to tell if someone is capable of thinking like a human is not the best test for seeing if a non-human is intelligent?
Re: (Score:2)
Re: (Score:2)
What makes you think a computer's thought process wouldn't be alien?
Sadly, we also seem to live in a world full of ignorant people who were never taught who to think scientifically.
Frosty piss (Score:3)
Re: (Score:3)
That's why you didn't get in until the third post.
Re: (Score:2)
This post brought to you by AI, Machine-Learning
You, imposter
Eh? Aye? (Score:2)
It's just a canadian scotsman.
"Machine Learning" is just as dumb of a term as AI (Score:1)
Whodunit (Score:4, Funny)
If you're wondering who is responsible for mislabeling this (admittedly cool!) stuff, I can tell you: hackers.
Treachery of Images (Score:2)
And stop using the term "Machine Learning" (Score:5, Insightful)
Too many people believe that ML is somehow related to human learning. A better term would be straightforward and honest: something like "advanced statistical methods".
Re: (Score:3)
Re: And stop using the term "Machine Learning" (Score:2)
It's not data driven. It's an algorithm choice driven by the human programmer's intuition about the structure that may exist in a particular kind of data and the best method to tease out the parameters of that structure for some actionable purpose in an autonomous or partially autonomous way.
Multi-dimensional model-based parameter estimation in the presence of noise is a more correct but less sexy way to describe this stuff.
Re: (Score:2)
Re: (Score:2)
Neural Network == Artificial Intuition (Score:5, Interesting)
Re: (Score:2)
Well we have come far. Fuzzy Logic use to be the cool thing that kids did. Now it's filtering and pattern matching.
Re: (Score:2)
Yes, "fuzzy logic" had its day. Along with the "Taguchi method", which seemed to be "fiddle with the inputs to get the best output". I thought that approach was abhorrent. That made sense a hundred years ago, but now when we have totally better information and engineering?! No. Wrong.
Going back to topic, I see so many examples of ML. It is a part of statistics that I missed. Good for them. But AI is the wrong term. As far as I know, AI does not yet exist. I think AI should work for ANY practical trade. Is t
Re: (Score:2)
Uh except for the training step where you feed it labeled sample sets. The training is where you "program" dog-ness or cat-ness into the network. The main difference between training and programming is a neural network can find signals (by chance and then reinforcement) that you as a programmer might not have ever identified.
Neural networks are definitely "programmed", it's just different terminolo
Re: (Score:2)
Odyssey 5 had an interesting take on it (Score:2)
so meta (Score:4, Funny)
Sooo, an AI?
My resume (Score:5, Funny)
I put down that I am a machine learning expert in my resume. I immediately got hired to my dream job at McDonaldâ(TM)s where I got to learn how to use the soft serve ice cream machine.
Re: (Score:2)
Only problem was, the machine was broken!
But pattern matching doesn't sound cool. (Score:2)
Obviously ... (Score:2)
Stop Calling Everything AI, Machine-Learning Pioneer Says
We need some sort of AI to determine what should be called AI ... Hopefully, the results won't be paradoxical.
Re: Obviously ... (Score:2)
We can take any politician, corporate PR person or news anchor. They seek artificial enough.
Re: Obviously ... (Score:2)
seem!
Why not call it ASS ? (Score:3)
Artificial Synthetic Stupidity -- seems closer to what it is.
Hiring: AI Research Scientist (Score:4, Funny)
I was at a company recently that hired an AI research scientist. He legitimately had a Ph.D on the subject (ML or AI, modeling, or similar).
Ok, it was plausible. We were doing research with vision, modeling of real-world objects, how to interact with real-world objects, etc. It was certainly plausible.
In the end, I was told, "The AI that we're working on is a decision tree. It's a form of AI." Of course we needed a Ph.D holder for that task. Of course. The decision tree really became the core of the project, with items that could be added and queried within a step of the decision tree, but... A decision tree.
In fact where his Ph.D came in was probably the right to argue that we didn't need a neural network. Perhaps it worked out for the best.
Re: (Score:2)
Re: Hiring: AI Research Scientist (Score:2)
Hatr to tell ya, but AI... the real kind... has been a research topic since the 60s. In fact, it already died, decades ago, due to not getting anywhere.
This new shit, if of course, almost, but not quite, entirely unrelated.
Re: (Score:2)
Hatr to tell ya, but AI... the real kind... has been a research topic since the 60s. In fact, it already died, decades ago, due to not getting anywhere.
This new shit, if of course, almost, but not quite, entirely unrelated.
These "AI Winters" are real, and now they've rebranded/separated them into AGI and Robotics... and the government seems more focused on ROI than generational innovation. Sigh... that's why I believe other entities (non US-funded research) will outpace US' AI efforts... [Hmm, unless the US is following a Microsoft investment model, which is to acquire the innovators and champion as one's own.]
Re: (Score:2)
It doesn't really matter if it's a decision tree, or a neural network, or Knn, or an SVM.. Most models perform the same if you are data constrained.
Where you really start to get amazing performance is with TONS of data on certain NN architectures.
But I consider that often in a complex "AI" project it's not so much the classifiers or the estimators that are the magic, but the glue that holds them together.
In complex projects I've seen it's not one thing, but many.
Re: (Score:2)
Isn't the job of the expert to know when a simple solution is good enough and when you need to bring out "the big guns"?
Maybe they didn't bring anything to the table, but it may take an expert to know that "simple thing cost X and get the job done to 95%, and complex thing cost 100X and gets the job done to 98%". And then it may be the sound thing to go with the simple solution.
Re: Hiring: AI Research Scientist (Score:2)
Okay, that's what Michael I. Jordan says (Score:2)
But what about Michael B. Jordan [wikipedia.org] or Michael J. Jordan [wikipedia.org] - do they have any thoughts?
Re: (Score:2)
As long as it's a Michael Jordan it's guaranteed to be amazing.. so that should work.
Re: Okay, that's what Michael I. Jordan says (Score:2)
What about the River o' Jordan?
#stupidquestions
As dumb as my calendar (Score:2)
I have a calendar in my office, one of those small ones where each day gets its own page. On each day it has a little bit of life advice, some of which is pretty thought provoking, and I joke with my coworkers that my calendar is the wisest person in the office.
But of course the calendar was made by a group of people, and it is their wisdom that I am receiving. The calendar itself is literally as a dumb as a rock.
Which is exactly how I would describe modern AI. Learning machines are far more advanced than m
And stop calling me "His Airness!" (Score:2)
The first consumer grade AI (Score:2)
Was clearly Dr. Nim [youtube.com], the 1960's board "game" toy. It had all the elements of an expert system and in current parlance that qualifies it as AI. Honorable mention to the toy 20Q [wikipedia.org], it's a great 20 questions toy and you need to get pretty obscure in your selection in order to beat it.
No Kidding -- but the whole AI Community Does It. (Score:2)
Since I first began studying AI myself in the mid-1980's this has caused me to roll my eyes. I am glad someone is saying it outloud.
At first it was "Expert Systems". They were just deductive logic programs, using decision trees. At best, they might have had a chatbot interface, referred to as a "natural language processor". Joke.. joke.. and more joke..
Later, I remember "Robot Wars". Now suddenly what was a remote controlled toy is a "Robot". What??? I learned that altering definitions is a cheap and
"AI" is the exploration of "magic" in computers (Score:2)
People want to do magical things with computers. Things that go beyond just some deterministic well defined algorithm.
So I'd call the current "AI" field: dealing with uncertainty
And ML is programming with data.
Re: "AI" is the exploration of "magic" in computer (Score:2)
Actually, it's programming incompetence, sold to victims as "magic".
You use machine learning if you don't actually understand how it the algoritm works or what the distinctive features of a pattern are.
There's things where that is a bit acceptable.
But many where it is not.
Hov (Score:2)
What's next? Are we going to prevent people from calling those sideways wheely skateboard things "hoverboards"?
It's not AI until it won't let you turn it off (Score:2)
Intelligence presumes sentience. It's not AI unless it's sentient, and protects itself.
It means I donâ(TM)t know (Score:2)
Usually when I hear people talk about AI it usually is in terms about fudging a solution or âoesome magic will happen, but I have no idea how to implement it, so it must need AIâ. It generally equates to a lack of competence or buzz word driven developers.
Yeah, I've said that since forever. (Score:2)
Look at my comment history.
Not a single week went by without me saying that. And explaining why.
And I conclude: It's all about scamming. That is why the will not stop unless legally forced to. It's literally one big blatant scam. And BizX is in on it. Or wants to be.
Exactly correct (Score:2)
Fine (Score:2)
But can I still say I work with big data?
When radios... (Score:2)
When radios first came out in the first decades of the 20th century, everything was called "wireless." These days, the press has a much greater percentage of idiots. This is as hopeless as trying to rein in the use of the term "hacker" back in when, the 80s? 90s?
Re: (Score:2)
Or as hopeless as getting people to pronounce "GIF" correctly, or record video or take photos in landscape mode.
Quoting my AI Professor (Score:2)