RefugAI (pronounced refugé)

•
8 min read •
by Joseph Perla
#ai#refugees#hardware

Contact: Joseph Perla - joseph@jperla.com - my number is +1.305.951.1381

Problem

UNHCR estimates there are over 100 million refugees today. As climate change continues to cause global issues, estimates are that 1 to 3 billion people will be displaced from their homes this century[1], making helping refugees a huge humanitarian crisis.

Refugees face significant challenges during displacement such as language barriers, lack of access to resources, and difficulty in navigating complex bureaucracies in their new environments. Furthermore, assimilation into their new communities is often hindered by a lack of relevant skills and job training. This situation necessitates a solution that can mitigate these issues and facilitate refugees' adaptation to their new lives.

One AI per Refugee

We envision a small, secure personal device that has a full LLM chatbot inside able to converse in any language with the intention of assisting refugees, with voice in their native language[2], help with navigating bureaucracies[3], help with advocating for themselves legally[4], and assisting them in their new jobs so that they can perform useful work in their new countries at competitive quality[5].

This device is how we can ensure that the most vulnerable among us won't be left behind by Big Tech's AI ambitions. We are a project from some of the top refugee and migrant management organizations and leaders including StartupBoat and Cosmopolis.

Vision

Manufacture and give out 100 million personal, private, offline AI personal devices to 100 million refugees by 2030. Economies of scale will lead to quality and cheap mass manufacturing cost.

These AI devices will not only act as a source of immediate assistance in translation and bureaucracy navigation but also as a long-term tool for self-empowerment. By providing tailored education and job training, we anticipate improved employment opportunities for refugees, thereby facilitating their economic independence. Furthermore, we expect these devices to aid in cultural assimilation by providing language education that bridges communication gaps. This project, thus, seeks to transform the refugee crisis into an opportunity for growth, resilience, and integration.

Device

  • Small handheld device with large black and white screen for text only, in every language.
  • Custom designed Neural Network chip for a 1 billion parameter LLM ChatBot.
  • Shared knowledge base of all useful knowledge will be collected and shared and integrated into the easy chat interface to help answer localized problems for refugees in each country and every language.
  • No keyboard, entirely voice input with a microphone.
  • Secure low-power device, no internet or wireless communication or on-board storage.
  • Camera in the back for OCR text reading input.
  • One USB-C cable for charging and data transfer for updates.
  • Solar pad to USB-charging option for quick solar charging.

Cost

We expect to get the cost to produce each device to as low as $2-$10 by 2029. Half of that will be paid for by the refugee to ensure that it is valuable to them and cared for, and the other half covered by donors.

We have already raised $500,000 and aim to raise another $500,000 for the initial exploratory phases and for prototype development and testing in 2024.

We will open source all chip and hardware designs and software, and the chip can be used in Artificial and Ambient Intelligence applications with many commercial applications to further scale up to billions in production and further lower costs.

Within 5 years we will develop a final product to ship 100 million units in 2029. Hundreds of millions of dollars will be raised from tech companies, governments, individuals, and NGOs in order to cover capital and half the costs to manufacture and deliver the devices for the refugees.

BOM (Bill of Materials) Budget

Costs for the first version will be kept low using the following principles:

  • Use strictly off the shelf hardware.
  • Scale of tens to hundreds of millions.
  • Cheaper materials to achieve target price and serve market of refugees.
  • No or minimal packaging.
  • No moving parts.
  • No wireless or wireless certification.
  • No hardware localization, only one color and SKU.
  • No commercial OS.
  • Distribution and marketing to refugees provided by local NGOs and government partners.

The only possible exception to off-the-shelf components is the neural network chip which will be manufactured and sold to other IoT (Internet of Things) applications in the hundreds of millions or billions to get to scale and become off the shelf. And we also try to convince partner manufacturers (like Qualcomm) to develop that market.

LLM ChatBot

Based on our research, we expect that by 2029 neural network algorithms will be heavily optimized and trained to the point that a 100 million to 1 billion parameter LLM will exceed the quality of a 2023 chatbot. We also predict that, by 2029, a targeted neural network chip of this size, manufactured at scale, will cost well under $5.

Alternative Versus Cloud ChatBot

Price

The leading cloud chatbot has a $20/month subscription, totaling $240/year and it has limitations on use. Our personal device is an order of magnitude cheaper, avoids cost of a mobile phone and wireless connection, is guaranteed to last for years, and has no limits on usage.

The poorest price sensitive users would always prefer the cheaper option, and the poorest would be ignored or locked out due to not being a valuable ad market or subscription market for a cloud AI company.

Reliability

A low power physical device is perfectly reliable, especially with solar charging. The cloud is not, it is dependent on both an electric grid and fast internet.

Refugees are the most vulnerable people on the planet. They often do not have reliable internet connections or even mobile phones or electricity at all. It's not uncommon for mobile phones to fall off of boats or otherwise be lost during migration.

Moreover, the migrant and border countries are most likely to have internet access cut off entirely. One, but not the only reason, is that war and political instability is a cause of both unstable Internet and refugees. For example, looking at the top refugee source countries (Turkey, Ukraine, Iran, Venezuela, and Afghanistan), they have all experienced these Internet and electricity cuts. Internally displaced migrants are also a huge problem (the majority of refugees are internally displaced according to UNHCR).

Finally, specific tools like ChatGPT have been banned and unavailable in the past in both source and destination countries: Ukraine, Venezuela, Syria, Italy, and Iran.

Privacy

Refugees are vulnerable at-risk people, and they also deserve privacy in their communications of personal thoughts with their chatbot.

In addition, many destination countries are in Europe which has the strictest data and AI laws. Many leading cloud AI companies refuse to or cannot legally operate in Europe.

Board of Advisors

We are looking for board of advisor members who can help us with key components and relationships. In particular, we would like to have people with high level experience at

  • One-Laptop-Per-Child executives and engineers
  • Major tech companies
  • AI companies
  • Experience with AI chip design
  • Refugee ambassadors

Donors

Possible donors can include

Why this avoids the many problems of OLPC:

  • Much, much simpler device (no moving parts, no keyboard, no wireless, no OS, no complicated interface). It means less design and setup, arguments, capital commitments, and cost.
  • Simpler device with no moving parts means less breaking and disuse.
  • No wireless means no wireless licensing per region which is much cheaper, simpler, safer, and less risky.
  • Much much (10X) cheaper target price ($10) makes it simpler and much more affordable to 100X more people at the same total capital cost.
  • No training required (trivial text only chat only interface). No training means more active use.
  • No images and no internet so no problems with illicit content, hacking, or security.
  • Does not compete at all in terms of price point, interface, or feature set of budget smartphones or laptops or cloud AI. If there are any, we will partner with commercially similar offerings instead of competing.
  • We will prove market demand for handheld personal AI devices for refugees with prototypes in 2024 before mass manufacturing.
  • No Operating System, so there are no political disagreements about open source versus proprietary operating systems.
  • Internationalized automatically (no localization needed, built into chatbot).
  • These days, much more advanced and standardized global computing and manufacturing infrastructure, materials, and tech components down to charging cables.
  • More specific and targeted niche product use case and global and growing problem (refugees). "Hair on fire" problem. Better product-market fit achievable.

Footnotes

  1. https://www.bbc.com/future/article/20221117-how-borders-might-change-to-cope-with-climate-migration
  2. https://about.fb.com/news/2020/10/first-multilingual-machine-translation-model/
  3. https://ianvanagas.com/2023/03/05/outmaneuvering-bureaucracy-with-ai/
  4. https://www.smithsonianmag.com/smart-news/the-first-ai-lawyer-will-help-defendants-fight-speeding-tickets-180981508/
  5. "The company's least experienced and least skilled workers, according to the research, saw the biggest gains from the AI tools and were able to complete work 35% faster with the tool's assistance." https://gizmodo.com/ai-chatgpt-customer-service-overachievers-productive-1850368672

Enjoyed this essay?

Follow me for more insights on technology, startups, and the future.

Follow on Twitter Read More Essays