The Missing Link in AI’s Evolution: It’s Not What You Think

Ian King

3 hours ago

My inbox is flooded with Black Friday and Cyber Monday deals already.

I saw an ad for 23andMe’s “Health and Ancestry” Service for $79, and it made me think how far we’ve come in discovering the secrets of the genome.

But we wouldn’t have this technology today if scientists in the 1980s weren’t interested in sequencing the entire human genome for the very first time.

Many institutions took a stab at this, but despite the advancements in DNA sequencing technology, the tools at the time were not ideal for such an undertaking.

No one university, research organization or governmental body could accomplish this task on their own.

And that’s how we ended up with the Human Genome Project.

The project was formally founded by the U.S. Department of Energy and the National Institutes of Health in the 1990s.

It was expected to cost $3 billion and take 15 years to accomplish.

On the government-funded side, efforts extended far beyond the U.S.

It was an international effort with a consortium of 20 universities and research centers across the U.S., the U.K., Japan, France, Germany and China.

And on the private side, companies like Celera Genomics took on the project.

The project was completed in 2003 — two years earlier than anticipated and cost about $0.3 billion less than expected.

Today, the success of this project has snowballed into a field that’s on the cutting edge of gene editing and drug discovery.

The only reason this great leap in science and technology was accomplished in the 90s was because of collaboration.

And now, in the 2020s the field that’s in need of collaborative efforts is artificial intelligence.

But with geopolitical tensions and private interests, it won’t be governments or companies leading this effort — it will be crypto.

Breaking Down AI’s Walls

Training AI models require vast amounts of data.

That’s something that mostly only giants like Google, IBM or Microsoft are capable of.

This has led to a concentration of AI resources in the hands of a few corporations, which has resulted in siloed AI models.

Data is everything to AI. Data is what powers accuracy and reliability in the answers AI gives you.

High-quality, representative data is essential for building effective AI models.

Since these companies compete with each other, they don’t share data.

And of course, that makes sense. Generative AI alone is expected to bring in $1.3 trillion in revenues by 2032, and they each want the biggest share of this market as possible.

But that also means that Google’s AI model could come up with a good answer to one question but Microsoft’s AI could come up with a good answer to another.

But you don’t get to combine the best of both worlds.

Worse yet, this keeps us from taking a very easy step forward with AI development.

Think about it like this: what if you bring a problem to AI and it requires two steps of problem-solving?

One AI model could be great in the first step, but it might not be as good as another AI model in the second step.

This is where crypto offers an obvious solution.

Blockchains provide an infrastructure where a person or a group of people can develop an AI model and allow it to be used by others.

Then based on experience using the model, developers would have an idea of what it’s good at and what it’s not so good at.

Then in true collaborative spirit, they could combine that model with another one that overcomes the first one’s limitations.

There are already crypto platforms that do this.

They encourage and financially incentivize people to develop AI models and are working toward combining such models to produce the best outcomes.

In this way, crypto democratizes access to AI.

You don’t have to be one of these large tech companies to build or operate an AI model.

You could be an independent researcher who goes onto one of these platforms and with the help of other people’s work, come up with your own AI model.

And unlike those tech giants, the data that you use to develop your models will be transparent and easily accessible.

Building Tomorrow’s Virtual Data Centers

But data is not all you need.

You need massive computing power as well if you are to successfully build AI models.

This is another area that hyperscalers like Amazon and Google dominate.

These giants are racing to build out hyperscale data centers so that they can rent out the computing capacity to users who want to use it to develop AI models.

In fact, U.S. data center demand is expected to grow by 10% a year until 2030:

Building out physical data centers that can handle these types of tasks is impractical for everyone, with the exception of giant corporations that have massive amounts of money to spend.

High-performance GPU chips that run these AI workloads alone cost a minimum of $10,000 each, and then you have the costs of the rest of the computing infrastructure that makes use of that GPU.

Not to mention the land needed to build warehouse structures that can fit at least 5,000 servers and all the associated equipment that helps them run smoothly.

However, while a physical data center network is out of reach for most, a virtual data center network is not.

Imagine this — you have a computer with a top-of-the-line GPU or a server rack for personal use loaded with these GPUs.

And although you do use some of this GPU capacity, generally most of that capacity goes underutilized.

So, even with everything that you do, you will only ever use a fraction of this storage capacity — a pretty bad return on the investment you made buying that hardware in the first place.

But what if you could sign up that GPU capacity to a network that allows you to make that unused GPU capacity available to users of that network.

Now imagine there are thousands of people doing exactly the same thing you are.

Together, the network that you all create results in a massive virtual decentralized GPU-powered data center.

Users can then just rent out that GPU capacity for a fee that gets paid to those whose GPUs are being used.

Crypto projects already utilize the blockchain to build such virtual networks.

That means an AI researcher can rent out this virtual data center capacity at a fraction of what they would pay a hyperscaler.

This is yet another way in which crypto democratizes access to AI.

With more people around the globe able to easily access both the data and the computing power necessary, the more unique and diverse contributions to AI development.

And like with the Human Genome Project, the more collaboration that we have, the more likely we are to lead AI into its next evolution.

The steps that the world of crypto has already taken in this direction are why I believe we are on the cusp of a convergence.