Take 5 for Tech

AI Fails & Cautionary Tales: AI Series #6 (2.21.24)

Beware of AI Failings

There are all sorts of examples in nearly all sectors of AI gone wrong. This newsletter showcases some of the AI foibles. Experts find that there are often similarities the errors. Knowing those similarities makes it easier to discern the fails. While there are too many error types to list in this newsletter, here are a few to get you thinking. To know more about these and others, take a look at the link above. It gives a brief, but great overview.

Unyielding thought or "brittleness"-- The inability for AI to see things differently. One example was not being able to identify an image of a bus because it was "turned on the side." While AI is always improving, it previously neglected to identify the rotated bus 97% of the time.
Promoting Bias as Fact-- AI regularly has bias innately baked in it. Among the biases, racial bias can be a widely prevalent consequence of AI. While algorithm programmers are working to make changes, bias against people with darker skin is still a widely-known issue.
"Catastrophic Forgetting"-- With the rise of "deepfakes," false images, videos and audio, AI eventually learns how to detect them. However, researchers have found that as AI learns new ways to detect deep fakes, it interestingly can "forget" how to detect them using older, previous methodology.
Lacking Common Sense-- This one is exactly as it sounds! A great example is the image attached to this section. Check out the picture, Is that the type of salmon you would expect to "swim" in a river? I don't think so!
Bad at computations-- Prior to my research I wouldn't have known that a calculator far outshines AI in math, but it's true. Researchers have studied why this is so. There are some theories that ChatGPT is bad at math because it's a model based on language and not numbers. Presently, the verdict is still out whether AI is improving its computations as studies have a wide discrepancy in findings.

All things considered, AI can be an excellent resource, but it clearly still has a wide host of limitations. If left unscrutinized, it can quickly turn into an unhelpful resource to varying degrees.

Deepfakes

zoom_out_map

If I'm being honest, while AI can be used in some very nefarious ways, deepfakes tend to scare me the most. Maybe it's the visual nature of the deception or maybe it's how real they now look, but regardless, this is what first comes to mind when I think of the frightening side of AI. It innately makes the spread of false information that much easier. Additionally, as it stands now, deepfakes are generally speaking legal except for a few states that created laws in some very specific instances.

The "I Deep Faked Myself, Here's Why it Matters" video gives an excellent overview of the entertaining & darker sides of deepfakes. It gives tangible examples and implications. If you overlook the one-minute pitch for Incogni between the 2-3 minutes mark, I'd give the video an A!

AI Gone Wrong- Updated & Eye-Opening List of Fails

Cases of Professional AI Misconduct & Mishaps

NY lawyer Sanctioned for Citing AI Generated & Nonexistent Cases

The Lighter Side of AI Fails (51 Funnies)

Deep Web Vs. Dark Web

You probably hear the phrases "deep web" and "dark web" often, but do you really understand what they mean? I found this handy resource from Tulane University to help break it all down & it's what I used to create this passage.

It's estimated that 90% of all internet is deep web, meaning the content is encrypted to protect personal data like in the case with emails or medical and financial information. A lesser known phrase, "surface web" is exactly as it sounds, visible to typical internet users. However, this content actually only consists of about 5% of content on the web.

That leaves the dark web, which then makes up the last 5% of internet content. Yes, you heard that right; the dark web represents approximately the the same amount of content as surface web. The dark web's purpose is to give users the ability to share and communicate anonymously. It consists of both illegal, and lesser thought of legal uses. There's plenty to learn about the dark web and deepfakes. The prevalence of these technologies provide even more evidence that people need to be critical thinkers to determine what is true.

No matter how you feel about AI, it's abundantly clear that we need to stay informed and to continue to teach students to scrutinize content with lessons like the SIFT Method. And while some things "may not be there yet" in an undetectable, perfect sense, they get closer to undetectable every day.