The real reason AI is difficult

This Christmas, my friend’s grandmother finally found out what her grandson had been working on for years. He’s a data scientist raised on English with a bit of Spanish that gets dusted off on occasional family occasions. His grandmother speaks only Spanish.

“Before today, my grandmother had no idea what I actually do for a living.”

The sci-fi-fuelled rumors of what data scientists work on — especially if we specialize in AI — attract a whiff of the ridiculous, so many of us find ourselves constantly having to explain our life choices. That’s why I was very touched when I learned that my friend shared a Spanish translation of my blog posts (La explicación más simple de Machine Learning que jamás habrás leido and ¿Qué diablos es Ciencia de Datos?) with his family while visiting them for the holidays, and his grandmother was finally able to tell him she understood him and was proud of him.

This was the first Christmas where she could picture what he really did at work.

For the first time she understood that his calling wasn’t something vaguely killer-robots-flavored, but rather something that’s deeply beautiful for its ability to make human life better. Machine learning and AI give you a second way to talk to computers. The old way was giving the computer explicit instructions, whereas the new way is to give examples — data! — instead. That’s the essence of AI.

Don’t let its simplicity disappoint you; levers are simple too, but they can move the world.

It’s moments like these that I’m most encouraged in my writing. The reason I write (and usually give up half of every weekend to do it) is to make data science and decision intelligence accessible and easy to understand for all humans.

I believe that basic literacy in the ideas is crucial for participating in the AI future that humanity is inevitably headed towards, and I shudder at the idea of anyone being left behind.

I don’t want AI to only belong to academics, experts, big companies, and those who speak the same languages as popular bloggers. It’s a lie that applied AI is brainy and mystical —here are the real reasons it’s difficult today:

(1) Pulling off automation at massive scale is hard. That’s the stuff I make all kinds of fuss about, but it’s not an AI-specific problem. Doing anything at global scale is always a beast of complexity, from making burgers to delivering search results. However, at individual scale it can be pretty easy, unless the tools aren’t user-friendly. Which brings us to the second point.

(2) Today’s tools suck. Yes, all of them.

What do I mean? Simply this: most five-year-olds can’t use today’s data science tools. The ideas aren’t difficult (examples go in, a recipe comes out), but the interfaces are. They might not look that way to the lifelong engineers among you, but they’re daunting to those who have never written a for loop before. The good news is that the tools are getting better quickly. Soon, non-experts will be able to use easy interfaces to cut all kinds of annoying drudgery out of their lives… as long as they know it’s an option. I don’t want to see people excluded from being architects of better lives for themselves with AI just because they can’t read English or they didn’t study math in college.

The good news is that the tools are getting better quickly.

(3) Testing an important application takes expertise. As a statistician, I probably rage about testing even in my sleep —being grumpy about lack of rigor is a deeply-ingrained instinct with my kind — so I have to remind myself that there are many applications where failure isn’t painful.

Sometimes failures are so funny they’re actually the best part — for example in this AI-generated recipe for Grilled Snailsed Butter, courtesy of aiweirdness.com. I challenge you to read it out loud without cracking up.

An AI-generated recipe, courtesy of aiweirdness.com. I dare you to read it out loud without cracking up.

Such applications don’t need testing rigor of the stratospheric standards that would make my fellow curmudgeons proud. There’s a difference between creating a lifesaving medication and creating a new butter recipe for my domestic amusement. I’ll join a protest this instant if we decide medication no longer needs statistical testing. As for the can’t-believe-it’s-butter… I do crazy stuff in my kitchen all the time where the only thing approximating a test is the fire alarm. (It sings me the song of its people often.)

In my talks and articles, I’m often championing industry use cases, but let’s not forget all the individual cute applications that bring a bit of personal joy or comfort. I have plenty of custom-hacked personal ML-based productivity tools that let me get away with acting like my day has 25 hours in it. These aren’t capital-I-Important but they make my little life a little better.

I’d love for everyone else to have access to these technologies — which is why I’m proud of be part of Google Cloud… making better tools and putting vast computing resources in the hands of all the people who don’t want to build their own datacenter is basically what we exist for. Unfortunately, people still end up excluded if they don’t know what AI is and that they’re allowed to join the fun without becoming a professor first. All that sci-fi nonsense about robots really doesn’t help matters.

I don’t want to see people excluded from being architects of better lives for themselves just because they can’t read English or they didn’t study math in college.

Everyone should be participating in the ideas of AI, and that’s what motivates me to write. I’m just one voice among many — and that’s a great thing, because every little bit helps. I’d like to say thank you to the people who are helping me take the message outside the English-speaking tech bubble that I’m stuck in.

To those of you who have taken a moment to share AI ideas — mine or anyone else’s — with someone who might otherwise be excluded, please accept a heartfelt THANK YOU. Everyone deserves to be part of this, no matter who they are or what they speak.

*******************************************************************

My second Medium account hosts community-translated articles in other languages. Here are examples in 🇦🇪 Arabic, 🇨🇳 Chinese, 🇳🇱Dutch, 🇫🇷 French, 🇩🇪 German, 🇮🇹 Italian, 🇯🇵 Japanese, 🇧🇷 Portuguese, 🇷🇺 Russian,🇪🇸 Spanish, and 🇹🇷 Turkish.

I wish I knew every language so I could write in all of them, but I don’t. If you want to go the extra mile in getting the word out, I always welcome volunteer translators. Google Translate gets the bare meaning across, but it loses much of the joy. I write the way I do because I think the joy is crucial…otherwise I’d just link to a textbook and call it a day. I’m extremely grateful to those who capture the spirit of my articles so they’re still fun when they land on the other side of the language barrier.

read original article at https://hackernoon.com/the-real-reason-ai-is-difficult-10b64a230c5e?source=rss——artificial_intelligence-5