Have you felt the need to implement artificial intelligence (AI) in your organisation? There’s no denying that pressure on companies to integrate AI into their internal processes is growing increasingly intense. In today’s rapidly evolving business landscape, there’s a pervasive belief that if a company isn’t leveraging AI, it risks falling behind its competitors. In fact, the mere announcement of AI integration can cause a public company’s stock to rise dramatically, as investors and stakeholders associate AI adoption with forward-thinking and innovation.
However, the reality of implementing Generative AI is far more complex and fraught with challenges than it appears on the surface. It isn’t a perfect tool and can produce inaccurate or skewed content that then gets fed back to AI, leading to a phenomenon called “model collapse”. This article will explore the weaknesses of AI-powered solutions in multilingual communication and consider why scrutiny and critique are crucial to ensure proper implementation of this technology.
McDonald’s failed experiment
The global fast-food giant McDonald’s attempted to automate its drive-through ordering process using AI, aiming to streamline operations and enhance customer experience. But the results were less than impressive. The AI systems struggled to understand accents, dialects and even background noise, leading to a frustrating experience for customers. In one TikTok video, two friends can be seen laughing as the technology mistakenly orders over 200 Chicken McNuggets for them.
However, it’s not as simple as just casting AI-powered products aside because of such errors. The pressure to scale up quickly using this technology is very real and other fast-food companies have had significant success with using AI for their drive-throughs, including Wendy’s and Panda Express. There’s no doubt that McDonald’s will explore other AI avenues in the future.
Yet this cautionary tale reminds us that we are still in the early days of AI-enhanced solutions. If McDonald’s couldn’t make AI work reliably for something as basic as ordering a hamburger, how can we expect AI to flawlessly handle the complexities of our businesses’ multilingual content? Try to ignore the hype and remain professional when you consider content creation that requires a deep understanding of context, nuance and cultural sensitivity – areas where GenAI often struggles.
AI and translation
This brings us to the topic of content generation and translation, one of the most promising yet challenging applications of AI. Advanced language models such as OpenAI’s ChatGPT have allowed businesses to streamline their content creation processes, generating marketing copy, reports, summaries, social media posts and even complex articles with the click of a button. AI-driven translation tools have been promoted to companies looking for quick, accurate translations that can be deployed across multiple channels simultaneously.
However, AI is not infallible when it comes to understanding the subtleties and intricacies of human language. While AI can produce grammatically correct texts, it struggles with idiomatic expressions, cultural references and the context-specific meanings that are crucial for effective communication.
Getting multilingual content wrong due to AI mistakes leads to individual instances of misunderstandings, not dissimilar to what happened with McDonald’s. But what happens when our content becomes so widely produced by AI that the incorrect results get fed back into the source and training materials, perpetuating and even worsening already grave cultural bias or linguistic mistakes?
What is model collapse?
One of the emerging challenges in the realm of AI is a phenomenon known as “model collapse”. This occurs when AI models are trained on data that includes AI-generated content, leading to a degradation in the quality of the output. They start to reinforce their own errors, amplifying mistakes and generating content that lacks originality, coherence or meaning.
Model collapse is not just a danger to the quality of AI-generated content; it poses a broader risk to the entire ecosystem of AI development. As more and more content on the internet is generated by AI, there is a growing concern that the training data used by these models will become increasingly tainted, leading to a vicious cycle of declining results. Some experts have even coined terms like “Model Autophagy Disorder” and “Habsburg AI” to describe this self-destructive cycle, likening it to the way the Habsburg dynasty’s inbreeding led to the deterioration of their genetic line.
This issue extends beyond text to other forms of AI-generated content and is perhaps best illustrated by what happens to images. When AI models are trained on images that were themselves generated by AI, they can start to lose the ability to create realistic or meaningful visuals. The result is a kind of digital inbreeding, where the AI’s outputs become increasingly detached from reality and less useful for practical applications.
Can AI provide a potential boost to human creativity?
Despite the challenges associated with AI, experts argue that the situation is not as dire as it seems. Human editors can review and refine AI-generated content, correcting errors, adding context and ensuring that the final output meets the desired quality standards. This hybrid approach, combining AI’s speed with human expertise, mitigates many of the risks associated with model collapse.
Furthermore, many believe that the rise of AI-generated content will actually increase the value of human-created content. As more content becomes automated, the unique qualities of human creativity – originality, emotional depth and cultural nuance – will become even more valuable. If AI is employed to create the bulk of the multilingual content, human translators and editors will be needed to check, refine and perfect the output, ensuring it resonates with audiences in different cultural contexts.
Integrating AI into your multilingual communication strategy
So, how should you be using GenAI? Implementing AI in content creation requires your teams to learn how to write prompts, set benchmarks for what good looks like, assess the output, consult reliable technology partners and have a clear strategy that actually saves time rather than just shifting the cost from one part of the content generation process to another. Without these, what was initially expected to be an advantage could quickly turn into a disaster, leading to operational mishaps, loss of customer trust and significant financial losses.
With the right approach – one that leverages AI’s strengths while acknowledging its limitations – you can use AI-powered tools to enhance your multilingual communication efforts. It’s interesting to examine where you save time and where you lose it, because whichever way you choose to generate your multilingual content, it must be quality controlled by a rigorous human review process.
If you do choose to implement GenAI in your organisation, talk to our language experts here at Sandberg and make them a part of your solution through post-editing services and quality assurance checks, ensuring that your content is accurate, culturally appropriate and effective in conveying the intended message.
Artificial intelligence, Content creation, Newsletter September 2024
