Artificial Intelligence – the here and now

Would you use an Artificial Intelligence? 

Siri and Google Assistant AIs are found on most smartphones.  They perform a small number of tasks and are getting increasingly good at them.  Speech recognition has come along way – though I still have the occasional argument with Siri.  AIs are accepted as useful  tools in our society.  A recording of a Google project made the news, where an AI, pretending to be human, used a phone call to book a table at a restaurant – complete with ums and ahs.  But rather than being useful, I thought this was a bit creepy – I would have preferred a more mechanical voice, so that I knew it was a machine.

So what is Artificial Intelligence, really?

At its heart, AI is a statistical model and whole lot of data.  The system has been trained, using thousands of examples, to build a model that can make a prediction when given data that hasn’t been seen before.  Think about digit recognition at Australia Post, for recognising postcodes. That model was trained using millions of handwritten digits so that it can now accurately predict the postcode that I write on postcards.  Importantly, no one has sat down and written an algorithm that details how to recognise a “3” – this learned ability to classify a digit between zero and nine comes from the trained model, not a coder. This is the distinction between AI and other software.

Facial Recognition, By National Institute of Standards and Technology via Wikimedia Commons

Why this recent boom?  

What was it in last 10 years that has made Siri capable of understanding my speech?  Although statistics have been around for ages, and some AI concepts date back to the 1950’s, the rise of AI came about when three things happened:

  • the internet made collecting and accessing large amounts of easy
  • storing this information became cheap
  • GPUs – graphical processing units  

GPUs were originally made for video games and animations, but they found another use – making large scale computations on data very quickly.


There has been a lot of investment in AI.  Many startups and many Google and Facebook projects.  In the last month a Machine Learning conference that would normally be attended by a few hundred devotees, sold out – more than seven thousand tickets in 12 minutes!

How can AI help me?  Will AI take all the jobs?

If your job is to read postcode digits, then perhaps your job has already gone.  Other job areas are ready to be assisted by, if not replaced by, AI. If I want to file a patent, I must check that my idea has not already been patented.  This means checking not just English language patents, but also Japanese and German patents. If you have used Google Translate, then you know that it can give adequate translations, but they are by no means perfect.  Translation is a difficult problem and an area of ongoing research.  The AI behind the language translator was given large numbers of texts that occur in both languages.  When the models are trained, they spot words like “grass” and “lawn” that are used in similar contexts, and make an association between the words.  Without being told the specific rules of English (or German), they can scan patents looking for these associations. I’m not sure that scanning through patents is the most interesting job in the world – so perhaps this is no loss.  However any job which is repetitive is a candidate for AI to play some part.


What if the texts that are in  both Japanese and English that we used to train the translator, contain gender bias?  Or racial bias? Google translator famously translated “She was a professor, he was a babysitter ” into “she was a babysitter and he was a professor” – switching the genders.  The model learned the associations (along with the bias) about who is more likely to be a professor.

Artificial Intelligence is not some scary unavoidable destiny.  Like an intelligent dog, AI is what we train it to be.

Jakub Hałun via Wiki Commons