* AI does best on images it's seen before. We know AI is good at memorizing stuff; it might even be that some of the images in the examples and benchmarks are in the training datasets these algorithms used. Giraffe With No Spots may be especially difficult not only because the giraffe is unusual, but because it's new to the internet.
* AI tends to sand away the unusual. It's trained to answer with the most likely answer to your question, which is not necessarily the most correct answer.
* The papers and demonstration sites are showcasing their best work. Whereas I am zeroing in on their worst work, because it's entertaining and because it's a cautionary tale about putting too much faith in AI image recognition.
The first and last points are basically how much of advertising and related things work. It is opposed to how science should work, but of course there's publication bias and so on. It's not how popular science should work, but in fact it is almost entirely how popular science journalism works (looking at you, LK-99, Cold Fusion, and Hot Fusion, too, and many others).
BTW the first point above is also known as 'data leaks', which is when you feed your software a multiplication table with single-digit numbers and then all you do is ask for the result of 4x5, 6x2 and so on. You then go and publish a well-received paper, "Now Computers can MULTIPLY!" where MULTIPLY is a fun acronym, and don't forget the exclamation mark. You did forget to mention that in the end, your software is, statistically speaking, incapable of multiplying any numbers, and we're not even talking rationals here.
* AI does best on images it's seen before. We know AI is good at memorizing stuff; it might even be that some of the images in the examples and benchmarks are in the training datasets these algorithms used. Giraffe With No Spots may be especially difficult not only because the giraffe is unusual, but because it's new to the internet.
* AI tends to sand away the unusual. It's trained to answer with the most likely answer to your question, which is not necessarily the most correct answer.
* The papers and demonstration sites are showcasing their best work. Whereas I am zeroing in on their worst work, because it's entertaining and because it's a cautionary tale about putting too much faith in AI image recognition.
The first and last points are basically how much of advertising and related things work. It is opposed to how science should work, but of course there's publication bias and so on. It's not how popular science should work, but in fact it is almost entirely how popular science journalism works (looking at you, LK-99, Cold Fusion, and Hot Fusion, too, and many others).
BTW the first point above is also known as 'data leaks', which is when you feed your software a multiplication table with single-digit numbers and then all you do is ask for the result of 4x5, 6x2 and so on. You then go and publish a well-received paper, "Now Computers can MULTIPLY!" where MULTIPLY is a fun acronym, and don't forget the exclamation mark. You did forget to mention that in the end, your software is, statistically speaking, incapable of multiplying any numbers, and we're not even talking rationals here.