AI is seductive, but those Black Swans are still floating

Sophisticated AI is a great tool, but it will always be a tool, not a replacement for sentient beings

I was having trouble finding people to comment on a recent story I was writing. I called everyone I knew, I trolled LinkedIn for people I didn’t know, desperate for any reactions. (I got a couple, eventually.) But one contact sent me back a response that left me speechless. I didn’t know how to react, even more so, I wasn't sure how to write a blog post referencing it as a learning moment.

The first idea that came to mind was that better media training was needed. Then, as you will soon see, I wondered how this episode fit into my ever-changing view of how artificial intelligence will impact our work and daily lives. Finally, my mind went to diversity and how the future of society needs it, in all its forms, now more than ever. Could I include all three on one post? (On, go on, Liz, let’s make it all three.) OK, here goes.

In my struggle to find a comment on a vague piece of regulation, I sent along a list of questions. I received a response from someone I know and respect, asking if they could have the weekend to respond. “Sure”, I said.

A few days later, I got an email. “I popped your questions into ChatGPT, here you go.”

I was alone, so no witnesses. But I stared at my computer screen, open-mouthed, for longer than necessary, at the response.

“Does this person not think I know how to use ChatGPT (or Google)?”

“Why did they need the weekend to do this?”

“Does this person understand that I can’t use any of what they sent?”

I thought this was a one-off, an anomaly. Until I saw Nick Huber's list on LinkedIn. I clocked “number seven” and thought. “Oh, shit, this is a trend.”

Let me break it down. First “media training”.

There have been times when I’ve looked for people to interview who can explain things to me. (I am the font of some knowledge, but not ALL knowledge.) I’ve written two chunky features on quantum computing recently. For the first one, I called professors at the Massachusetts Institute of Technology and Oxford and started with the line: “I have high school level physics, please explain this to me.” And they did, and I got interesting angles to use in my final story and cool quotes to lend the feature gravitas.

However, 90 percent of the time, what I am looking for is a quote, a comment, a viewpoint. What I need is a well-known expert in the industry or a senior staffer at a bank to talk to me so I can use their name next to these marks (“ ”) in an article. Half the time, I know the answer to my question. What I need is a quote.

I have 30 years of experience in this industry, which concerns itself with how financial entities use data and technology, so I have developed my own views and knowledge. (And I have shared them, often, as my personal opinions). But as a journalist writing an article, my viewpoint – informed or not – is irrelevant.

Sidebar media training: I once had to explain to a freelancer that they couldn’t simply write a sentence like “banks don’t serve small businesses very well” without attributing it to someone. While that might statement may *feel* like a fact to many people in this industry, it is still just an opinion. You need a quote that pins that view to a person, or highlight a story from an SME that illustrates that point. OK, sidebar media training over.

To finish this “media training” portion of the post, I can’t quote “ChatGPT” in an article. The following section explains why.

Artificial intelligence is a tool that is fast emerging as a valuable asset in people’s work lives. I’ve admitted on stage that I’ve used generative AI in my work. I was running short of time, so I uploaded a four-hour roundtable into a chatbot to ask for an outline. This outline helped me identify the main points of the discussion and locate interesting quotes to incorporate into my story. The gen AI bot didn’t write my story, and I didn’t use all of the “main takeaways” it listed. But it did save me four hours watching people discuss “tokenisation” in slightly different ways until my head went numb.

That is a useful “tool”.

As sophisticated as AI will soon become (or already is in some corners), it will always be a tool, designed by humans, whose “intelligence” is fed by the information we choose to share with it.

AI is not, and never will be - no matter how good it gets at human imitation - a sentient being. It is a product of what it is given to learn from.

If we start overriding our research, viewpoints, and informed opinions to the call of the gen AI chatbot, which churns out analysis based on past “research, viewpoints, and informed opinions”, aren’t we walking into a future where facts and commentary are regurgitated like chum? No interesting context, no pioneering ideas, no sparks of genius or joy.

I once had an editor who advised that if you couldn’t decide which story should lead and which should be “on the other side of the fold” (;-) ), always err on the side of the story about “a person”. Even the most stoic amongst us will turn their head from a bland list of statistics and facts to a real-life story about a person who “did a cool thing”. (Even if it’s about something like corporate actions processing.)

Which brings me to the third point – diversity.

AI, in all its forms, is being viewed as so powerful and revolutionary that the impact it is expected to have on work and our everyday lives is both scary and exciting. However, please remember the lessons from above. AI is a tool, and its development, learning, and usefulness are fed by what we feed into it. That information should be accurate, comprehensive, and diverse.

The other week, amid protests against immigration actions in California in the US, several politicians, including Gavin Newsom, the governor of the state, reposted a picture from the San Francisco Chronicle showing National Guard personnel sleeping on the floor. The photo was intended to illustrate how the federal government had deployed these troops to the state with minimal planning and care.

To verify whether these photos were legitimate (a reasonable question in this age), many people submitted the photo to a generative AI chatbot. However, the answer many people got from these chatbots was inaccurate. Many inquiries said these images were from “the war in Afghanistan” and not from recent developments in California. This, of course, led many on one side of the political spectrum to accuse Gavin Newsom of spreading false information. But it wasn’t the governor who was wrong; it was the chatbot.

Some might use this story as an example of “right-wing bias” in the development of the gen AI. However, I believe the flawed analysis is a result of a lack of comprehensive and diverse information.

Take the most famous “surprise” information of them all. Around 300 years ago, people in Europe believed that all swans were white because all swans in the Northern Hemisphere are indeed white. However, a consequential visit from Captain James Cook to Australia in 1770 revealed the existence of black swans. Does that mean that black swans didn’t exist before colonial powers ventured Down Under? Of course not, their ability to float around in a pond in Australia, in all their goth glory, does not depend on who plants which flag on a foreign land. However, some 18th-century gen AI chatbot (bear with me, please) would deal with a myriad of questions about swans and never deviate from the “fact” that all of them were white.

This makes diversity, in all its forms, more critical than ever. We need diverse contexts, experiences, research, content, and viewpoints. It is only through embracing all of these “data inputs” that we will ever understand nuance, discover new angles, and get more accurate “facts” from our sophisticated AI tools.

Sparks of genius or joy rarely come from regurgitating chum. Don’t tell me there are millions of pebbles on Brighton Beach; show me a perfect, round, smooth stone that your five-year-old son found and painted a ladybug face on. Show me the anomaly, the tiny crack behind the sofa, the faded tablecloth your grandmother crocheted, stained with coffee rings and ancient ketchup drops. Show me the humanity that still exists everywhere in our world, which should always override insights gained from millions of lines of code and fragmented data centers.

***

You knew there was no other choice.

Swan Lake – Entrée and Adage from the Black Swan pas de deux (The Royal Ballet)