I really struggle with AI in general. I have yet to see something it actually does that hasn't already been done for many many many years.
Using voice recognition and Apple as an example, I distinctly remember when Speakable Items was added to Macintosh System 7.1.2 some time in the 1990s. It was so cool, it worked really well, and it was programable by performing AppleScripts with custom commands. TBH, it worked quite a bit better for some use cases than Siri does today.
Years later in MacOS X in 10.3, they reintroduced a lot of the capability that was lost in the OS X transition.
In 10.8, they added dictation, with offline dictation being added in 10.9.
In 10.15, they claimed "Voice Control" as a new feature that had actually been available for longer than 20 years on macOS. The claim was that users could now give instructions to to control applications - to a user of voice control since the 90s, this wasn't new to me. Maybe the need for AppleScript was eliminated through new APIs?
Now, as we move into the AI age, everyone is trying to convince me that voice recognition and the ability to say verbal commands and get responses is something new and complex.
To use another example:
Tools like chatGPT - what I observe, and have been shown by people that are wowed by its capability, is a simple input, response chat bot.
I distinctly remember chatting with Smarter Child - one of, or perhaps the very first, chat bot designed for instant message platforms in the 1990s (AIM, MSN, etc). At first, it was a toy - you could ask it it's name and have a basic "how are you today?" conversation and it would give silly or appropriate basic responses. By the late 90s it was a useful tool to do things like get movie times at the local theater (which at the time would have otherwise required making a phone call to the theater or checking the newspaper), getting phone numbers and addresses for businesses, and things like that.
As far is information is concerned, it seems I can get the same information by doing a Google search that I can by asking chatGPT. The information results from chatGPT are often more difficult to get (as it keeps asking to be more specific) and difficult to verify.
I just don't get it. AI isn't doing anything for me that I haven't already been able to do for decades. I don't understand all of the apparent compute power required to do things that are at best only slightly better than what was being done in the 1990s on chips measured in MHz and RAM measured in KB.