Ethicalgo

We're trying to ease the impact of current AI disruptions. For example, AI Generated Content has taken the world by storm be it Large Language Models or Art Diffusers. However many people's livelihoods are at risk as many of these models are trained without consent on artwork and AI Art is flooding the internet and reducing the authenticity of human expression. Ethicalgo releases AI models (Sniffusion) to differentiate between content and thus helps to maintain the value of human expression and safeguards people from the detrimental impact of fake content.

AI is a technology that knows no borders so the work that we do is aimed to help everyone. When it comes to artists, our classifier is forecasted to help 17 million artists and even more people who are affected by AI.

We also have other initiatives like:
EXPRIMA: Database of human artists along with certification.
GRAI: Vulnerability disclosure for AI models so that AI can be safer.
AI Axioms: Community-based approach on AI regulation.
APART: Movement to watermark AI generated content for increasing detectability.

Ethicalgo's Sniffusion is a binary classifier designed to detect AI Art from Human Art.

Let's start with what diffusion is.

Diffusion is the most common method used for generating AI images. This is done by removing noise from a random image and essentially "scattering" them until it looks like the prompt the user asks it to make.

Example of diffusion: 

Meanwhile, human artwork is done stroke by stroke.

During the diffusion process, there is some noise which is left in the images and the neural network detects the noise.

The difference between human art and AI art is that AI art has homogeneously distributed uniform noise.

Demo of the Sniffusion model in action can be seen in the MIT project video:
(TIME STAMP: 00:47)

In general, Ethicalgo aims to benefit everyone, but Sniffusion primarily caters to artists so that their expression is protected in a world where human art is losing its commercial value due to the inexpensiveness and speed of AI Art.

Sniffusion affirms the authenticity of human art and detects AI art.

This solution aims to prevent posers from flooding human art forums with AI art and human art can never be mistaken for being the product of an algorithm.

I'm the team lead and I used to animate as a hobby and my team also consists of artists who are the stakeholders of this phenomenon.

We try to get people from various backgrounds so that we have a collective perspective that lacks any bias.

Artists are free to test our network, provide feedback and report any bugs.

We need help with outreach and we believe that MIT has a good network of professionals that can help us to improve our technological aspect and improve outreach. Additionally, we also need help to bridge the gap between the academic community and the general community to have a truly universal solution.

We believe we're focusing on the current impact when the academic community is focused on long term impact like corrigibility and goal-misgeneralization.

Our project's goal is not loyalty to abstract AI ethics but rather an initiative to make people's life better with AI by reducing the negative impact of AI.

We want to encourage other communities and companies to take a more proactive approach when it comes to implementing AI ethics.

We want to change the market and the general approach by placing a more practical emphasis on AI ethics instead of a superficial approach facilitated by other companies right now.

We want to start treading into resolving issues of Artificial General Intelligence within the next 5 years so that we can impact all sorts of people and ensure that AI is for all and designed to do things that are balanced between what we want and what is necessary for us. We want to lead the world and create an ethical framework to pave the way for future AI developments in an altruistic manner.

We're trying to measure the impact by number of people affected, number of supporters for our community and even the impact of our online movements like APART.

Our theory is that if everyone takes a practical approach for AI ethics everyone will start implementing systems to ensure that AI is used for good.

For example with Sniffusion these are the steps we want to take.

1. Separate content on the internet using a solid technique.

2. Now that content is segregates, it gets easier to regulate the impact while human expression is quarantined.

With APART:

1. Protest to large companies to watermark their content and release services to detect whether content is generated or human-created.

2. This will improve step 1. of Sniffusion's objective.

With AI Axioms:

1. Take back control of AI regulation from autocratic companies to the stakeholders - that is the general public so that AI ethics can be defined with the democratic interests in mind.

For Sniffusion, we're using TensorFlow and gathering datasets from several AI image generators and we've created a binary classification neural network to help with the segregation process.

For EXPRIMA, we're using encrypted steganographic watermarks and MySQL for creating a large database of artists.

We're trying to get people, no matter where they're from because AI needs to be unbiased, and hence AI ethics needs to be too.

We're aiming for non-profit. We're currently self-sustained but we might need some donations only for hosting the web-services that we release for executing AI ethics like API hosting for Sniffusion and EXPRIMA etc.

Donations and funding from fellowship organizations. We don't incur a lot of expenses because all the volunteers are contributing for free (bless them) and we don't aim to create any money out of this.

API and server hosting will be under 200$/year when we're fully releasing our services, assuming that we don't start any other resource-intensive initiatives.

We're not really a resource-intensive organization so, so far we don't have any requirements for financial sustainability.