On the 12th of March, two representatives from YUZEN Translation attended an event at the British Chambers of Commerce Japan. The speaker was Darren Cook, data scientist, and director at QQ Trend, who has more than 20 years of experience as a software developer, data analyst, and technical director. The purpose of the talk was to fill members in on the current progress and basic workings of AI. The word seems to have become something of a buzzword, recently, thus it was a very timely talk.
AI has been around for a while now, and its very implications have made people nervous about being replaced or even outsmarted by AI. Large corporations have seen the attractiveness of AI and as a result we have seen a tremendous amount of investment in this industry. Indeed, there appear to be many benefits associated with it: from personal assistants to self-driving cars and ships. It is now even possible for us to take an aerial video of ourselves skiing down a mountain thanks to a drone equipped with AI.
A major shift in the development of AI in recent times is the amount of data that is now available and corporate investment has seen this increase significantly. The more information that gets shared, the more everyone gets to benefit, so long as this information is shared responsibly and openly. The speaker emphasized that information should be openly available, so that all can benefit, and that if this were not the case, that it would be costly and a waste of human resources to re-invent.
AI’s basic principles are found in deep learning. Everything gets changed into numbers. For example, as with an image, that image is first converted into three channels (RGB), then each pixel is converted into a number. These numbers are then fed into an algorithm to be analyzed. Words are turned into numbers, too.
The way words are organized is through a set of properties and the relationship between those properties. Basically, machine learning learns the meaning of the words depending on their relationship to others. For example, the king is in the castle and the queen is in the castle. It will know that they have something in common because of the castle.
Human language is difficult. All languages are difficult. People can deal with this. It is always changing because it is forever being re-created. While machine translation has made significant advances over the recent years, the current model does not account for grammar which will greatly affect the accuracy of a translation. Also, when it comes across a word that is unrecognized, it will omit that word. Again, this is also a problem for texts that need to be transcribed accurately, which many do.
AI can be likened to stage magic. The audience is wowed at the magician pulling a rabbit out of an empty hat. But, ask that magician to pull a white dove out of that same hat: he’ll be in trouble. AI is very good at doing what it is programmed to do (this makes it more automation that AI in the philosophical meaning). For it to be more accurate, much much more training data will be needed. However, an infinite amount of data will not necessarily mean improved AI.
While anyone who owns a digital assistant will know the struggles AI has with really understanding language, translators and writers can rest assured knowing that their jobs will not be replaced any time soon. However, as we discussed in small groups towards the end of the session, it was pointed out that although self-driving cars have made mistakes, the proportion of mistakes made by humans has been greater.
The session was very informative and we would like to thank Darren Cook and the BCCJ for offering us this unmissable opportunity.