Emailing robots

By Louis Eksteen

Over the holidays I brushed up on my ping pong skills with my godson. He had already downloaded an iOS game before we left, which we used in preparation. Real life is different though and where we played, the tables were outside, so wind was always a factor. You can imagine the light ping pong ball swerving even more through the air, difficult to control with spin and wind thrown in.

By the end of the holidays, I’m somewhat proud to say I made it through to the semi-finals of that week’s Tuesday tournament. Just to be comprehensively beaten by a decidedly older, but remarkably better player. It was so much fun we even tried some artificial-lawn tennis, with a lesson learnt in just how hard this game is. Hats off to Nadal.

Upon returning home, the 2019 edition of the annual CES showcase of all things gadget featured the incredible artificial intelligence-powered ping pong robot from Omron. Named Forpheus, the robot has been much improved this year. Omron clearly uses Forpheus to demonstrate how its artificial intelligence prowess improves over time.

Playing with a ping pong robot:

This is the most important thing about AI. It gets better over time, as the robots learn.

The new AI email writing component baked into Gmail is another excellent example of machine learning. If you’ve been using it and think it gets better over time, you’re right. It does.

Tired of misspelling people’s names upon a reply email to an unfamiliar, difficult to construct first or last name, no worries. Gmail will complete it for you when you start typing “Good morning…” Your fondness for a specific salutation is also easily taken care of.

It starts becoming a bit scary when your email system starts suggesting familiar types of words and even phrases in the body text of mails. On the fly. What I’m most scared of is probably how dependent I’ve already become on Gmail assisting me to write quick, succinct and correct emails.

Our genius head of development at The Toast, Antoni Urban, is an AI expert. He sees it not as any threat, but rather for what is really is: Humungous expanding databases of knowledge and learning that make life easier, when used properly.

The thing about these databases is their unlimited ability to immediately access the correct knowledge, often powered by connectedness. AI is therefore not some sort of inaccessible magic, but rather collected pieces of static data sewn together at lightning speed into something useable. The latest thinking in AI development concentrates on teaching programmes to learn through trial and error, allowing the robots to remember and differentiate between errors and successes.

This is how the machines become better over time. If Gmail suggests a certain word or phrase a few times, but I do not ever use it, the system will learn that it needs to provide a different word or phrase. When I then act upon the new suggestion, it will remember this and offer it again. It does not have to be a specific word or phrase though, it also remembers patterns, style, substance, grammar, etc.

The earliest incarnations of these kinds of assistants were the early spell-checkers. These days spell checking happens as you type. The AI trick comes in when the simple task of looking up a word in a dictionary becomes an intelligent suggestion of how to construct emails that you become better at as time goes by.

My latest interest is in artificial creativity. Call it AC. There is no doubt in my mind that the creative arts are already influenced by predetermined patterns of behaviour. Whether this can be coded is clearly open for debate, but I believe it can. Over time, at least, creativity will become heavily influenced by databases of previously gathered knowledge and experience and by the normal rules of AI trial and error, a certain kind of AC will become more prevalent over time.

Or maybe I’m just dreaming of electric sheep.

The quest for AI creativity: