What if we could get training data for AI and raise money at a time?

Ko
4 min readJan 3, 2018

I have learnt AI for a while. Naturally I want to start an AI startup.

I believe AI is one of the chances in life because with AI you could develop services which were impossible even a couple of years ago. I think being able to improve something is good but being able to develop something which has not exist before is huge.

But when you start thinking about an AI project or a startup you realize one thing soon…

No data, No fun.

It seems data is becoming king…

Rob May, the author of Inside AI and CEO of Talla inc., even wrote in his email magazine that we may need data neutrality.

It makes me wonder if we need a “data neutrality” movement.

Bradford Cross talked in his blog and youtube that proprietary data was more valuable than models.

It makes sense because in order to deliver not just cool but real benefits we need a specific model to solve a specific problem. To train a specific model we need proprietary data.

One of articles of Economist insisted that world’s most valuable resources was data.

In a nutshell to train AI model, we need data, talents, and computation power. For computation, we can use GPU instances on Amazon and Google. GPU is not cheap but accessible. For talents, we can learn AI by ourselves thanks to great online resources on Youtube, blogs, and Github. But annotated data is difficult to obtain. We need to come up with out-of-the-box solution.

What if we can issue tokens to obtain data?

Here’s my proposal to obtain annotated data for free with blockchain incentive mechanism. You can also access money and acquire early adopters as side effects.

Step1. Issue tokens to obtain data.

Similar to Amazon mechanical turks, we can give our own tokens instead of dollars to data contributors. Data contributors can contribute annotated data to you in exchange for your tokens.

You can get data for free.

Step 2. Train a model

let’s train an AI model. There are many online resources which we can learn AI by ourselves. Siraj’s videos are the best I know.

We can even tweak good models on Github.

Step 3. Release the product

Once you release your product, the token holders can enjoy your AI service with your tokens. Now even though your tokens still may not have any monetary value, users can use it. You already have kind of like eco system.

Token holders are not just data contributors but also early adopters

This is almost obvious. If you contribute data to an AI project, you want to try it once the service is ready. This is one of greatest side effects of this framework. We, AI entrepreneurs, can not just obtain data but also acquire early adopters.

Step 4. ICO

Maybe more than 99% of ICOs could be unintentional scam because most blockchain founders do not have any track records in business and startup.

In non blockchain startup world, founders know developing product is easiest part, user adoption and scaling is much harder. But it seems blockchain startups are focusing only on technology.

So I agree investing in tokens is a joke.

But at the same time I have a few points to believe in ICO itself.

First, the current situation of ICO market is reminiscent of dot-com bubble. Startups raised too much money and they were focusing only on technology. The result was because most dot-com startups never generated any money, bubble just bursted. But the companies after the bubble generate profits.

So I believe next generation of blockchain startups also will generate money.

Second, ICO is clearly democratizing fundraising.

In Silicon Valley anyone can access money if you are talented. Maybe you can access investment in big cities in developed countries too. But in most developing countries or even in rural areas of the US, it’s just hard to access investment no matter how special you are.

If anyone living anywhere can access money like in Silicon Valley, this will be awesome, isn’t it?

And I believe using blockchain to develop AI has potential to generate real startups.

I mean just making something decentralized is bullshit because decentralization itself is nothing to do with user benefits. But if you use blockchain to develop new AI models, you can bring real benefits to users and blockchain can be part of it.

What do you think?

I’m starting an open source project, called Ratel Network, to make this process repeatable.

I will develop ground works for blockchain incentive and UI/app parts for data collection.

I’m looking for partners.

If you are interested in building an AI startup with blockchain incentive mechanism, please contact me from Ratel Network website or connect on linkedIn.
Any feedbacks and suggestions are welcomed.

Here’s live demo and source code of a proof of concept example.

Thanks,

--

--

Ko

I'm a serial entrepreneur. I enjoy AI, UI, and blockchain. I like history and reading too.