I have to admit that it’s a standard – but very artificial – framing. “You have to pick one of the other,” I always cheerfully say. Having wrestled gamely with the false choice, MIT Sloan professor Eric So pushed it back to me, saying “I’d love to know where you are on this.” I hadn’t reckoned on that, but one of the joys of hosting the podcast is that, more often than not, you find out what you actually think about something.
Eric recently published ‘The Collision: What AI Does to Us’. We had discussed the ideas behind the book, as well as the research and insight which drove him to warn of the impact on the human brain when we outsource our thinking to AI. My first observation was that I’d just asked a question that was pure “attention economy clickbait”. That runs counter to my personal belief that there aren’t many situations where there’s a pure right or wrong answer. For sure, whether AI is good or bad is not a binary choice; but it’s a helpful way to challenge guests…
What I feel is two things at once. Sometimes I’m awestruck by what it can do. Then I’m befuddled by how stupid it can be. Stupid isn’t quite the right word, but you’ll know what I mean if you use it every day.
Good at what it’s good at
A lot of my work is writing and editing, so here’s a concrete example. After I’ve recorded a podcast, I’ll feed the transcript into Claude and ask what it makes of the conversation. What are the key points? What’s interesting about it? What topics are discussed? What’s the flow and what are the sections, the topics?
Claude does a decent job of reviewing the conversation. However, if I ask it to write the show notes – What’s the hook? What’s the title? What’s going to get people to read this? – I never get something I’d call good.
Of course, that’s not what it should be doing. Its ability to understand what resonates or jars enough to gain attention is low. Spotting the what’s exciting about a conversation is my job. Trawling through an hour of transcript, sectioning it up and telling me what’s in there is its job. And it’s pretty good at it.
So that’s where I ended up with my answer to Eric: it’s good at what it’s good at, and bad at what it’s not good at.
The hype is coming from somewhere unusual
Something else has been nagging at me and it came out in the same monologue. Normally, at the outset of our adoption of a disruptive technology, its proponents are outliers. Whether it was the personal computer, the world wide web or the mobile phone, the disruptors begin with little or no power and leverage. They’re out on the fringes shouting, “hey, this is great, you should use this”. It’s completely understandable that they hype it up, of course.
I would argue that AI is a very different disruptive technology in that its being driven by very powerful organisations with very deep pockets. And they are grabbing attention and hyping a technology that isn’t ready yet. Or at least, it isn’t ready to deliver on the promises being made for it.
There are amazing things it can do, like dealing with large volumes of data and understanding language and responding to it. As Eric So said, it’s able to mimic the output of intelligent operators, but it’s at its most powerful (today anyway) when it amplifies the ability of experienced humans who can frame the problem, direct AI, and assess its outputs.
Eric likens students using AI to get the answers to work that is meant to build their capabilities as being like taking a forklift to the gym. The weights still go up and down, but you’re seriously missing the point.
There’s plenty more in the conversation: whether AI is really thinking at all, what the firefighters’ story is about, what happens when two people in an argument each ask their own AI, and how optimistic Eric is when it comes to it.
Listen to the full conversation:
https://podcast.curiousbusiness.co.uk/e/eric-so-mit-the-collision/
And check out Eric’s book, The Collision: What AI Does to Us by Eric So:
At amazon.co.uk | At uk.bookshop.org (affiliate links)
Where do you stand on AI - good or bad? 😀
