Artificial Intelligence has existed since the 19th century, but the latest iteration — generative AI (GenAI) / Large Language Models (LLMs) — has changed AI output drastically. For the first time, machines are making decisions like humans. Instead of the machine producing predictable results, as has been the case until now, the technology’s output is indeterminate, leading to a new era of trust. During our webinar, Embracing AI: A New Era of Trust, Lionbridge’s moderator, Will Rowlands-Rees, led a lively panel discussion focused on AI trust issues during localization with Scott Schwalbach of Amazon Web Services (AWS), Jane Faraola of Cisco, and Vincent Henderson of Lionbridge. If you missed it, you can watch the session on demand. This webinar was the fifth in a series on generative AI and language services. To view the recordings of other webinars, visit the Lionbridge webinars page. Short on time? Read our recap blog for the essence of the discussion. What Exactly Is the New Era of AI Trust? The new era of AI trust entails the AI engine’s ability to produce valuable outputs and much more. It encompasses who’s using AI, how they’re using it, what the AI system does, and how you can convey this information to your company’s stakeholders. Companies can think about AI trust holistically through the following key aspects: Partner — Which vendors are you choosing to partner with? Can they be trusted to manage AI responsibly? Process — What processes are being followed for localization? Are these processes reliable and working toward your goals? Systems — What technology choices are being made to support localization? Is that technology safe, or is it using your data to train itself? Management — How can you assure internal stakeholders that their content is processed and managed appropriately while tempering their expectations? Where Should You Start? Panelists unanimously agreed that now is the time to embrace AI. It has undergone an enormous change, and one must lean all in to realize the transformative speed and cost-saving benefits the latest AI can achieve. This technology can enable enterprises to reach more people in more personalized ways through additional language variants and ultimately increase profits. However, it is essential to exercise caution. So, where do you start? For Cisco’s Jane Faraola, the first step is to be curious. “The sky’s the limit, and I think people’s creativity is really going to drive what we do with AI going forward,” she said. AWS’s Scott Schwalbach advises companies to be deliberate about using GenAI solutions. “[It involves] understanding where it is you want to be and then working backward from there on a process to get there,” he said. Importantly, panelists underscored that people will always be involved, working to bring AI processes into their systems. How Can You Thoroughly Address AI Trust? We’re introducing a memorable way to think about trust with our TRUST framework, an acronym centered around five key measures: Transparency Reliability Usefulness Safety Timeliness Together, these factors comprehensively address trust within the scope of localization. Transparency involves knowing how GenAI/LLM models are used during localization. It goes a long way toward assuring internal stakeholders that their content is being processed and managed appropriately. Ask: Is your language partner being transparent with you about its use of AI tools? Intervening at the beginning of the process achieves reliable output. Determine what you want the machine to do, prompt the engine well, and conduct test pilots. Iterate, test, and confirm satisfactory results before proceeding to the next step. Use GPT-4 for an ongoing dip-sampling assessment of output segments for added confidence. Our tests found GPT-4 assessments to be more accurate than human evaluation. Watch the video to learn how Lionbridge helped Cisco run a pilot project on its content to help it make strategic decisions. It’s necessary to be deliberate about identifying purposeful projects for GenAI. These projects must be useful and propel your company forward so that you can capitalize on the technology’s time-saving and cost-enhancing benefits. Protect your intellectual property by ensuring that LLMs are not absorbing and using your content for training or other purposes you do not condone. Also, be mindful of the geopolitical domain of your LLM’s data center and its potential biases, especially when using specific machines for regulatory compliance or government-sanctioned models. GenAI/LLM technology can expedite your localization process, but rushing its implementation may lead a company to overlook key success factors. The need to tune the LLM, build a Retrieval-Augmented Generation (RAG), and test it may consume time and resources, impacting how long it will take to see a return on your investment. GenAI in Action: Why It’s Worth the Effort To Get Trust and Processes Right AWS, responsible for creating courseware, uses GenAI to personalize content further and drive better engagement. For instance, AWS now builds eLearning courses in the native language instead of always starting in English. It also customizes the tone of its courses to match its audience, such as by prompting the AI engine to make the subject matter more humorous or technical material less dry. “Does it matter that every i is dotted and t is crossed? Not so much,” says Scott Schwalbach. “Our measure is, ‘Does the student open the course? Does the student take the course? Does the student finish the course? And does that student go to the next course? And even more importantly, does the student evangelize for us and tell others these are the courses they should be taking?” An added benefit beyond improved engagement is significant time savings. By incorporating GenAI and using new processes, AWS aims to reduce the time it takes to build a course from 90 days to two weeks. Trusting Your Language Partner: What Should You Look For? Choosing the right language partner is critical to achieving your goals via AI initiatives. First, be clear on what you are trying to achieve. Once you have identified your goal, ask the following questions during the RFP process to determine the strengths and weaknesses of your potential partner: What technology choices is your prospective partner making? Are you being asked to purchase complex technology, such as a Translation Management System (TMS), that you don’t need? Will the vendor protect your intellectual property, or will your data be used for training? Is the language provider data-driven and able to provide objective, actionable data to drive decision-making around when to use Gen/AI and when to pass? Can the vendor provide data to inform which languages you should address today and in the future? Is the vendor transparent about their processes and putting your needs first? The above considerations will enable you to create a thorough evaluation with the right criteria. Equally important is your ability to identify the appropriate team members within your organization who can assess the qualifications of the partners you are working with or are considering working with. View our TRUST framework cheat sheet for additional guidance on AI Trust. Get in touch Lionbridge is leading the industry in AI implementations, serving nearly 500 customers with tailored GenAI solutions and many more engagements in the works, even during the early stages of these latest technological advancements. Ready to embark on your GenAI journey with a trustworthy partner? Reach out to us today.