top of page
  • Writer's pictureNathan Temeyer

Democratizing AGI

Updated: Sep 25, 2023

Building Tools for Global Inclusion

In our journey to harness the potential of Artificial General Intelligence (AGI), one critical challenge stands tall: inclusivity.


Our target audience—micro business owners, migrants, and speakers of minority languages—find themselves on the periphery of global discussions on AGI, largely due to a lack of hands-on experience with this transformative technology.


Tools like ChatGPT are a great start. They've showcased the promise of AGI and opened the door to its potential applications. But to be frank, they're not enough. For many people around the world to truly engage with AGI, experience it firsthand, and contribute actively to global discussions, we need to push the boundaries even further.



Why is this so crucial? Because AGI might just be the most impactful technology in human history. The decisions we make about its development and use will shape the future of our world. Therefore, it's of paramount importance that these decisions are informed by a diverse array of perspectives.


We must strive to ensure that as many languages and cultures as possible have a chance to actively participate in these discussions. This is not just about providing access to AGI; it's about ensuring that everyone, irrespective of their language or location, can shape the way we utilize this powerful technology.


Meet Bizzy

As we begin our search for the potential positive intersections of minority languages, business support, and generative AI, we created our first live experiment: Bizzy.


Bizzy is a multilingual experiment that combines multiple models and methods in a search to find the limits of current AI tools. Our team employed various strategies to assist these models through bespoke prompt engineering, and a web app delivery with a friendly robotic owl to draw users in. We were a little skeptical of what results we might find - everyone has those Google Translate horror stories, and we assumed we'd find something similar, but wanted to test it to be sure.


Internal testing with our multilingual staff and with real users found that as expected, translations were imperfect, but also surprisingly usable in our target use case. Slang words continue to pose a challenge, and we definitely viewed some funny interactions that occurred from unexpected slang usage. However, the AI actually proved to serve a self-correcting function in these cases! Through context, the AI functionality would find sources of error and gradually correct them as the conversation continued. This is basically the opposite effect that we've seen in previous Google Translate tests where you can translate a phrase back-and-forth like a game of telephone until the results reach a comical level. And it was an incredibly encouraging result for the usage of this technology moving forward!


Additionally, some of the common issues with chatbots being overly rigid and constrained were gracefully handled through the LLM chatbot functionality. In our experience with chatbots, we often find users encountering frustrating errors like typing "three" instead of "3" and leading to an error message. Or wanting to go into more depth, but being stuck with only the preassigned content options that may not fit their needs. AI unlocks an incredible degree of personalized flows, and we saw numerous conversations reaching important levels of depth that are impossible with traditional methodologies - and all in the user's heart language!



Further work is required to democratize the delivery of the channel to a system more approachable for our target users. Our delivery mechanism of a webapp proved to be a challenge for many users - highlighting the fact that this problem is far bigger than a simple technology question. Additionally, we have many opportunities to further explore the impact of the interactions between prompt engineering and machine translation. We're only just embarking on the journey in this area!



The French philosopher Voltaire once said - 'The perfect is the enemy of the good.' And our early tests have clearly demonstrated that while imperfect, the results have reached a usable level for hundreds of languages around the world. This exciting development means that unlocking incredibly powerful and personalized business support through AI and machine translation is now within reach - massively reducing costs while improving quality and giving us an amazing tool towards our goal of providing a coach for every business.









Join Us!

We invite you to join us on this exciting journey. Let's work together to democratize AGI, to make it more than just a concept discussed in global tech conferences. Let's bring it to the hands of the people who make up the rich tapestry of our world, allowing them to actively participate in and shape our collective future. Because when it comes to AGI, no voice should go unheard, and no perspective should be left out.


12 views0 comments

Recent Posts

See All

Comentarios


bottom of page