The death of a pedestrian by an automated car in Arizona and the harvesting of Facebook personal data by Cambridge AnalyticaOpens a new window for political clients have highlighted the importance data privacy, risk assessment and the role of trust and transparency in the real-life application of big data. These two developments emphasize the need for the convergence of artificial intelligence and blockchain technologies â€“ bookends of the powerful emerging tech continuum.
Artificial Intelligence: The Important of Audit Trails Opens a new window
Data is only as good as its reputation. If the integrity of data is called into question by an incomprehensible or inaccessible audit trail in an artificial intelligence-powered application, it will be perceived as a risk, whatever the real danger it poses, and adoption of its capabilities will be derailed.
A key barrier in the adoption of AI is demonstrating how a decision is made. Artificial intelligence that runs on the current black-box model will be restricted by mistrust: Users are unlikely to have confidence in things they don’t understand, and corporate leaders won’t invest in applications that provide no evidence of their decision-making.
The disclosure of Cambridge Analytica’s haul of Facebook data and the death of a pedestrian in Arizona in a collision with an Uber autonomous vehicle have reignited public mistrust in what data is being used, how and by whom.
They indicate that without public trust of how data is being compiled and utilized for artificial intelligence purposes, as well as framing news and information, the entire paradigm of big data technology is subject to public approval and potentially at risk at its very outset.
Is there a solution?
A successful audit trail has two demands from an algorithm: data â€“ what happened previously â€“ and a definition of success.
Algorithms are, in crude form, opinions of what defines a successful outcome, embedded in code. The first rule of algorithms is identifying who is in charge â€“ in short, whose opinion determines the definition of success.
Success is also subjective. An imperfect analogy: A parent seeks to persuade a child to eat three different vegetables at dinner. The child might not agree that represents a successful outcome but it’s the parent deciding that three-vegetable-consumption is the successful outcome.
In the same way, the perceived complexity of algorithms is often no more than a marketing trick to leave the impression that algorithms are true, objective and scientific.
Like people, though, algorithms can go wrong, or be born of good intentions yet have destructive effects. A badly designed car that crashes is there for everyone to see â€“ and perhaps more importantly, becomes part of the narrative that determines a product’s success or otherwise.
But a badly-designed algorithm can silently wreak mayhem, and its flaws go undetected for a long time. There are algorithms built by private companies for private ends. So how can trust be engendered in something that’s not well understood?
The power of artificial intelligence lies in having machines conduct educated guesswork on a precise scale that outperforms human ability. It is a probabilistic method, with the end result being machines learning to make decisions based on what they determine to be the most likely reality. If a â€œwrongâ€ decision is reached, the application of more data to the equation allows artificial intelligence to adjust the algorithm, with the aim of â€œimprovingâ€ the outcome.
Past and Present Data Use in AI
Big data has transformed the modern approach to artificial intelligence.Opens a new window In the 1990s, AI research was largely based in academia, generally consisting of a (typically small) fixed dataset from which an algorithm was proposed and then distributed, often in a journal or conference setting.
The work of Microsoft researchers Michele Banko and Eric Brill showed in 2001 that by substantially increasing the magnitude of datasets and shifting the algorithms from natural language processing (small datasets as they comprise fewer than one million words) to memory-based algorithms, error margins hovered around 5% â€“ good enough for real-world applications in some industries.
But the black-box model of artificial intelligence audit trails and algorithms have increased concern about data privacy. In 2017, DeepMind Health, an offshoot of Google’s artificial intelligence arm, developed a digital ledger called â€œverifiable data auditâ€ identifying not just when data is accessed but how and why it’s used.
The Current Value of Artificial Intelligence
The value of artificial intelligence in businessOpens a new window in 2018 is in augmenting human productivity rather than creating entirely new industries. Adding value to existing enterprises by detecting fraud, enhancing the resilience of supply chains and enabling managers to focus on analysis are tools artificial intelligence technology is providing enterprises.
By automating processes that are too advanced for older technology to execute, enhancing business value by identifying previously overlooked trends in historical data and strengthening human decision-making by articulating forward-looking intelligence, AI is now adding valuable support to human functions.
But organizations are facing increasing pressure from regulators and end-users to open their black boxes by making AI processes transparent, explainable and provable. Vendors will need to share previously protected information, and previously incomprehensible AI will need to be explained by the designers and creators of deep learning.
Explanation is hardly straightforward, and new techniques may have to be deployed to reframe them comprehensively. The process of step-by-step documentation and explanation also comes at a cost, making the process slower and more expensive. But if the trade-off is reducing risk and establishing trust, it is a vital next step.
Towards Decentralized Intelligence
While an audit trail is a desirable trait in artificial intelligence decision-making, the convergence of AI and blockchain â€“ two major technologies each with technical complexity and business implications â€“ may be able to reshape the entire process from scratch.
Basically, artificial intelligence is the brain to blockchain’s body. Artificial intelligence’s machine learning methods find opportunity and improve decision-making, adding intelligence and insight (albeit using guesswork), while blockchain automates the verification of the transactional process, providing the necessary integrity and decentralization.
The application of blockchain centers on the facts, while artificial intelligence is about the creative element.
In its simplest form, according to the Bank of England, blockchain is â€œa technology that allows people who don’t know each other to trust a shared record of valuesâ€ â€“ a securely distributed fixed index shared by all members of a network.
Transactions are validated through a variety of mechanisms, but the connection of the blocks means that without network consensus, it is extremely difficult to modify any of the information established within the chain.
As opposed to traditional, clunky audit-trail software in computer security, blockchain has obvious strengthsOpens a new window . But technology-related obstacles and their origin in the technologically-outdated financial services sector mean that pairing blockchain with artificial intelligence will unlock the potential of both.
Artificial intelligence will enhance the effectiveness of blockchain technology in a number of ways: decentralizing the technology for increased scalability, economizing on energy consumption, enhancing security, democratizing privacy issues, increasing efficiency and enabling the capacity to track and sort data in the volumes now available.
Conversely, blockchain can have a profound impact on the development of machine learning systems by helping AI technology to explain itself and increase its effectiveness, cleaning and organizing of personal data to lower market barriers to entry and shrink the competitive advantage of incumbent tech giants. Artificial trust would also be increased.
While the conjoining of these two technologies that currently bookend the technology spectrum is inevitable, it appears to be happening more slowly than expected. AI’s probabilistic, changing nature often hinges on algorithmic guesswork, while blockchain’s permanent, deterministic approach and strength lies in the recoding of reality, not speculation.
AI-blockchain convergence is far from complete, given the number of companies actually working at their intersection. The focus appears to be on working on decentralized intelligence, and slightly less so on conversational and prediction platforms and intellectual property.
But many of these companies, notes MediumOpens a new window , have â€œlarger advisory boards than teams. It might be an early sign that the convergence is not fully realized yet and there are more things we don’t understand than those ones we know.â€