Stop Calling Everything AI, Machine-Learning Pioneer Says
spectrum.ieee.org
spectrum.ieee.org
I won't be surprised when AI toothbrushes come out.
if brush_time < 2.minutes
return "Brush longer"
else if brush_time > 5.minutes
return "Brush less"
endhttps://www.theverge.com/circuitbreaker/2019/10/25/20932250/...
In French: Intelligencive Artificielle? ÉmAI brillant? Recarie Neural Networks? MolAIre.
Exploding Stove Polish was apparently a problem. Gotta wonder what chemical it was causing the problem.
> LOS ANGELES, Feb. 24 Mrs. Sylvia Olds wife of George Olds, of Gardena was probably fatally burned by the explosion of a bottle of stove polish today.
"Of course it's organic, it comes from a plant!"
"That's not what organic means..."
"Guh, wat?"
Our toothpaste contains 0% raw sewage!
(Haven't googled it yet, I swear!)
Which is to say, sure, today "AI" is just a marketing term but that's 'cause they don't have it, not because real intelligence isn't something that would be very useful in many, seemingly trivial places.
A hint appears partway through: "computers will not be able to match humans in their ability to reason abstractly about real-world situations". Does human intelligence distinguish itself by its "abstractness" and by its application to "the real world"? There's also "the systems do not form the kinds of semantic representations and inferences that humans are capable of". Seems like a promising direction for some definition of intelligence.
For my money, we will never consider any machine intelligent as long as we mass produce it. The only way we'll accept machines as intelligent is if, as the singularity theorists say, the machines build themselves. Then we aren't really mass producing them, we're just kicking off a process that we don't totally understand, a bit like gestation.
We won't consider machines intelligent until they have and demonstrate the capacity to consider us not-intelligent.
It's an open question of how long we will consider anything after that point.
I don't think it needs to go that far. Birds that can solve simple puzzles are obviously intelligent, so are cats who can open doors, etc. I think the human perception if intelligence is hugely based on the notion of, "is this thing self-aware?" or "is there someone in there?", "does it have a mind of its own?"
I would say we'll consider machines intelligent when they're able to have some amount of independence. Right now, they can't solve problems they're not specifically designed/engineered to solve. We don't have any kind of general-purpose artificial intelligence. You might try to argue that reinforcement learning is that, but no, RL doesn't really work outside of a sandbox with an artificial reward function, and it's very sensitive to hyper-parameters.
If you could have a chat bot that can pass the Turing Test with a high degree of success, I think most people would consider that intelligent. Same thing for a robot that can do your laundry, wash your dishes and make some mac and cheese without days of training and without fucking up more than a typical human child would.
I can't see how this claim is even slightly illuminating aside from it's pure "quip-ness". A robot playing, say, the game of go, might spit-out "you are not intelligent" whenever it wins. Is that "the capacity to consider us not-intelligent"?
You could even set-up a micro-world where virtual entities interact, evolve and rate their opponent. It wouldn't be hard for these entities to compete with humans in this arena, to do better than humans and so to give human poor poor evaluations but the entities to be pretty simplistic and far from what humans consider intelligent and far from fit outside the micro-world (you could do it with adaptive game-theory solvers, say).
If someone could do that, it would be a significant stride towards producing an intelligent entity.
As far as "AI research" we're somewhat stuck in the position that humans "can't define intelligence but know it when they see it". But if we abandon this intuition based approach, it seems like we wind-up with the argument, "this toaster is intelligent 'cause you can't prove it's not".
>One common insightful definition of AI: Artificial Intelligence is everything a computer can't do yet. As soon as it can, we call it obvious.
If the term is used like that it does make sense to me.
https://seths.blog/2015/04/the-noise-in-our-head-and-artific...
Like most clever things marketers do, that quote is borrowed nearly verbatim from a much older quote without reference.
"AI is whatever hasn't been done yet."
From Douglas Hofstadter quoting Larry Tesler in 1980
Anyone who has studied AI and its history at all knows that similar quotes to Godin have been going on since the 70s an even that quote from Tesler.
There's an entire wikipedia article on the AI effect[0] if you want to dive deeper.
Also, "Algorithm" might be a little too narrow, "computer program" fits the colloquial use of AI a little better. :)
Artificial Intelligence – The Revolution Hasn’t Happened Yet (2018) - https://news.ycombinator.com/item?id=25530178 - Dec 2020 (120 comments)
The AI Revolution Hasn’t Happened Yet - https://news.ycombinator.com/item?id=16873778 - April 2018 (161 comments)
Others?
It's odd how under-appreciated this is. There is no "intelligence" here as we know it, not even a hint. There are algorithms operating on data for very prescribed use cases. When you break it down its pretty primitive stuff really. Clever sure, but don't think we're building a mind here, that language just confuses things. So much of the public conversation around this stuff is polluted with this sort of sci fi nonsense claiming all sorts of properties for AI and implying some sort of singularity is around the corner. Pure fantasy.
I agree with your general point, but if you break it down, our brains a nothing "intelligent" either. It's just a network of neurons searching for patterns in the sensor input. Sure, it can reconfigure itself, but from a really abstract perspective it's nothing more than a ultra-scaled neural network as we currently use them. So this really boils down to the definition of intelligence, which is by no means easy.
You state this as if it is an undisputed/obvious fact, but it isn't.
I tend to think that because computers are the current peak of civilisation / engineering, and in some ways, "alive", we think we might be like a computer. In the Victorian machine age they talked of humans as machines. Maybe some, hitherto unthinkable invention is around the corner, and we will say humans / brains are like that too.
One obvious feature of intelligence is that it can abstract from particulars. We form and reason about universal concepts. I know the concept of "squarishness" which means something definite without being some specific square (of which there is an infinite number). I can imagine a square, but those are always particular squares, whereas the concept is universal. But material things are only ever particular. Can you show me a material thing that is "squarishness"? No. Computers are physical objects, right? So how can the universal "squarishness" exist in a physical object without being a particular square?
Even here we often fool our selves when we fails to recognize that computer programs, like the text in a book, need humans to grant them their identity and make sense of them because their meaning and identity are assigned by the observer and not intrinsic to them like mind-independent things. A "Square" class, defined as record with two fields corresponding to width and length, is not the concept of a square. Instances of that class are not squares. Even "class" and "instance" are mental models which do not exist "out there" like real entities or phenomena. What we call computers just have affordances that allow us to configure them to correspond and cooperate with the models in our minds. But strictly speaking, computers don't really have an objective reality and neither does the computation within them. The machine is real in the sense that we have arranged a bunch of stuff to be a certain causal ensemble, but this causal ensemble is not objectively a computer nor is it computing. Rather, we are using it to compute. But if you erased all human beings from the planet, and left a bunch of computers running, objectively, they would only be arrays of things with magnetic and electrical components changing physical state. You need someone to interpret what computers do to make them useful. Just like three beads on an abacus doesn't mean "3" and pushing a bead toward those three doesn't mean "adding 1". You need a person to assign those meanings to it. Computers don't magically break free of that reality.
I remember when 'GANs' were 'marketed' as 'deepfakes' to the public and watched it enter the english lexicon.
You're lucky it didn't end up being "<Company> AI Blockchain."
[0] git, that is. Yes, git is technically a blockchain.
Basically, it seems that whenever someone mechanizes something that previous only a thinking person could do, then its viewed as AI.
Often the AI or machine learning that is being sold to their customers (or if it’s a startup, to their investors) is in fact a team or a group of teams creating static rules. If it is image recognition the they will often have a large team of manual reviewers. Yes there maybe some AI or machine learning models that are assisting in the decision making, but they are usually much less effective than people realize.
Today whenever I hear AI or machine learning, my default is to assume it is marketing speak.
I don’t doubt there are models, I just doubt their effectiveness. But saying “we use AI to do X”, sounds much better than saying, “we have a team of experts who help us do X really well”
With that said I have worked and continue to work with some amazing data scientist who really are pushing the limits of what can be done with machine learning.
Sure, the risk is that they can't make it work in the medium term. But without knowing much about the complexity of the orders / domain, that doesn't sound like a huge risk, given the current SoTA
I am just about to hit my 40 year anniversary for getting paid to be doing "AI" related work, with a lenient definition of AI.
On the other hand, Artificial General Intelligence is a much stronger term, and is something that I don't expect to see in my lifetime.
But why that? There isn't a principled reason for singling those out. It's just a matter of convention that we call those AI.
Pretend you never heard of AI before and you came across these algorithms. No amount of analysis or poking around would lead you to think "oh, man, this is artificial intelligence!". No, you'd see a bunch of statistical techniques that do X, Y, and Z, none of which would jump out at you as "wow, this is AI". All that computers give us is the ability to run tedious computations over larger data sets that would be impractical for human beings to perform.
"AI" is what we read into these things. There really is no such thing as such.
I imagine if we really wanted to build our final invention[1] then we would have to be under some existential pressure to do so. In other words: we would build a 'run away' AI if it could potentially save humanity. Also worth reading this: https://en.wikipedia.org/wiki/Ethics_of_artificial_intellige...
Btw, reminds me of the old joke that goes something like this: AI for marketing, ML for recruiting, Regression for design, multiplication for implementation.
The problem is the term never had a technical definition. It was always just a hand waving marketing phrase for “clever algorithms”.
One of the funniest cases was a start up that claimed to be using AI to process their customers' requests but was actually just farming out the work to contractors who were doing everything by hand.
The technical people involved have since moved on to using AGI as the term for "actual AI" for years now.
How much more am I as a human really than the machine learning algorithms are?
Of course, computers might be able to imitate some things and far exceed some things that our brains do. That's different though.
Personally, I go to the Wikipedia page for any particular term and follow the crowd wisdom, plus a grain of salt.
¯\_(ツ)_/¯
Actually, you raise a genuine issue. Terminology isn't clear, and that makes discussion difficult.
I believe that the notion of AI was predictive, and emergent, and retro-fitted. It emerged around the 1950s (arguably), gained some degree of formalisation, meanwhile the goalposts shifted and the field diversified, and now the term is applied haphazardly. I think it's now the responsibility of the user of the phrase to define their terms... which probably indicates that it's no longer fit for purpose.
To say nothing of the fact there is a risk of another AI winter when NN/ML continues to not produce (the promised) results.
We're at least 30 years away from that though, if not 100. Immortality tech, may be easier to accomplish.
AI is not a precise term. If you can make a product that feels intelligent to the user, why does the implementation matter?
pattern recognition? matrix multiplication? fancy statistics? or maybe enhanced statistics?
gah, might as well just say 'algorithm' and just be done with it.
And then I clicked on the article.
HAHAHA.
I learned Bayesian inference models from him.
Every time I see a new company preaching endlessly about 'AI' to address a non issue at this point is straight-up begging to VCs to fund their $0 revenue so-called 'tech startup'.
Doesn't have to be deep learning though: https://www.newyorker.com/tech/annals-of-technology/the-past...
I had people respond to me in a gdpr thread saying that they never accept cookies
example : https://news.ycombinator.com/item?id=26360345
What do such ppl do here? just not read the article?
Of course, sometimes you direly need some info and can't skip it, but these situations are extremely rare.
from their privacy policy https://www.ieee.org/security-privacy.html
> What information do we collect?
IEEE collects the following personal data in line with the use purposes explained in a subsequent section:
- Your name and contact details - Date of birth - Online profile data/usage - Emergency contact information - Social media profile information - Copies of identification documents - Education and professional information - Communication information including IEEE Online Support and Contact Center communications - Purchasing and payment information - Registration and participation in IEEE events and activities - Subscription preferences - Information about the device(s) you use - Information about service usage - Cookies - Authentication data - Location information - Author and peer review information - Other information you upload or provide to us
How do we use your information?
To engage with third parties. IEEE may share your personal data with third parties in connection with services that these individuals or entities perform for or with IEEE.
In this case no, as I did not need to click "accept" in order to read articles. With an intrusive pop up, I most likely would have. Overall, though, I'd probably need to make an exception for IEEE anyway as it hosts a lot of papers for my field of study.
To be fair, though, I'm not a hardcore data protection activists. It might be possible to start a complaint or check whether the website does tracking without clicking "I accept" (and I'm pretty sure most do), but unfortunately my life is only so long and the effect is probably rather small. I highly dislike that this tactic works for them, but I'm only willing to sacrifice so much. Also, I still run no script and ad blockers, so the effective tracking is hopefully limited anyway.
The MiniHack app on iOS does this as well.
So in the mid 1900s a calculator that is able to do arithmetic was considered "AI", and in the late 1900s chess playing machines were considered "AI". Today those things are not considered "AI".
What is considered "AI" today will not be considered "AI" a decade or two from now.
More: https://ai100.stanford.edu/2016-report/section-i-what-artifi...
Is this true? The citation you link doesn't seem to clearly state that.
I see the linked page says:
>A simple electronic calculator performs calculations much faster than the human brain, and almost never makes a mistake.[4] Is a calculator intelligent?
But I don't see where it says anyone in the mid 1900s thought a calculator was AI.
I am no historian, but I can think of numerous examples of AI in science fiction throughout the 1900s and perhaps earlier. Metropolis, Rossum's Universal Robots, 2001: A Space Odyssey. Heck, one could probably even draw the line all the way back to Frankenstein.
Maybe the lower bound for what gets called AI to sell stuff has changed over time, but it seems like there's been a pretty clear and consistent (if poorly defined) goal of human-like intelligence.
You don't think performing arithmetic and playing chess could ever have been thought of as requiring human-like intelligence?
See https://www.leanix.net/en/blog/what-is-the-difference-betwee...
Excerpts:
Artificial intelligence has been around for ages, with it even being used in the Greek mythology to express human like machines which copied man’s behaviour. At the beginning, even early computers were considered logical machines as they were able to reproduce “intelligent” capabilities such a arithmetics and memory… Engineers of that era saw it as an attempt to create mechanical brains.
Nowadays, something as “trivial” as a calculator function, does not go hand in hand with our current concept of Artificial Intelligence.