It’s hard to believe now that computers, the internet and artificial intelligence (AI) in general did not even exist 100 years ago. In fact, AI was only officially founded as an area of research in 1956, and the aspirations for it then were very different from the results being seen today.

Back then, the hope was that one day AI would be created that mimicked the way the human brain works. By achieving this goal, machines could take over many of the jobs and tasks humans do, leaving people to focus on other things – like living the high life, if the majority of sci-fi movies are anything to go by. Although the field of research was novel, the desire to create AI was nothing new, and everything from Mary Shelley’s Frankenstein to Fritz Lang’s Metropolis to A.I. Artificial Intelligence by Steven Spielberg follow these themes.

However, a recent Sydney Morning Herald article by Douglas Heaven reports that AI today is somewhat different from how people envisioned back then. The reason for this is that the human brain is such a unique phenomenon in nature it remains a subject of numerous unsolved mysteries. When humans still have yet to fully understand the workings of the human brain, it seems an impossible task to replicate it artificially.

Artificial intelligence

Computers work differently to human minds and the progress now being seen in AI research is possible because this has been taken into account. As Mr Heaven points out, computers learn through the use of data processing and statistical analysis. By using these as the building blocks of artificial brains, achievements are now being made.

Take machine translations. The most popular machine translators are arguably those provided for free by search engines and social networks and are also known as Statistical Machine Translation. As the name suggests, the system works by analysing endless online documents and learning the language pairs that are statistically most likely. This knowledge is used as the foundation of the translations the machine produces, and the more examples it has of different language pairs the more accurate the results will be.

Although these machines are fast, for now, a professional human translator will always provide a more reliable service than a machine. That’s because Statistical Machine Translation still struggles with language pairs less frequently used online. Of course, developments are being made all the time but for reliable results, businesses should still choose a living, breathing professional if they require a translation.

One area of development that is prompting excitement among AI commentators is that of real-time interpreting. At Microsoft’s 21st Century Computing event, the software giant’s Rick Rashid showcased the latest technological breakthrough made by the company. Taking to the stage, he addressed the audience and a few moments after he finished speaking his words were broadcast in his voice, but spoken in Chinese.

It’s important to note that this process was not as instantaneous as it seemed on stage. Mr Rashid was not ad-libbing but reciting a pre-written speech, which the system had learnt in order to be able to interpret. However, even this early step is to be admired and suggests that the world is closer to seeing the invention of a Star Trek-like universal translator.

This breakthrough is a good example of the sea change in AI development. The bulk of AI research had been dedicated to teaching a computer much like you would teach a child; by establishing rules and expecting the computer to use these to learn more knowledge. Once scientists stopped attempting to replicate their own brains and looked to a computer’s proven strengths instead, they were able to progress with AI. Nello Cristianini, a professor of AI at the University of Bristol, tells the Sydney Morning Herald: “As soon as we gave up the attempt to produce mental, psychological qualities we started finding success.”

Machine translation

Taking machine translators as an example, the computers do not understand language like a human does. Unlike a human, who would consult their mental language rulebook before forming the answer, machines simply work through their databases and calculate the result that is most statistically likely. Of course, this means errors do frequently occur, as the answer is not always the most common one. Processing context is also a stumbling block for computers, which lack the knowledge to understand it.

A case study carried out for Stanford University’s Gendered Innovations revealed that machine translations have demonstrated gender bias as a result of relying on statistical analysis. Whether or not the subject of the text was male or female, systems like Systran and Google were found to “massively overuse” male verbs and nouns. As a result, defendants are frequently referred to as male in translations, and nurses as female, no matter what the context says the gender is.

Another case of the statistical analysis model backfiring was highlighted by Jorge Rivas of ABC Univision’s Fusion channel. He wrote an open letter to Google to draw the company’s attention to the tendency its translator had of changing the Spanish word ‘indocumentados’ to ‘illegal immigrants’. Rivas explained journalists are increasingly using ‘indocumentados’ in articles as it is a more politically neutral term, and by translating it to ‘illegal immigrants’ rather than simply ‘immigrants’, negative connotations were added. Product communications specialist for Google Ricardo Blanco responded: “Since the translations are generated by machine, they’re not always perfect, but we’re constantly working to improve the quality of our algorithms.”

This is just one example of AI noticeably lacking the human touch. A human’s mind is able to process all the information in a document in order to work out whether the subject is male or female, whether the writing is biased or neutral – and the plethora of other points that make up that one particular document. From here, an accurate translation can be produced. The difference between the way a machine and a human translator works is that while the machine will discount an option because it is statistically unlikely, a human knows it only takes one.

AI will keep progressing and one day the world might more closely resemble the sci-fi movies of 50 years ago; with machines doing the work so people can get on with something more fun. However, it looks like it will be a few years before human translators are putting down their pens.