Mistaking a teapot shape for a golf ball, due to surface features, is one striking example from a recent open-access paper:
The networks did “a poor job of identifying such items as a butterfly, an airplane and a banana,” according to the researchers. The explanation they propose is that “Humans see the entire object, while the artificial intelligence networks identify fragments of the object.” News, “Researchers: Deep Learning vision is very different from human vision” at Mind Matters
“To see life steadily and see it whole”* doesn’t seem to be popular among machines.
*(Zen via Matthew Arnold)
See also: Can an algorithm be racist?
3 Replies to “Artificial intelligence: Machines do not see objects as wholes”
It looks as if the “deep learning networks” weren’t trained for the situations the investigators explored. Perhaps if the networks were trained in such a way that they were presented with deliberately confusing images such as the investigators used, they would of necessity develop overall shape discrimination ability. These training images would be pictures of objects where the surface has an image of something else. I don’t think the limitation in deep learning networks discovered by the investigators is some sort of fundamental limitation of AI. “Recognizing” objects by their shape under a lot of varying conditions should be achievable by AI, but this is no indication that AI can achieve any form of self aware consciousness.
Not too long ago I might have been inclined to say the same thing, but as I have learned more about the profound differences between mind and materialistic processes, today I would be much more inclined to believe this ‘problem of seeing wholes’ for computers is far more intractable than someone would be inclined to believe at first glance.
As Michael Egnor stated, “Your computer doesn’t know a binary string from a ham sandwich. ”
Simply put, computers don’t do context. A subjective immaterial mind is required in order to take an overall context of a given situation into consideration.
For prime example, Richard Dawkin’s infamous Weasel phrase, (which Dawkins used to try to prove the feasibility of evolutionary processes, and which William Dembski debunked), simply does not make any sense without taking its proper context into consideration
This ‘problem of seeing wholes’ and/or the ‘problem of discerning contexts’ for materialistic processes is far more problematic than many people realize at first glance.
This ‘problem of discerning contexts’ is clearly illustrated by programs that seek to translate languages. As the following expert put it, “ambiguities at the lexical (word) level,,,, required “holistic interpretation.”
In the following article, Erik J. Larson gives a simple example of something requiring “holistic interpretation” that would be irresolvable by extant AI systems:
In the following video Pastor Joe Boot, although he is talking about the universe as a whole, illustrates the insurmountable problem that ‘context dependency’ places on reductive materialistic explanations:
And Pastor Joe Boot’s observation that materialists have no overarching design plan ‘to connect the dots’ is now rigorously established as a valid principle.
Specifically, Gödel’s incompleteness theorem, which can be stated as such, “Anything you can draw a circle around cannot explain itself without referring to something outside the circle—something you have to assume but cannot prove”,,,
Gödel’s incompleteness theorem,,,, has now been extended to physics.
In the following article entitled ‘Quantum physics problem proved unsolvable: Gödel and Turing enter quantum physics’, which studied the derivation of macroscopic properties from a complete microscopic description, the researchers remark that even a perfect and complete description of the microscopic properties of a material is not enough to predict its macroscopic behaviour.,,, The researchers further commented that their findings challenge the reductionists’ point of view, as the insurmountable difficulty lies precisely in the derivation of macroscopic properties from a microscopic description.”
To put it mildly, this is not a minor failing of the Darwinist’s reductive materialistic worldview. Besides, as witnessed by the abject failure of the inflationary model, i.e. the inability of reductive materialist to be able to give an reasonable account for why the universe, or any object within the universe, may have the precise structure, shape, and/or form that it does,
Besides the inability of reductive materialist to be able to give an reasonable account for why the universe, or any object within the universe, may have the precise structure, shape, and/or form that it does, the ‘bottom up’ reductive materialistic framework of Darwinian evolution is found to be grossly inadequate for explaining how any particular organism might achieve its basic form.
“Bottom up’ reductive materialistic explanations of Darwinists simply don’t have the capacity within themselves to explain why any particular organism may take the precise shape and/or form that it does.
As Stephen Meyer stated, “the information needed to code for complex biological systems vastly outstrips the information in DNA”.
The following gives us a small glimpse as to just how far “the information needed to code for complex biological systems vastly outstrips the information in DNA.”
In the following video, it is noted that the information to build a human infant, atom by atom, would take up the equivalent of enough thumb drives to fill the Titanic, multiplied by 2,000.
And the following video states that “There are 10^28 atoms in the human body.,, The amount of data contained in the whole human,, is 3.02 x 10^32 gigabytes of information. Using a high bandwidth transfer, that data would take about 4.5 x 10^18 years to teleport 1 time. That is 350,000 times the age of the universe.”
Obviously that staggering amount of ‘programming information’ that specifies what form and/or shape an organism may take cannot possibly be stored in the embryonic cell, nor can it be stored, least of all, in the DNA sequences as is presupposed in the reductive materialistic premises of Darwinian evolution, but this staggering amount of information specifying what form and/or shape an organism may take must be coming into the developing embryo ‘from the outside’.
To back up that obvious conclusion, at about the 41:00 minute mark of the following video, Dr. Wells, using a branch of mathematics called category theory, demonstrates that, during embryological development, information must somehow be added to the developing embryo, ‘from the outside’, by some ‘non-material’ method.
The following article adds weight to Dr Wells assessment and states: “the process of development should be thought of as being controlled by an “algebraic structure outside space-time itself”
To provide further evidence for information coming into the developing embryo from ‘outside space-time itself’, it is also important to note that quantum correlations somehow arise from outside spacetime, in the sense that no story in space and time can describe them,,,
And these quantum correlations which somehow arise from outside spacetime, are now found in molecular biology on a massive scale. In every DNA and Protein molecule,,,
Thus in conclusion, whereas Darwinian evolution is found to be grossly inadequate for explaining how any particular organism might achieve its basic form and/or shape, the Christian Theist is sitting very comfortable in his claim that God has formed each of us in our mother’s womb.