“I’m your personal AI assistant. I can help with any questions about your site or account.” Anyone who knows me well is aware of my above-average verbal fluency. But that comes from a lifetime of communication with fellow members of my species, where with a little give and take, a little clarification, a smile, a gesture, I can get my point across, even if it involves a little effort on both sides. But with greater and greater frequency, I am (make that ‘we are’) being asked to have a dialog, a conversation with something non-human, in which you ask a question and an answer appears on your phone or computer. I’m not as clueless about AI as the current U.S. Secretary of Education, who confused AI with A-1, a bottled marinade you can slather on a steak, but I’m never satisfied by outperforming the dumbest person in the room.
Some of the information that can be produced makes sense to me. I can understand that a computer can examine the possibilities of the next move and the move after that in a chess game with greater speed and fluency than any mere mortal. When we thought we might need to find another hospital to perform laser surgery on my prostate, Richard Levine used ChatGPT to produce a useful list of likely hospitals and surgeons in a jiffy, which thankfully I didn’t need after all.
But an actual conversation with a computer? That’s another level of complexity, as you will soon see. How was I going to explain what was vexing me, that I was having a software-related issue beyond my admittedly low level of competence? I had just finished my latest article (The Oy Gevalt Group, if you are keeping track) and hit the ‘publish’ button, sending it out to my most devoted followers, those who are actually subscribers. After that, I copied the link to the article to another group of devoted followers, the readers I reach in an email group called ‘blog,’ after which I posted the same link on my Facebook page for the stragglers to locate. That’s what I’ve been doing since 2015 when I started using WordPress to publish my ramblings. One more thing: I proudly announced to Barbara that ‘literary history has again been made,’ or some similar grandiose statement, to alert her that she should look in her inbox for my latest effort. Which is what she did, except there was nothing there.
The reason I subscribe to my own blog is so I can check on what others are receiving from me. Let me take a look, but to my dismay, I hadn’t gotten my post either, meaning that it was unlikely that any of my faithful subscribers would know that I had sent anything out. Something must have gone wrong. WordPress must have a help desk, somewhere. Sure enough, there in the lower right, waiting patiently for business, was the AI assistant. And so it was, in my eighty-fourth year, that I would have my first conversation with a machine.
My initial attempt at outreach was, admittedly, lacking in clarity: ‘published article but it didn’t go out,’ so I can’t blame my AI assistant for scratching its virtual head, trying to figure out what I was trying to explain. I will give credit to AI for one thing: it never once responded, That’s not what I asked you, stupid!, or some even less polite retort. Instead, it suggested, Did I hit ‘publish?’ To which I responded, ‘hit publish no email notification to subscribers,’ which was closer to what I needed to say.
There was nothing for my pal at AI to do but to start at the absolute beginning, as in verifying that I have actual subscribers, because you never know, especially if you are a machine. But here’s the thing, as a former U.S. president was wont to say. AI doesn’t actually ‘know’ anything; It simply parrots back the information – every last bit of it – that has been fed into it.
We spent a lot of time and effort, my new-found friend and I, looking for an elusive ‘small paper airplane icon,’ which might have gotten turned to ‘post only’ instead of ‘post & email.’ Except I couldn’t find said icon anywhere where it was supposed to be, which totally baffled AI. Nothing that had been programmed into it would explain how an icon could simply disappear from the face of the screen. Finally, it thought to ask the obvious question, once it prompted me to reopen the article I had sent out: ‘Let me know what you see in the top right area.’ To which I responded, ‘There are several icons: view post, view, jetpack arrow that says newsletter…” Then it dawned on me. Could it be that what I thought was a pudgy arrow was actually the elusive paper airplane icon we have been searching for the last half hour?’ I feel confident that if I were talking to someone of my own species at a help desk we would have figured out the problem a lot quicker, because a human would have immediately considered it more likely – or at least as likely – that the icon was there all the time, but I didn’t recognize it. And sure enough, somehow it got set in error to ‘post only,’ so that, while the article was visible on my WordPress site, nobody else would automatically see it, which sort of defeats the purpose.
Now what? I now knew enough to make sure I wouldn’t screw up again, but…? I’d sure like my faithful subscribers to see what I had already written. Can I simply resend the article. It appears not. So I asked: ‘Can I copy the contents of the post I sent out into a new post and send that out?’ (Notice it’s me doing the asking, trying to find a solution.) Answer: I can duplicate it, which would create a second version of the article, then make sure it says, ‘post & email, and only then hit ‘publish,’ sending it out to my loyal subscribers, who have been wondering where I’ve been all this time.
I took the trouble to tell AI that ‘This time it worked. Thank you.’ To which it responded, ‘Awesome.’ That’s called me making AI’s day; glad to be of use.
Now that I’ve gotten my toe wet using AI, maybe it was time for some virtual skinny-dipping on the internet. My parents had to adjust to how television took over our apartment in the 1950’s, my grandparents to things as radical as radio and the phonograph were back a hundred years ago. I had to learn my way around computers and the internet. AI? Just one more thing coming down the pike for those of my generation to watch out for. I couldn’t pick up a newspaper (remember them?) without some reference to this new phenomenon. Most of them were as informative as reporting that some people think it’s great, some think it’s terrible, as in ‘thank you very much.’
More intriguing is an article entitled ‘How to use AI to talk to whales’ in The NY Times by David Gruber, the head of Project CETI (Cetacean Transition Initiative) a 50-plus person team… ‘working to listen and decode the communication of sperm whales.’
They just published a study that showed sperm whales have what appears to be vowels and diphthongs, and use them in ways similar to how humans do. The team will soon be releasing Whale Acoustics Model, ‘a novel AI system that translates any audio into sperm whale vocalizations – allowing human to potentially experience, for the first time, what it might be like to interact with a whale in its own language…’
I can barely communicate with my Israeli neighbors in the local language, so being able to chew the fat with cetaceans is probably overkill for me. Still, I should be able to find some use for AI in my own life. Let me at least download an AI program onto my computer and see what I can see.
If at first you don’t succeed…. I learned fairly quickly that my vintage-2017 iMac (powered by now-obsolete Intel chips) didn’t have what it takes to use Chat GPT. But my phone? OK, it’s an iPhone 12, but it can handle the lates IOS. Sure enough, I was able download AI on my phone and fire it up. Let’s see what it can do. Start with something simple. What year was the Empire State Building completed? ‘1931.’ Let’s up the ante. What year was P.S. 80 in The Bronx built? (In case you’re wondering, that’s the school that Frank and I went to from K-9th grade, and our sister, Marilyn June, graduated from after our parents moved to 77 E. 208 St. in 1941.) ‘1957.’ Nooooooooo, you doofus. The correct answer, according to a plaque inside the building, is 1927.
Is there a way to argue with AI? Why don’t I ask AI itself how you do that? How does one correct AI? Here’s what it has to say for itself:
‘Correcting AI is an ongoing process that involves several key elements:’ And then it lists Feedback, Data Quality and Diversity, Model Architecture and Training, Testing and Evaluation, and Ethical Considerations and Bias Mitigation, each of which has several sub-categories like, ‘Algorithmic Improvements’ and ‘Reinforcement Learning.’ I’m so glad I asked! You know the old adage, which goes something like, If you can’t outdo them with brilliance, dazzle them with bullshit.
Let’s try again. How many recordings did John McCormack make? (Who is John McCormack? Click here for a recording of Stephen Foster’s Jeannie with the Light Brown Hair.)The one word answer was ‘John.’ That’s all AI could come up with, not even the singer’s full name. (Just for the record, if you go into Wikipedia and scroll down to external links, you can search ‘John McCormack recordings at the Library of Congress’ (166 entries) and ‘Public Domain Recordings of John McCormack at the Internet Archive’ (192 entries). I’m beginning to understand why some people think AI is the bees’ knees and others are not even underwhelmed. I guess it depends on who you ask or, better still, what you ask – AI that is.
I have one task that AI should be able to perform and hasn’t so far: traffic control. Consider the following situation: there are four cars waiting for the signal to turn green. (I’m thinking about the intersection in Jerusalem where Route 1 meets Bar-Lev, but any busy intersection with a traffic light will do.) When the light finally changes, three of the four vehicles get to go through before the light changes back to red, leaving the last car waiting and the driver exasperated. A traffic cop, one from our species, would figure out to let the fourth car go through as well, but the signaling system on hand goes by time, not by need, being oblivious to the facts on the ground. So what’s the problem? I know whom to ask!
‘There are several reasons why AI doesn’t fully control all traffic yet, and why it’s a complex challenge:’ Are you ready?
Complexity of Real-World Scenarios, Data Requirements and Sensing, Safety and Reliability, Ethical and Decision-Making Dilemmas, Infrastructure and Cost, Public Acceptance and Trust, and Regulatory and Legal Frameworks, each of which has (you guessed it!) several sub-categories like ‘Ubiquitous Sensing’ and ‘Data Fusion.’
You know what I say? Excuses, excuses.
Just to be fair, I repeated the question about John McCormack, but this time I specified ‘The Irish tenor,’ not to be confused with all the other random John McCormacks floating around. AI hemmed and hawed but finally reported that ‘it’s generally estimated that he made over 600 recordings.’ That’s more like it. Now if we can persuade AI to pay a visit to my old stomping grounds in The Bronx, it can read for itself the plaque that confirms that P.S. 80 was built in 1927. Maybe it can figure out the traffic patterns along Mosholu Parkway, while it’s at it.
