However machine learning is nowhere near what we consider as an AI, equivalent of our intelligence.
You can compare machine learning with training a hamster to jump on a command. If you will repeat learning process a lot of time hamster will jump. But change anything in the environment and he won't.
Machine learning is just a hamster that is trained thousands of times.
It can do one thing, sometimes quite good, but still it is as intelligent as a hamster.
Machine learning does not aim to become an intelligence. It is just a well trained hamster. Nothing more.
It is just a fuzzy algorytm.
That is why it is so easy to fool the algorytm. I hesitate to call machine learning any kind of AI just for the reason it generates such confusion.
If it comes to developing real AI we are nowhere near currently. However we enjoy machine based models that are easier to brute force train with processing power he have today
At a mile high conceptual level, AI is nothing but a program created by a computer based on the data it is provided
Which is why it is extremely easy to fool using techniques that it is not trained to handle, today, but might be able to handle tomorrow. It is a race...
I found a quote from Geoff Hinton where he talked about this last year.
From [1]: “I can take an image and a tiny bit of noise and CNNs will recognize it as something completely different and I can hardly see that it’s changed. That seems really bizarre and I take that as evidence that CNNs are actually using very different information from us to recognize images,” Hinton said in his keynote speech at the AAAI Conference.
“It’s not that it’s wrong, they’re just doing it in a very different way, and their very different way has some differences in how it generalizes,” Hinton says.
[1] https://bdtechtalks.com/2020/03/02/geoffrey-hinton-convnets-...
This. In a nutshell every sort of algorithm we call "AI" today is reductive pattern matcher. This limitation isn't due to computational capacity or even, IMO, algorithm design, but due to our collective lack of understanding of how intelligence itself works. We'll get there eventually, but not for a long while.
This a common refrain but fairly obviously untrue. It assumes there's some secret sauce in human brains that makes us "intelligent" whereas AI is "just a machine".
It's pretty clear that human brains are just programs. Extraordinarily complicated highly optimised programs, sure. But nobody has even found a shred of evidence that there's anything fundamentally different to programs in them.
Thinking otherwise is along the same lines as thinking that animals don't have feelings.
Every time there's an advance in AI the "it's not really intelligent" goalpost shifts. Clearly intelligence is a continuum.
My general thought is, that since our brain is made of matter, it can be dissected and understood and eventually copied. Except, we are reverse engineering millions of years of evolution, which is exceedingly hard! We have had access to the information about all the proteins that make our brain cells for almost two decades, and still, their function has to be teased out in year long experiments. Not to speak of understanding the workings of the Homo Sapiens brain as a whole.
It is also an analog 'computer', the performance of which we have no perceivable chance to match, at least in the near and most likely also distant future.
Artificial neural networks don't have "memory addresses" to store data in the same way that a conventional program does either. But they can still store data. GPT-3 knows the first page of Harry Potter, but if you feel through its weights you won't find any of the text. The knowledge is distributed somehow (in a way that we don't fully understand).
Despite that GPT-3 is clearly a program.
That's not the case. Or rather, we don't know, we only have models and some are useful. In your statement there's a whiff of you having only a hammer, and everything looking like a nail.
I'm not saying I entirely disagree, only that the computer analogy is only an analogy, and has problems and detractors.
For example, this from 2005: "In cognitive science, the interdisciplinary research field that studies the human mind, modularity is a very contentious issue. There exist two kinds of cognitive science, computational cognitive science and neural cognitive science. Computational cognitive science is the more ancient theoretical paradigm. It is based on an analogy between the mind and computer software, and it views mind as symbol manipulation taking place in a computational system (Newell and Simon, 1976). More recently a different kind of cognitive science, connectionism, has arisen, which rejects the mind/computer analogy and interprets behavior and cognitive capacities using theoretical models which are directly inspired by the physical structure and way of functioning of the nervous system. These models are called neural networks—large sets of neuronlike units interacting locally through connections resembling synapses between neurons. For connectionism, mind is not symbol manipulation. Mind is not a computational system, but the global result of the many interactions taking place in a network of neurons modeled with an artificial neural network."
Google is making it nearly impossible for me to get a URL for this, here's a link I hope works:
https://scholar.google.co.uk/scholar?lookup=0&q=Evolutionary...
Your brain is but a preprogrammed, biological computer that reacts to data obtained from its interfaces and attach peripherals.
So it's a little more sophisticated than just adding random noise, it's adding very specific quantities of noise to very specific locations, which are based on perfect knowledge of how the predictive system (the deep model) works.
Is this stuff interesting? Absolutely. Is it worth studying? Yes, again. Does it mean that CNNs as we know them are poor computer vision systems and fundementally flawed? No. It's a limitation of existing deep models, and one which may be overcome eventually.