Large Language Models (or LLMs) are the hottest emerging technology on the block but what tangible difference do they make? There is no shortage of dystopian takes. However, right now the focus needs to be on the societal implications the widespread integration of LLMs hold. Today, LLMs already have the power and ability to transform and improve industries and contribute positively to our society.
That said, we need to use LLMs to address potential issues across industries. While the technology is very promising, it doesn’t come without any perils. For starters, the positive and negative long-term effects LLMs may bring to the table. Below, I walk through different ways LLMs can be used to transform unconventional industries and society as a whole.
Using LLMs to Expand Language Accessibility
When it comes to LLMs, all languages are not created equal. The vast majority of models are only trained in English and other widely spoken languages, leaving out low resource languages, such as Swahili or Icelandic. This language gap introduces an emerging inaccessibility issue which creates a global barrier to truly adopting responsible and inclusive AI.
New efforts, such as Cohere for AI’s Aya, are beginning to introduce more inclusive models that have fluency beyond the world’s most common language, but there is still a long way to go. There are more than 7000 signed and spoken languages in the world, but only 7 can be considered high resource languages with a significant amount of training data. A key ingredient to making the possibility of globally accessible LLMs into a reality is enlisting human expertise within the training and evaluation phases to ensure the accuracy and helpfulness of the models. Human insight helps ensure LLMs use local language accurately, and gives models important cultural context that varies by country and is heavily influenced by language.
By enabling LLMs to produce results in less common languages, we are removing barriers to information, expanding access to critical information and education, democratizing the use of AI and contributing towards a more inclusive and fair society.
How LLMs Can Connect Citizens and State
The second direction where LLM’s will have a profound impact is bureaucracy in the public sector. Recently, France announced an ambitious plan to use LLMs to simplify interactions between citizens and the state. The idea that the public sector could benefit from LLMs applies to every country. Already in 2021, 99% of government services in Singapore were end-to-end digital, and we can expect other countries to follow suit. For a sector known to be slow to adopt new technologies, LLMs present an opportunity to make strides in a short period of time. It is only a matter of years or even quarters before we see LLM-based solutions making public services simpler, faster and far more intuitive all around the world.
Will LLMs Hurt or Help Our Mental Health? It Depends.
Lastly, one of the preferred topics dystopian fearmongers like to explore is the impact of AI on loneliness and psychological health. How many times have you heard something to the effect of: “All this technology is making us crazy” You definitely remember a stellar performance by Joaquin Phoenix and Scarlett Johansson in HER. You might even remember an eerie feeling you had after watching the movie, but have you ever heard of a paper in the Journal of Medical Internet Research that showed how AI start-up Replica helped people curtail loneliness? Probably not, but the results are very promising.
With the arrival of LLMs a whole new industry is now on the rise. Companies like Woebot, Wysa and Limbic are using LLMs to offer therapy, resources and information to better deal with mental health related concerns. To further enable adoption of these solutions in the health space we need high quality, highly specialized data. There is a general intuition behind any machine learning project: bigger and better data brings better results. LLMs are early on this journey but it is clear that engaging field experts to create high-quality industry specific datasets is imperative to improve LLM-based mental health solutions even further.
Of course, every rose has its thorns and no solution is ever perfect. This same concept applies to LLMs. The future of what LLMs look like is very much still up in the air and largely depends on how we choose to implement it. Some things we know:
- In order to increase accessibility and adoption, we need more efficient and affordable models
- To make sure the models are safe, accurate and reliable, we need to make sure datasets used to train models come from expert, reliable and responsible sources
- To ensure the responsible use of LLMs, we need regulations and government oversight
There will be many challenges as LLMs continue to evolve. However, one thing is certain, the impact of LLMs is far greater than help with text editing or code generation.. This technology has the power to transform the ways we learn, teach and organize our societies, as well as how we deal with the ultimate challenge: our human condition and our wellbeing. If we shift our focus away from dystopian fear, and towards how LLMs can improve different industries and societies, we will be able to face any challenges head on and come out stronger than ever before.
About the Author
Dr. Ivan Yamshchikov leads the Data Advocates team at Toloka, a data solutions partner for AI development. He’s also a professor of Semantic Data Processing and Cognitive Computing at the Center for AI and Robotics, Technical University of Applied Sciences Würzburg-Schweinfurt. His research interests include computational creativity, semantic data processing and generative models.
Sign up for the free insideBIGDATA newsletter.
Join us on Twitter: https://twitter.com/InsideBigData1
Join us on LinkedIn: https://www.linkedin.com/company/insidebigdata/
Join us on Facebook: https://www.facebook.com/insideBIGDATANOW
Speak Your Mind