Machine Learning and the Luck of the Irish

This week the world celebrates all things Irish, with St Patrick’s Day coming up on the 17th of March. We’ll talk about Irish things soon enough, but first, while attending the #AFI2018 conference last week in San Francisco, there was some worthy discussion around A.I. and machine learning. A very interesting session lead by Springboard Auto’s Khalid Ayadi, made me think about what happens when information is incomplete or error-strewn or not even there. How does machine learning cope? Further, what is the current state of it today and what are the possibilities in the future?

This isn’t just a question as relates to ordering a car and it’s financing – via Amazon’s Alexa (nice demo by the way #SpringboardAuto), but also to other areas such as autonomous driving i.e. road, weather, traffic conditions etc. How do AI/machine learning systems cope with imperfect and incomplete or even non-existent data?

I want to share a seemingly unconnected story about Ireland I had planned to discuss at AFI 2018 with Royal Media‘s Natalie Matilla, but ran out of time. I was on the BBC News website that morning before our Fireside Chat, when an article with an unusual title caught my eye.

‘No Address’ parcel delivered in Donegal.

With a title like that, of course I needed to find out more about it (website link is at end of this article).

To explain my own Irish connection, some disclosure for those who don’t know me. Although born and raised in faraway (and sometimes mythical Middle Earth) New Zealand, I have lived in (Southern) California for the past 20+ years. After briefly going back to NZ for the Millennium celebrations (i.e. 2000), later that year back in the U.S., an Irish friend and family kindly invited me to go to the other end of the world, to Ireland (Co. Donegal specifically), for Christmas and New Year’s. From then on, my life has forever been tied to that most amazing, friendly, charming, and indescribably beautiful county and country.

After 18 years now, of going there for more Christmases, NY’s, summertime trips, many weddings – including the one to my wife, I can just about tell the difference amongst 4 or 5 of the 30-odd Irish accents. It probably took me about six years alone just to decipher the Co. Donegal accent to an acceptable level.

Truth be known, although I still probably mistake what has been said sometimes – and what was meant, at least I can claim to understand 10 different forms of English now (i.e. including other countries versions). Irish (ok, all forms of,) Gaelic is still a mystery, although my children have been picking it up for years. Multi-dialect/Machine learning indeed…

Back to the article,and here is a picture below of the parcel with the address written on it, well scribbled mostly, and pretty much wiped off it.

Irish parcel #1 – Successfully Delivered

How could a parcel with no discernible address, ever find its way home? (Hint: have you ever been to Ireland and asked for directions?)

The parcel did indeed make it to the correct person, in Ireland, in Co. Donegal. To a Shane Crumlish, who lives in Carndonagh. (I don’t think I’ve been there). There was a note attached to it that said simply “Try Carndonagh, Co Donegal.” Now, the reason apparently it made its way to Mr. Crumlish he said, was that two local postmen had figured it out.

” I’ve been talking to the postman, John from Moville,” said Mr. Crumlish. “Between himself and another postman they have done a bit of head-storming and came up with the idea: ‘It must be for Shane.’“

For Shane. Of course, it was for Shane…

Apparently, Mr. Crumlish sometimes received similar looking parcels, though probably with the full, legible address attached. You might be thinking “only in Ireland” and you might be right. Machine learning indeed…

There’s another case back in 2015, where the Irish Post, in Co. Donegal again, delivered a letter to another, seemingly obscure address. The reference to the other case is ‘The boy with the glasses’, and the full address this time was at least legible:

“ Your man Henderson, that boy with the glasses who is doing the PhD up here at Queen’s in Belfast; Buncrana, County Donegal, Ireland.“

Irish parcel #2 – Successfully Delivered

(Growing up in a small island country, I’d also bet good money Ireland was written in full caps as “IRELAND”, on the full address).

Yes, the address was legible, but not exactly the precise description you were used to. Without going into details, this letter was also delivered successfully, to your man* Mr. Henderson, in Buncrana. Well, to his wife, Roisin, who’s office it was delivered to. Machine learning – and the luck of the Irish, indeed…

So, my question about machine learning is, at what stage of understanding has machine learning evolved to? Could it deliver and/or identify Your man Henderson? Perhaps it is more advanced still and can already deliver to your other man, Shane Crumlish? Or is it only about as ‘good’ as my understanding of the Irish accent, even after learning and listening to it for over 18 years? For myself at least, the phrase, “divided by a common language” comes to mind…

Turning the question around a bit, what can we learn from analog learning methods? Does the Irish way of thinking and the results described earlier, illustrate some other aspects of learning and intelligence that can’t or haven’t yet, been entirely captured by A.I.? (No misquote about Freud and the Irish need apply here.**)

This isn’t my area of expertise, so these are open questions addressed to the tech community at large. How are the algorithms that are being written and curated, managing to include this type of seemingly unrelated information? Big data, and neural networks alone aren’t a full or even partial answer, so can anyone clarify or comment on this to an extent that might illuminate us? Are these even the right questions to ask?

At the very least, I hope I shared an interesting anecdote connecting Ireland, machine learning and analog learning. Maybe even increased your own learning?…

Thank you to the entire Royal Media team, for a such an impressive conference. I’m looking forward to the next one!

Matt Traylen is the Head of Financial Services at Faraday Future.