So what counts as AI this week and would you let an LLM into the doctor's office with you?

When do we stop calling it AI?

Dust off your memories of the ‘00s and before. What’s that 🌟 wow 🌟 AI development you recall being enthralled by which now feels more like a building block? 🏗

Back in 2016, Google Translate relaunched itself, shifting from statistical machine translation to the then cutting edge neural machine translation. With the incorporation of Word Lens and the by then seemingly ubiquitous mobile phones, you could seamlessly navigate your way round a city without even the ability to read the script. It felt astounding.

Today, according to Wikipedia, Google Translate is used by 500 million people daily. And we take it, and the computer vision that enables it, 100% for granted. Few would call it AI.

This phenomena of magic to just a tool is so common there’s a name for it and I found reading the article AI and the Frontier Paradox by Konstantine Buhler a great pause-and-reflect opportunity.

The history of #ai is littered with accomplishments that have worked well enough to no longer be considered sufficiently intelligent to earn the aspirational monniker.

A good read and a timely reminder to founders, investors, executives and NEDs alike to be curious and skeptical in the best scientific sense. 🔍🔍🔍

There’s a lot to find intriguing about Google’s Med-PaLM2

Personally I’m extremely bullish about the possibilities offered by AI-assisted #healthcare. Even in Australia, where I’m lucky enough to live, which has great access to publicly funded healthcare compared to many places, I’m regularly frustrated and dismayed by how little time and attention my primary care physician / general practitioner can afford to spend with me at any visit and how the cost imperatives of the system work against any meaningful conversations about preventative measures. Surely there is a better way than ‘wait until sick and then treat’ 🤢🤢🤢 - but that’s the way our current systems, by and large, are designed.

So using LLMs combined with search and summarisation to make health and medical information more accessible and understandable to the general population feels like a huge plus ➕➕➕

I believe it will help many of us become much more active 🏃‍♀️ 🏋‍♀️ participants in our own health journeys! 😀

Still, it’s disconcerting in this Verge article on early testing of a medical AI chatbot utilising Med-PaLM 2 to see the lead researcher from Google say that he “wouldn’t want it to be a part of his own family’s healthcare journey”. Um, why not exactly? 🤔

And my other niggle is that while the preprint on the new model release says in the abstract that

Med-PaLM 2 achieves an accuracy of 86.5% on USMLE-style questions, a 19% leap over our own state of the art results from Med-PaLM

you have to read to page 3 in the ‘Related work’ section to see that the new one is just barely beating vanilla GPT4. Or at least that’s how it reads to me.

Where do you land on AI in health? Do the pros outweigh the cons? Let me know in the comments below.