Our Analysts' Insights

Listen to This Blog
 

Are you already fielding multiple requests related to switching from machine translation (MT) to generative AI (GenAI) to reduce costs? Or maybe an executive or two has asked about your plans to integrate artificial intelligence to increase team efficiency? If not, your time is coming. Here are six ways to prepare to handle these queries, based on a recent full-day symposium on generative AI (GenAI) for the Global Leadership Council facilitated by CSA Research.

• Prepare a synopsis to explain what your team is doing. If you are already investigating the use of GenAI, craft a brief and cogent explanation that’s easily understandable for colleagues and leadership teams. For example: Testing a beta version of ChatGPT to produce translation in your target languages and comparing the results to output from current machine translation engines.
• Share the areas that look most promising for exploration. Even if you’re not allowed to experiment with GenAI at work, document use cases that look the most promising for investigation by your team – and others. Note that these tasks require careful prompt engineering.

• Publicize ramifications for lack of appropriate training data beyond English. Colleagues may not realize that only a few languages – Chinese, Spanish, and a few other European languages – have anywhere near the same amount of training data for large language models as English does. Example: Your web or app developers want to implement features that depend on multilingual capabilities of GenAI that don’t yet exist. Be ready to work with them to bridge the multilingual training data divide.
• Help people understand that GenAI tools don’t store images or text. The engines simply consist of hashes and vectors of data. The latter is not a copy of what they are trained on, but simply a copy of probabilities within the text. Ergo, placing a signature on images doesn’t protect them. The squiggly signatures that are produced on GenAI-generated images are not a rendition of original (human) signatures; rather, the algorithms have recorded so many squiggly-type signatures on images that they produce an approximation of a signature. However, the problem is that the image may look like a 100% copy to the human eye.
• Stay informed on privacy, data security, responsibility, and ethical issues. Depending on the GenAI engine in question, data may or may not remain private. Most tools have data destruction clauses. However, there still appears to be a lot of wiggle room across the various engines, so company lawyers will need to review policies in detail to provide the necessary guidance. Examine error responsibility provisions in the small print when using GenAI: if an error occurs, is there any recourse or is it the user’s responsibility?
• Assign someone on your team to conduct tests and monitor trends. Have them cover what’s going on externally to your organization, as well as internally within your own company. Experiment with various use cases to identify which can be integrated into your workflows.

Localization teams have much more experience with natural language processing (NLP) than most, if not all, people in their companies. By preparing to field a growing number of queries related to GenAI, you can leverage that knowledge during internal discussions to raise visibility and strategic value within your organization.