The Future of AI Ethics: Blockchain
How blockchain enables data transparency and ethical AI
Blockchain, Bitcoin, and NFTs have all become “buzz-words” that many people are eager to learn about. However, only 0.5% of the world’s population is currently using blockchain technology.
Blockchain is set to be one of the most disruptive and transformative technological advances in the modern-day, even being compared to the invention of the internet and “the biggest opportunity set we can think of over the next decade or so,” according to Robert Greifeld, chairman and former CEO of Nasdaq. Blockchain has the potential to benefit every industry and vertical, including ethical AI.
What is blockchain?
A blockchain is a decentralized ledger system digitally distributed among different nodes or “blocks” on a computer network used for recording transactions and tracking assets. Once a “block” of data is formed and filled, it links to the previous “block” and becomes immutable or unchangeable.
Along with each node or transaction being individually encrypted, the distributed nature of the blockchain ensures that all network participants have a copy of the ledger. This is especially effective from a security standpoint as the network has to unanimously approve the transaction or “node” before it is added to the blockchain. This protects digital assets from being reproduced infinitely by maintaining proof of record, a phenomenon known as a value-exchange protocol.
Summed up in the words of Adam Draper, founder and managing director of Boost VC and successful angel investor in companies such as Coinbase, Amplitude, and PlanGrid, “The blockchain does one thing: It replaces third-party trust with mathematical proof that something happened.”
Blockchain’s role in enabling data transparency
According to Forbes, not only have we created more data in the last 2 years than in the totality of human existence, but by 2025, 150 zettabytes (150 trillion gigabytes) of real-time data will need to be analyzed.
These large amounts of data are key to powering the machine learning and deep learning models of the future. Monica Rogati, ex-VP of Data at Jawbone and one of the early members of the LinkedIn data science team, places data and data collection at the base of her “Data Science Hierarchy of Needs” pyramid, the most important aspect of data science.
As the amount of data that is ingested by an AI model increases, the model has more input and perspective to understand the story or pattern that the data is showing, which in turn can correlate to the accuracy of the model. From an AI ethics standpoint, this is important, because often the focus of resolving bias is around maintaining data transparency.
Data transparency can be key to determining if a data set should or should not be used for training or testing purposes for a model. Data transparency, a seemingly simple term, entails many dimensions. Data transparency is defined as “the ability of a subject to effectively gain access to all information related to data used in processes and decisions that affect the subject” by Bertino et al. in a 2019 IEEE paper. So for AI models, the relevant data would include: what is the training dataset, how the training was implemented, what was the source of the training dataset and other relevant metrics, and whether or not ethical requirements are met.
The innate characteristics of blockchain make it the ideal candidate for maintaining transparency in the datasets that feed into artificial intelligence models. The blockchain will secure the data and the metadata of datasets by the very nature of its infrastructure and design. AI companies and organizations can directly obtain trustworthy datasets from the original source without having to worry about them being altered.
In fact, Gartner estimates that by 2023, 30% of media used in news will be backed by blockchain to establish the authenticity of the digital asset. Therefore those same digital assets (images or videos) that may feed into datasets would be authenticated, removing deepfakes and associating the asset to the physical world such as tagging the location of an image.
How can blockchain prevent AI from growing unchecked and uncontrollably?
In the 2008 action-thriller movie, Eagle Eye, a government-created AI meant to assist in managing the United States government, according to its own calculations, decides that the Secretary of Defense, sixth in the line of succession to replace the president, is most fit for the role of being the President of the United States. The government AI then proceeds to blackmail two individuals into attempting to end the lives of the six presidential successors in front of the Secretary of Defense. Despite this being a fictional plot, it does beg the question — how can we prevent AIs from gaining too much control and autonomy?
As AI-based robots and algorithms continue to experience unprecedented growth and advancements, it is important to explore methods to put checks and balances for AI-based decision-making processes. In fact, it is vital that all AIs are kept under close human supervision, and no decisions are made solely by an AI.
Current Efforts: Humans.ai
Humans.ai, a start-up aimed at building the first blockchain platform to develop and govern AI at scale, has created a functionality they call Proof-of-Human which ensures 3 main functionalities for AI ethics: governance, consensus, and verification. The Proof-of-Human system uses human biometric data and private keys on the blockchain to ensure that any changes to a particular AI/ML model are verified and still under supervision.
This multifaceted platform also uses non-fungible tokens (NFTs), a non-interchangeable unit of data stored on a blockchain with a unique ID, to track assets such as data, AI’s, and algorithms and their respective contributions. These NFTs’ rights to generated revenue and governance are represented through the platform’s Ethereum-based token, $HEART, an integral player of the platform’s ecosystem.
For example, if someone uploads his or her data inside of an AI NFT such as a facial scan, he or she can create rules as to how that facial scan can be used, such as not allowing the data/scan to be used for deepfakes or any obscene, offensive content.
To uphold these predefined rules, the platform uses a Proof-of-Stake (PoS) concept where there are “validators” who secure the blockchain network. To be a part of the governance, validators analyze the incoming request to the AI NFT and then “vote” using cryptographic signatures based on whether or not they feel the request is in agreement with the rules or not.
Within the Humans.ai ecosystem, anyone possesses the ability to become a validator, as long as they stake a certain amount of $HEART tokens. Staking is when a validator’s vote is weighted according to the amount of $HEART tokens committed from one’s wallet, which are locked in as collateral. Then once the votes are in, the validator will either be rewarded with more $HEART tokens for their time and efforts, or if they are found to be “cheating” or “dishonest” can not only potentially lose their stake, but also be sanctioned and lose their validator privileges.
Blockchain will no doubt play a pivotal role in data transparency and ethical AI. However, it is not only important to acknowledge its shortcomings, but also to recognize the government’s role in determining blockchain’s ultimate assimilation into mainstream AI practices.
Given that corporations are often commercially motivated, it is the duty of the government to take steps to create regulations and policies that are in favor of the average person. In fact, the European Parliamentary Research Service, a research center for the European Parliament, has done its own study and published a white paper in which they recommend blockchain-based solutions to allow companies to comply with its Data Governance Act, a big push by the European Parliament to regulate and govern AI.
Although the future of blockchain in maintaining ethical AI is bright and promising, it is important to proceed with caution. As with any great technological advancement, it is vital to be preventative and take into account what has yet to be discovered and not be blinded by the unique advantages of such a technology. Although blockchain has made strides in the distributed architecture realm, it still has areas in which it has yet to mature including scalability, energy consumption, and implementation costs which make it difficult to justify from a business perspective.
I look forward to seeing brilliant developers and judicious lawmakers working together to allow blockchain to become a de facto standard for maintaining ethical AI in enterprise-scale systems.
Tanishq Sandhu is pursuing his Bachelor’s Degree in Computer Science at Georgia Tech and is passionate about ethical intelligence and full-stack development. To connect, make a suggestion, or learn more, visit Tanishq’s website at www.tanishqsandhu.com.