Christopher Barnatt asks whether artificial intelligence poses a real threat to translators’ livelihoods
As 2020 approaches, we stand at the dawn of the Cognitive Computing Age. What this means is that a few years from now every type of digital technology will be able to possess internally, or access remotely, some level of artificial intelligence (AI). In turn, this will allow computers to undertake a widening range of cognitive tasks, including complex natural language processing (NLP) and translation.
It is critical to understand that most AIs do not follow hard-and-fast rules. Rather, many of the latest systems are based on artificial neural networks (ANNs). These employ layers of ‘artificial neurons’ that learn to recognise and react to specific patterns of data input. A lot of ANNs also employ ‘deep learning’. This means that their design includes many layers of ‘hidden’ neurons, which establish complex patterns of connection that are very hard for a human being to interpret. In some senses, ANNs function a bit like the human brain, with many able to make decisions by comparing data input to vast stores of learnt or sample data. Future translation AIs may, for example, draw on their knowledge of hundreds of thousands of books in order to deliver the most fluid and accurate output.
Many reputable organisations have attempted to predict the impact of AI. In 2016, Citi and the Oxford Martin School estimated that about 35% of jobs in the UK are “at risk of automation”1. A year later, a report from PwC was in broad agreement, noting that 30% of UK jobs are “susceptible to automation”2 by 2030, although PwC did stress that many jobs are destined to change, rather than be lost, due to the rise of AI.
When considering the impact of AI on employment, we would be wise to focus on the range of work tasks most likely to be automated, rather than the potential replacement of entire jobs. Using this approach, my own estimate is that around 20% of work tasks will be candidates for AI automation by 2030, with these activities spread across at least 80% of human occupations. I would suggest that only a small percentage of people will lose their job to AI. However, the vast majority of us will see our work altered and restructured to some extent, as cognitive computing takes hold.
The advance of AI
Already AI is advancing rapidly and radically. The development of artificial general intelligences (AGIs) able to ‘think’ and act like humans remains a distant pipedream, yet AIs do not need to mimic people in order to automate a wide range of cognitive tasks.
Recently, AI has started to be delivered over the internet. For example, IBM, Google, Microsoft and Amazon now offer cloud AI services using standardised APIs (application programming interfaces) that are turning AI into a pay-as-you-go, plug-and-play utility. Only two or three years ago, anybody wanting to integrate AI into their business needed deep pockets and a technical knowledge of machine learning (ML) techniques. Today, new users simply require some basic coding skills, and either a purchase order or a credit card that will be charged on a per-transaction basis.
The cloud AI services on offer from the goliaths of computing include language processing and translation utilities that can be plugged together like Lego. IBM, for example, has invested more than a billion dollars in an AI platform called Watson. This includes modules called Language Translator, Natural Language Classifier, Natural Language Understanding and Tone Analyzer. These facilitate the interpretation and translation of text in 13 languages, with the AI able to recognise and understand the emotion, sentiment and communication style within a passage of text. Watson has been used to create chatbot interfaces for the Royal Bank of Scotland in the UK, and Staples in the US.
Google indicates how its cloud AI services can be used to “inject AI into your business”. For example, the company’s Natural Language API can be employed “to reveal the structure and meaning of text”, while the Google Cloud Translation API can be plugged into any website or AI module to offer translation in more than 100 languages. Back in its early days, Google Translate was not very accurate. But the self-learning system is vastly improved, with Google charging just US$20 for the translation of 1 million characters.
Microsoft’s online cognitive services are offered from its Azure cloud platform. Current modules include Language Understanding (LUIS), the Translator Text API and Linguistic Analysis API. Again, these can be mixed with other components (including decision-making, and speech and vision recognition modules) in order to create bespoke AI systems.
In early 2018, the world’s largest cloud computing provider, Amazon, announced a suite of AI tools, which included a new version of Amazon Translate, as well as Amazon Transcribe, Amazon Comprehend (natural language processing) and Amazon Poly (text-to-speech services). Amazon Translate charges US$15 to shift 1 million characters between two languages, with the service intended to provide on-demand translation of user-generated content, including “real-time translation for communications applications”. In common with offerings from Google, Microsoft and IBM, Amazon Translate gets smarter over time. It works by processing the meaning of a passage of text, rather than simply recognising and translating single words or short phrases.
Some companies, including Hotels.com, already rely on cloud AI services to translate user-generated web content. The results may not yet offer the polish and accuracy delivered by a good human translator, but sometime in the 2020s it will become impossible for most people to distinguish machine-translated content from documents translated by a human brain. Cloud AI is set to have as big an impact on the translation business as spreadsheets had on the accounting sector some 30 years ago.
Working with the machine
This does not mean that all or even most human translators will be made redundant. Even today, many translators use AI tools to assist with translation activities, and – as in many other occupations – the creation of human-AI teams is going to be increasingly common in the 2020s. Already, the translation, localisation and e-learning specialist Lionbridge is employing Amazon Translate to great effect. As its Chief Technology Officer, Ken Watson, has reported, “Human translators armed with machine translation help companies localize more content, faster, more affordably and into more languages. Based on our experience, pairing Amazon Translate with a human editor, we believe we can produce cost efficiencies by [sic] up to 20 percent.”
Almost certainly, the rise of AI will require human translators to redefine their business in order to compete with or capitalise on the new technology. In many instances, the best option will be to integrate human and machine translation, as Lionbridge have signalled, and to operate as a value-added intermediary. For example, a human translator may use cloud AI services to provide real-time translation of a client’s web content, but with the added benefit of a guaranteed human check and edit within 24 hours.
Other human translators may decide on a niche, non-AI strategy, with their business redefined as a communications skills or cultural understanding service. Just as most accountants now spend little time calculating totals, so in the future many linguists may spend very few hours a week on basic ‘language processing’ activities.
Regardless of which strategy is adopted, the most important thing is to avoid an ostrich approach. Sadly, across all industries, many people appear to be caught in AI denial. Some are clinging to the notion that AI does not exist, or will not exist for some time to come. The creation of general AI does, indeed, lie far in the future, so there is a strong philosophical argument for this. Unfortunately, however, the pragmatic case is very weak, with Amazon et al already selling and developing AI systems able to take on an increasing range of cognitive activities.
Other ‘AI deniers’ have decided that, while AI does exist, it will have no impact on their particular work activity. Some may be correct. I would, however, conclude with the observation that every industry I have ever worked in believes itself to be special, with new technology only presenting a threat – or opportunity – to everybody else.