A Framework for Culturally Empathetic AI
All good design depends on the designers empathy, as they must understand the needs and desires of their users. The global market presents a research challenge for Natural Language Processing (NLP), as the nuances of culture are sometimes subtile and deeply ingrained. It’s common to assume that designing for foreign markets only requires a change in currencies and language translation. Sometimes that is enough, however, cross-cultural design can be much more complex and nuanced, and NLP must take that into account. Designing for global markets predates the digital industry, as trade is something humans have been doing for a very long time. It is important to think about how a culture thinks and communicates, because that often determines what they think about your product.
Anyone can understand that a difference in language is a hurdle to consider carefully. Translating is infamously inaccurate, or can miss the point by being too accurate technically and totally wrong in practice. But even within the same language, culture plays a huge part in how we use speech. So when a company wants to sell something internationally, say clothing, they need to account for some dramatic differences in meanings. Most of us know that in America a “fanny pack” is a belted bag you attach to your hips, but in England it is referred to as a “bum bag” as “fanny” is another term for female genitals… And that a “ wife-beater” is just a U.S. colloquialism for a sleeveless undershirt. It’s obvious to see where the confusion could arise here, but cultural understanding goes much deeper than this.
Natural Language Processing is not currently advanced enough to tackle everything. Synonyms, and extraction of entities have been done, but mixing dialects, registers, styles and slang are more difficult. Another classic comedic example is the difference in meaning between “forgive me father for I have sinned” and “sorry daddy I’ve been bad”, the words might mean the same thing, but oh-boy does the meaning change when the register is taken as a whole. Here’s where Artificial Intelligence comes into play. Rather than trying explicitly to figure out the exact model for how to best communicate with all people in the same way, machine learning methods, given a lot of data (like age, location, language, gender), can find patterns and make predictions on how to best communicate with the one person they are talking to. Just as humans know to code switch when talking to different people, we can train our NLP to do the same… with a little extra work.
Since most AI today is build to complete a task, or make our work easier, in order to understand what people need we must first understand where each cultures values lay. Much of modern day cross-cultural design is rooted in the work of Trompenaars, who is widely known for “The Seven Dimensions of Culture” which he came up with after interviewing over 46,000 managers in 40 countries. Rather than distinguishing cultures simply by language, they established seven differentiating qualities.
- Universalism versus Particularism: Do people place higher value on the rules, laws, and dogma? Such as in the U.S., and Germany. If so be sure to provide clear instructions, processes, and procedures. Or do they believe that each circumstance, and each relationship, dictate the rules that they live by? Such as in China and Latin-America. Then realize that the response to a situation may need to change based on what’s happening in the moment, and who’s involved.
- Individualism versus Communitarianism: Do the people value personal freedom and individual achievement? Such as in Australia, and Canada. Then give them autonomy and expect them to want creativity. Or are the needs of the group greater than the individual? Like in Japan and much of Africa, where it may help to allow people to involve others in decision making.
- Specific versus Diffuse: Are peoples work and personal lives kept separate, typical of Scandinavia and the U.K.? Then be direct and to the point, focusing on people’s objectives. If work and home life overlap, common to India, and Argentina, make sure to find out as much as you can about the people that you work with and be open to the topic shifting from business to social interactions.
- Neutral versus Emotional: Do people freely express their emotions, such as in France and Italy? Make sure to use emotion to communicate your objectives, adding value to needs. Or is emotional language limited in business settings, such as in Sweden and the U.K.? Then watch people’s reactions carefully, as they may be reluctant to show their true emotions it’s important to pick up on subtile clues.
- Achievement versus Ascription: Are people valued for what they do, no matter who they are, such as in the U.S. and Scandinavia? Then use titles only when relevant, as otherwise it can come across as “kissing ass”. If power, title, and position matter to the culture, such as Japan and Saudi Arabia, these roles define behavior, so use those titles, especially when they clarify someones status in an organization.
- Sequential Time versus Synchronous Time: Do they want work to happen in a striated sequence, placing high value on punctuality, planning, and staying on schedule, such as in the U.S. and Germany? Then focus on one activity or project at a time, setting clear deadlines. Or do they think that the past present and future are an interwoven continuum, often working on several projects at once, like Japan and Mexico? Then allow people to be flexible on tasks and projects, where possible, realizing creativity is often not a straight line.
- Internal Direction versus Outer Direction: Some cultures want to control nature and the environment to achieve goals, such as in the U.S. and New Zealand. So allow these people to take control and engage in constructive conflict. Other cultures believe they must work with their environment to achieve goals, and they avoid conflict where possible, such as Brazil and China. So let them know how their actions are affecting their environment by balancing negative and positive feedback.
This model can be used to build AI that understands people from different cultural backgrounds better in order to work with them more effectively, preventing some misunderstandings. But its critical to recognize that people are still individuals, and there are many factors that will have a bearing on how to design communication and interactions.
People are such beautifully complicated creatures, lets embrace the diversity inherent in the world, adding richness and depth to our AI.
