The case for decentralized artificial intelligence

4 min read


Not very long ago, artificial intelligence was a geeky sic-fi fantasy. Today, it’s become an inherent and transparent part of life, and we often interact with it without knowing it. When you Google a term, artificial intelligence scavenges the web for relevant content. When you log in to your Facebook and Twitter account, AI determines the content of your newsfeed. When you’re in YouTube or Netflix, AI suggests which video you should watch. And almost everywhere you go, AI algorithms are shadowing your every move and bombarding you with ads.

But how much do we know about these AI and machine learning algorithms that are working in the background?

Very little. The reason for that is that, AI applications have taken on the characteristics of the centralized environments and organizations in which they’re developed.

And that can be problematic because under the current model, as AI becomes more and more prominent in every aspect of our lives, we continue to lose control over the content we consume and actions we take. That is unless we start moving toward decentralized AI platforms, which promote transparency and give everyone the chance to be an active member of one of the most impactful technologies of our time.

RELATED: What you need to know about the different flavors of AI   

algorithmsThe problem with centralized AI

Companies like Facebook, Google and Amazon depend very much on data hungry machine learning and deep learning algorithms to run their businesses. However, they each have their own walled gardens where they develop their algorithms, collect their customers’ data, train their models and run their applications. The company with the largest data store of customer information will have the greater chance of gaining the competitive edge over others. That’s partly the reason you see some of these companies make acquisitions that are way out of their league.

How can this go wrong? First, users must replicate their data across each of these platforms. You can have totally different profiles in Facebook and Twitter, each reflecting different sets of preferences, connections, friends, etc. If you sign up to a new network, you’ll have to rebuild your digital profile from scratch. Part of the reason users become locked into large platforms is that the barrier of entry to a new one is too high.

Second, the entire data collection and AI application is very opaque. You never exactly know how much data Facebook and Google collect about you and what they use them for. Your only option is to hand over your data and trust them to use it wisely. If Facebook’s AI algorithms use your data to feed you fake news or manipulate your feelings, you have no way of verifying. Facebook won’t tell you why it makes specific friend suggestions because it considers its algorithms its trade secrets. Likewise, Google might decide to start manipulating your search results in ways that improves its own business, and you’ll likely never find out about it.

RELATED: How obscure artificial intelligence algorithms are hurting users

Innovation too suffers from centralized AI models. As long as data and algorithms remain behind the guarded silos, companies and organizations can’t cooperate. Every new company that enters the field must reinvent the wheel, create its own data store and develop its own algorithms. This will not only put strain on users, who must give up more of their privacy and data, but will also create a lot of redundant work and inconsistencies between different companies.

The innovation problem also gives rise to a competition problem. With AI data being in the monopoly of companies such as Google and Facebook, startups and new players will have little chance of success developing their own solutions. Consequently, they’ll have no other choice other than to integrate into the inaccessible platforms of one of the big tech companies, and hand over their data to those companies to be able to use their AI. This is how platforms such as TensorFlow work.

The centralized model also forces AI applications to remain mysterious black boxes, which give little or no hint about their inner workings. If you can’t predict—or at the very least explain—how a technology will make decisions, you won’t be able to give it more critical responsibilities. This will prevent AI from being able to take on tasks such as driving cars or treating patients without the help of humans.

How decentralized AI can address current problems

The concept of decentralized AI is simple. Imagine a data store that has no specific owner. All the involved parties can contribute to it while at the same time every one of them can use it to train their AI algorithms. Since everyone is using the store to train their algorithms, they all have a stake in keeping the data clean.

Meanwhile, with everything being transparent, the data will be auditable by participating competitors. This will be a great step toward countering the “black box” effect and will also incentivize against using biased data sets.

RELATED:  The ethical, economic, legal and privacy challenges of AI

The problem with decentralized AI is creating a store that isn’t owned by a single party. One solution to this problem is the use of blockchain, the distributed ledger that underlies cryptocurrencies. Basically, blockchain is a database that is replicated across thousands of independent nodes. Blockchain uses cryptography to prevent the tampering of data, and is transparently visible to everyone. Anyone can audit the data contained on the blockchain.

A blockchain data store will have all the characteristics that decentralized AI requires. It will also provide several other use cases that traditional cloud storage doesn’t provide. For instance, users will be able to own their data as opposed to giving it up to a tech company. They’ll be able to port their digital profile to different applications, and monetize it by making it available to different AI algorithms.

However, blockchain applications must overcome some challenges to become AI-ready. Transactions and queries on the blockchain are slow while AI applications require fast access to data and are very compute intensive. Scaling for very large datasets, as AI applications require, is also a problem. But we’re seeing solutions emerge to these problems. For instance, some projects use blockchain to store and update the digital signatures of data while storing the actual data in off-chain storages. Another project named BigchainDB converges the advantages of traditional distributed databases and blockchain to provide a fast and scalable solution.

RELATED: Blockchain’s role in a data-driven world


Some argue that decentralized AI will hurt competition by making everything available to everyone. On the contrary, I believe that decentralized AI will enable more players to compete in the space. For instance, several startups can create a shared pool of data and algorithms to speed up their development of AI applications. As more companies and users join the network, they’ll collectively be able to challenge the might of big tech companies.

At the same time, it will help propel innovation forward by enabling companies to share expertise, knowledge and algorithms. We’ve seen this at work in projects such as Numerai, an AI-powered hedge fund which has created a decentralized market for trading algorithms and has seen great success since its launch.

AI is a fast-evolving landscape, and the way things are moving, we should avoid putting too much power in the hands of a few actors. A decentralized AI industry can ensure that we make the best of innovations while minimizing the tradeoffs.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.