Bittensor: The Next AI Revolution - Why Investors Are Betting Big on Decentralized Development

Good morning. If we all got together to play a round of tech buzzword bingo about the last decade, I am sure crypto and artificial intelligence would be on everyone’s tech buzzword bingo card.

It can only be a natural progression that investors are now pouring millions into the next hidden gem - a combination of crypto and artificial intelligence.

Or better - Decentralized AI.

Crowdsourcing technical challenges where everyone is rewarded for their contribution sounds like a wonderful idea, so I became curious about how it could work.

As my excitement erupted, I started to write this week's post. I hope you are in for a technical chat that will reveal the potential for a decentral AI market where everyone can find opportunities to build and make money.

5-minute-read

When one looks at emerging companies around the AI (LLM) revolution, one can’t help but notice that the landscape is already dominated by a few tech giants planning to own it all.

What’s missing in this landscape is the impact that small, independent developers can have. They can't bring their ideas to the table if they don’t work on one of the big systems.

New technologies and successful disruptions usually have one thing in common: They help democratize. They can attract contributors if they enable them to do something they couldn’t before because only a select few had control.

In the case of AI, developing good AI models is controlled by a few companies with access to funds and resources.

But imagine AI becoming a decentralized network where anyone can contribute and work together to build the best models.

This is the vision of Bittensor, a growing platform that aims to rewrite the playbook for AI development.

The Problem: AI is already dominated by a few

Developing AI models requires immense investments in compute and datasets.

Thus, a few large players emerge and drive the development of AI models. With these high entry costs, the select few that are enabled to train and develop AI models remain an exclusive country club for the companies that can afford it.

Furthermore, the different AI models can’t learn from each other, and each will have its own weaknesses and strengths. Thus, each has to improve independently, limiting their quality and what they can do.

Let’s assume you can get the money together for the hardware needed… Don’t assume you can take out your credit card and order… You will hit the next barrier—the limited supply of GPUs. Large AI companies are trying to control compute and make life for new entrants extra hard.

Other challenges are:

• Limited transparency

• Limited collaboration

• Bias and limited diversity

• Decision-making dilemmas with AI power in the hands of only a few

The Opportunity: Cryptocurrency as a motivator

What comes to mind when you hear the word decentralization and the relentless endeavor to overcome the authorities of a central few? It’s crypto.

Crypto is still early in the adoption cycle. However, we have seen that organizing distributed systems and resource networks around it can be a good idea.

The motivator here is that any entrepreneur with ambition and ideas around artificial intelligence can use a system where everything is accessible in a decentralized network of compute, data, storage, predictions, and models.

If we shift machine learning into a tradable commodity, we can shift control from big corporations to a wider community and hungry entrepreneurs.

Meet Bittensor.

Sidenote: More companies are entering this field, like Gensyn. However, I focussed on Bittensor here because I found their documentation better.

Bittensor overview

After their central pivot in March 2023, AI researchers Ala Shaabana and Jacob Steeves focused on developing their blockchain as a mechanism to incentivize a global network of ML nodes. This is the underlying fuel for the decentralized approach to AI development.

Each node can train and learn; adding additional resources increases the network’s collective intelligence and computing power.

Let me try to make it simpler:

It’s like shared task solving…

So Bittensor is like a giant brain trust where shared knowledge and collaboration rule. The system gets smarter than any single system could be on its own.

The secret sauce

The last piece - why would anyone contribute to a shared, decentralized network?

Bittensor rewards contributors with cryptocurrency when they share what they learn with each other. Every developer contributing to the system and sharing their models gets a financial incentive.

For this, Bittensor developed the TAO Token Ecosystem. TAO is the decentralized currency provided to subnet contributors.

So, is it an AI platform or a blockchain?

Bittensor has one blockchain, and many platforms are connected to it. These platforms are called subnets.

To be clear, machine learning is not done on the blockchain. The blockchain rewards people who contribute computing power and train machine learning models.

What are subnets?

A subnet is like a market. Anyone can create or participate in a subnet. In a subnet, you can be a miner or a validator. You provide your computer and wallet. Given that your computer has sufficient computing resources, you can install the validator or miner module.

Every 12 seconds, a new TAO is minted and distributed amongst all subnets based on each subnet's performance.

Can everyone be a miner or validator?

It depends on performance. Requirements for mining and validating can vary widely between subnets. And since it is a competitive market, subnets might remove poor-performing contributors.

Bittensor rewards based on the quality and accuracy of their AI outputs. “Proof of Intelligence” is the consensus mechanism.

How does it work end-to-end?

Subnets can focus on specific fields. Subnet examples are:

• Text prompting

• Image generation

• Prediction

• Music generation

• Healthcare

Subnets cover the long list of use cases already offered by the leading models. The full list of existing Subnets is on Taostats.

The Bittensor API communicates with the Bittensor platform. This is where the Blockchain nodes sit, responsible for executing the consensus mechanism (“Proof of intelligence”) and doing the administrative work.

If you are interested in going even deeper into the concept, please check their documentation here: https://docs.bittensor.com/questions-and-answers

Challenges and Future Adoption

• Security: A recent hack forced the team to yank the cords for the network. The breach affected multiple wallets. The network was put into “safe mode” on July 2nd. The TAO price immediately dropped by 15%. Security incidents like these can damage investor confidence and hurt the network.

• Adoption Barriers: As with all nascent technologies, Bittensor must overcome skepticism and the dominance of established systems.

Use Cases Summarized

Decentralized AI Training

AI model developers can contribute to decentralized training models. They don’t need to rely on a central server; instead, AI models are improved across a distributed network of nodes.

This is the biggest benefit for me. It reduces the need to build out massive centralized infrastructure before one can get started, massively lowering the entry barrier.

Real Monetization Opportunities

Anyone who contributes can be compensated for their help in improving AI models via the native tokens ($TAO). A node in the network can earn tokens by contributing compute power, data, or training.

Cross-collaboration

Contrary to the big models (OpenAI, Google…, etc.), owners of models in the Bittensor network can exchange knowledge and train models together. This can benefit overall model improvement, accuracy, and robustness.

Sharing and Reusing AI MOdels

Pretrained models can be shared, allowing other developers to help fine-tune them for specific tasks or adjacent tasks. I think this is a big benefit for the network. Developers can build on top of existing work, which saves time.

Democratizing Access to AI

Smaller companies or individuals who see an opportunity but lack the computational (or monetary) resources can access models that solve their problems. There is no need to build an AI sub-division for it.

It levels the playing field and will help drive more niche AI use cases faster.

AI Model Verification

The entire process of developing the model, training it, and continuous improvements can be audited because it is tracked. This makes it very transparent.

It lets users easily validate if the models align with expected technical or ethical standards.

Unlimited Amount of Applications Across Industries

You can take any industry you want: finance, robotics, transportation, healthcare. They all can benefit from a decentralized model, in which different institutions contribute their specific data to create a diverse and powerful model.

The kicker: The institutions wouldn't even have to reveal the dataset they use to train the model because it might include private and sensitive data. Just think about the health industry.

Key takeaways and benefits

I can see that decentralized AI development is becoming a big movement. Although there are still challenges to overcome, we found several key benefits:

• Democratization - A low barrier of entry lets anyone contribute and earn rewards for their contribution of compute, data, or AI models.

• Reduced bias - The point above enables a wide range of views - this clearly reduces the bias that a single entity would bring into an AI model.

• Competitive environment for better quality - quality and accuracy are rewarded, which leads to high-quality work. The ecosystem will be based on consistent improvements and innovation.

Instead of relying on a select few companies that provide limited access to needed compute, it would create a free market over all existing compute.

Contributors to the network are getting paid for unused resources, so even if you are not the greatest AI model builder, your MacBook or gaming PC could still allow you to contribute and make money while you sleep.

Is there a massive opportunity aside from AI?

Aside from AI, I think there could be even more opportunities for solving hard tasks in a decentralized network and getting rewarded.

Isn’t that what we do at our day jobs in the knowledge industry? Is there an opportunity to completely overhaul the employee/employer/contractor relationship? I will leave this open and need to think more about it. Let me know what you all think.

How did you like this week’s edition and slightly technical deep dive?

Have a great rest of the week,