Capturing the value of AI content

In an episode of “Black Mirror”, the hugely popular futuristic television series, soldiers searching for enemies wear a small device with a flashing blue light. When a witness gives information, this Instant Translator detects the language being spoken (Danish) and immediately translates it into English for the soldier. A fluent conversation takes place between two people, seamlessly facilitated by Machine Translation (MT). Artificial Intelligence (AI) at its finest.

We’re not there yet in 2018… but then again, we’re not far off.

Every day, new use cases for artificial intelligence (AI) emerge, showcasing its vast, yet untapped market value potential. McKinsey Global Institute recently published a report with insights from hundreds of use cases in which they estimated that AI could create between US$3.5 trillion to US$5.8 trillion in annual value in the global economy, which is approximately 40% of overall impact of all analytics techniques.

The commercial awareness and subsequent adoption of AI seems to be following a similar path to institutionalization of other (now ubiquitous) developments in Big Data, Predictive Analytics, and Connected Devices. Many of the use cases have focused on top-line-oriented functions such as Marketing and Sales, and in bottom-line-oriented operations functions such as Supply Chain Management and Manufacturing – where it’s relatively easy to visualize and quantify the impact of AI, from personalization to automation.

Great strides have also been made in AI in content-related areas such as Translation Services and Transcription Services – which also can greatly benefit organizations’ revenues and profitability.

In Translation Services, companies like Lionbridge, earn hundreds of millions of dollars by ensuring their customers’ online content is accurately represented in all relevant languages. They traditionally did this via armies of local employees or contractors who could help maintain brand identity and communicating clients’ value propositions with the right local nuances – whether it is in Mandarin or in Malayalam.

This business works because of its cost arbitrage model, and Lionbridge’s ability to source a knowledgeable, localized, specialist workforce which ensures quality. For many, the current electronic alternative is not particularly appealing, whether it’s Facebook’s translate button, or Google Translate, or the Safari translation add-on. Scalability does not reach the quality threshold, yet.

However, things are changing rapidly given the advances in MT by technology companies like Microsoft, which recently claimed to reach a historic milestone – namely using AI to match human performance in translating news from Chinese to English. So it’s not surprising that Lionbridge teamed up with Amazon Web Services to incorporate Amazon Translate, a neural MT service designed by Amazon Web Services.

Other competitors such as SDL have also taken note of this type of technological progress, and have come up with their own combined human/machine translation offerings. Many of these algorithms need not be developed from scratch – there are at least 10 open source toolkits for neural network machine creation that exist online. SDL has focused heavily on Linguistic AI, where they are building algorithms that make sense of masses of unstructured data by structuring the data via the creation of taxonomies and metadata, which in turn help understand, reason and learn in order to help their customers’ top-line performance.

Transcription Services have already shown bottom-line benefits to organizations by making workflow more efficient. For example, Scribble is a virtual scribe service adopted by the Massachusetts General Physicians Organization (MGPO). With such a service, doctors no longer need to write, type or dictate any notes; their conversation with the patient is recorded, transcribed (by medically trained professionals in India) and then returned to the physician for final check with full anonymity.  Scribble is not the only service, others exist as well, e.g. Nuance, (which states that 90% of its transcription work is done via its Dragon software) and Augmedix.

On the positive side, by using Scribble a physician is able to dedicate more attention to the patient – as opposed to talking with her back to the patient while typing into a computer. However, the usually personal, connection-based dialogues may be replaced by more clinical and somewhat awkward conversations. Occasionally, the doctor would have to say, “For the purposes of the scribe, I note that…” in order to emphasize a certain point or give context, much like an author describing a scene or an action in a book.

It’s a win for the doctor in terms of efficiency, but is it truly a win for the patient? The objective was to be more customer-centric, which it was. However, it takes away some of the intimacy of the doctor-patient relationship, since the patient is aware of a third party – the “fly on the wall” who was an actual person, who is listening to private medical history. Even when technology becomes advanced enough to completely substitute this third party with a machine, this potentially opens up another set of issues around data security; what if that information is accessed and altered by unauthorized parties during the workflow?

Ultimately seizing the value of content generated and processed by AI in areas such as transcription as well as translation will depend on 4 important factors. As a company interested in utilizing AI in content operations, you will need to test yourself (or more likely, if you are outsourcing this capability) prospective vendors on:

  1. Customer Centricity The service needs to nail the “Job to be Done”. In evaluating vendors, companies must test whether the vendor has designed a solution with a crystal clear understanding of who their customer is, what they would find acceptable, and what they would not. During implementation, the solution needs to be rapidly tested and iterated, not just in the initial phases, but continually. And very importantly, personalization options need to be incorporated. For example, in the medical field, if the aim is to generate better patient outcomes via better, more intimate understanding of the patients’ ailments while simultaneously reducing physicians’ administrative workload, more thought must be given to what will make the patient most comfortable and free to discuss.
  2. Product Quality Vendors’ AI-generated services need to be accurate, reliable, repeatable and relevant to your needs. Your carefully developed brands are at stake, consequently you should be highly discriminating in selecting one provider versus another. Should you use an off-the-shelf application? Or a custom-built application? Should it incorporate additional services? Can the vendor understand how to classify and interpret slang, bad language, or even emotions that are discernible in the written or voice content? Determine your needs and your quality specifications very carefully and regularly test against the vendor’s deliverables. AI is at its infancy, but there are already several tiers of providers with varying qualities. Below is an example of a base, consumer-level application, the “auto translate” app within the Safari browser, which instantly translated a Fortune Japan article into English. It may be sufficient for some, but unsatisfactory for others.
  3. Responsiveness Speed of Delivery and Service Level need to be appropriate for your business operations. The right solution requires a keen understanding of the type of responsiveness required. Responsiveness comes in two forms:
    1. Speed of Delivery – much like in financial information systems companies such as Thomson Reuters and Bloomberg, where data streaming is characterized as Real-Time, End of Day, or another unit of time
    2. Level of Service – represented by the support your vendor will be able to provide you on an ongoing basis, e.g. customizing the product, or delivering additional services as requested.
  4. Security Your content is private, valuable and a source of competitive advantage –and should be managed securely. Given that every step of the translation or transcription process is an area of exposure for unauthorized parties to gain access, companies should work with vendors that will provide transparency and a full audit trail of all their data, a “chain of custody”. They should have a documented risk and mitigation plan, and aim to comply with the overall principles of the new General Data Protection Regulation (GDPR) which are being set up in Europe, but ideally would apply globally.

These four factors, if prioritized sufficiently, can accelerate the speedier adoption of AI-based solutions in content translation and transcription and provide significant top and bottom line benefits. The Instant Translator from “Black Mirror” may seem like science fiction, but it may be possible sooner than you think.

    Engage With Us

    Join Our Community

    Ready to dive deeper with the Digital Data Design Institute at Harvard? Subscribe to our newsletter, contribute to the conversation and begin to invent the future for yourself, your business and society as a whole.