This site uses cookies to improve your experience and to provide services and advertising. By continuing to browse, you agree to the use of cookies described in our Cookies Policy. You may change your settings at any time but this may impact on the functionality of the site. To learn more see our Cookies Policy.
OK
Dublin: 3 °C Tuesday 12 November, 2019
Advertisement

These are some of the odder training methods used to make smarter computers

Training up artificial intelligence takes a lot of work, and it means putting it in sometimes unconventional situations.

Image: Tilemahos Efthimiadis/Flickr

WHETHER YOU LIKE it or not, artificial intelligence (AI) is going to play a major role in our day-to-day lives.

You already see it in simple ways, on your phone with personal assistants, video games that allow a single-player mode, customer support and security to name a few.

But training it up is another thing entirely, and there are numerous experiments out there which look at doing mundane things as well as humans. Here are some of the more unconventional tests AI has been put through.

Reading fiction (and trashy romance) novels

Of the many players that are working in artificial intelligence, Google is easily the most prominent and it shows in a number of projects. One of them comes from its deep learning section, Google Brain, which did research into helping its search and apps understand and adapt to the way people speak.

How did it do this? By feeding it 11,000 novels - including close to 3,000 romance novels – to an AI system. It then tasked the system by giving it two sentences from the books and asked it to generate sentences that would make sense. Here are two examples (original sentences in bold):

he was silent for a long moment.
he was silent for a moment.
it was quiet for a moment.
it was dark and cold.
there was a pause.
it was my turn.
it made me want to cry.
no one had seen him since.
it made me feel uneasy.
no one had seen him.
the thought made me smile.
the pain was unbearable.
the crowd was silent.
the man called out.
the old man said.
the man asked.

Playing board games against world champions

Google (again) with its AI company Deepmind, took on the World Champion of Go and won 4 – 1. The reason why Go was chosen was because it’s an extremely complex game, with a vast number of moves that could be made as the game progressed.

Many experts felt an AI playing at the top level, let alone winning, was years away from becoming a reality, but now that it’s achieved this, DeepMind is looking further afield with robotics, healthcare and smartphone assistants being considered.

The Week That Was in Asia Photo Gallery Go World Champion Le Sedol played against Google's DeepMind in a best-of-five game. The AI won 4-1. Source: AP Photo/Lee Jin-man

Chatting with people on Twitter

An example of where things can go badly. Back in March, Microsoft unveiled a chatbot called Tay which was designed to talk like a normal teen. Since it was learning from the people it spoke to, it didn’t take long for the internet to take advantage of it. Within a few hours, it was repeating inflammatory and racist remarks before it was taken down.

It returned shortly after but the account was then made private and Microsoft’s CEO Satya Nadella saying “it was not up to the mark“.

Learning to play video games

An area ripe for the taking, gaming is something that many AI have been tested on immediately and for good reason. While you can programme it to know what moves to make, a harder task is making sure it has the skills to learn and adapt to a game it has never seen before.

Another project from Google’s Deepmind, it trained its AI to play 49 different games on the Atari 2600. The researchers set up a rewards system so it would know when it was improving, and by the end, it outperformed humans in 22 games like Space Invaders and Breakout.

Such experiments have expanded to more complex games like Minecraft, research that Microsoft is heading itself.

Games E3 Microsoft Minecraft is one game that is being used as part of AI research. Source: AP Photo/Damian Dovarganes

Creating art

Yet again, Google feature in this after teaching its AI to recognise objects within images which then turned into a way of serving up trippy artwork.  In its quest to look for patterns in an image, it will process an image over and over again which makes it look like it’s hallucinating or looks like it’s on LSD.

Some of the images were eventually sold off in a charity auction in San Francisco where it raised $84,000 (€74,100).

Acting as a teaching assistant in college

AI is already used to help provide quick answers to questions or schedule, but in one case, an AI was used as a teaching assistant in Georgia Institute of Technology in the US.

Designed to work alongside the other eight (human) teaching assistants, Jill started working in January but despite the first few weeks being shaky, it started to improve and could answer questions with 97% certainty and didn’t need assistance by the end of March.

The students weren’t aware they were interacting with a robot until it was revealed to them in late April.

Read: Will this guy appear in cinemas again? Nintendo is certainly working on it >

Read: Android users, this is the quickest way to access your important WhatsApp chats >

  • Share on Facebook
  • Email this article
  •  

About the author:

Quinton O'Reilly

Read next:

COMMENTS (1)