Trends in Blockchain and AI
An enjoyable chat with moderator from Refinitiv, I elaborated on historical trends in AI and blockchain and how these mega-trends canents' professional and advisor. I made a point of the seminal paper in 2012 which provided a breakthrough in performance of machine learning systems, the vicious cycle of more data, better algorithms, better UX/UI, resulting to even more data - the fuel of modern AI.
In terms investments in AI, my point is that there is no justification to not invest in the biggest commercial opportunity available today. Market research and analyst predict a multi-trillion dollar industry, so it can't be ignored. The entry barrier is also becoming lower with a plethora of resources to get started, at least to a level that one understands the applications, limitations and in general technology landscape of AI today. I was asked about success stories, and we have a lot of them already in so many industries to list to list here but only looking at breakthrouhs in image processing and natural language processing is enough to get a feeling of the transformative power of AI.
However, there are limitations too: a lot of focus on a specific technique which is deep NN -but there are limitations there, such as lack of contextual awareness, a slight change in an image composition could cause maladaptation of the algorithm; data intelligence need to weed out the data you don't need so as not to introduce bias and duplicates. The recent data deluge it is an Achilles heel for modern AI, you need them, but you need good quality data not just big and lots of data. You don't get that much value anyway as you bit the limits of the law of diminishing returns - beyond a point, more data doesn't improve performance.
We also talked about blockchain, of course. Again, from an historical perspective, I pointed out to the origins and early attempts to have some sort of a digital only currency - back in the 90s - a long time before the term blockchain and crypto-currencies were coined. Decentralization, disintermediation, and cryptographic level security are the core elements of blockchain and use cases derive from tokenization of our digital economy, tokens can represent everything that can be represented in a digital fashion.
With my fellow co-panelistsfrom AvidXchange and Constellation Labs we debated trends and new technologies in fintech. I mentioned that AI can act as intelligent assistent for the investments' professional and advisor. It is becoming a commodity these days with successful applications vary from robo-advisors to RPA (robotic process automation) tools. It is an intelligent aid, not a replacement, we will always have human-in-the-loop systems.
In terms of technological trends, I posited that GOFAI (good old fashio AI) and its contribution with semantic technology is coming together with modern deep learning techniques, so that take into account context in our automated machine leanring tools. We are going to need to tackle the issue of explainability as a key prerquisite for AI systems to be applied at large scale in a professional and regulated investments' environment.
An interesting and thought provoking question from the moderator was on what do I perceive to be the most pervasive and mass market adoption solution for the financial services going forward. Interesting question indeed. I mentioned that the writing is on the wall for a number of years that the massive incumbents in the financial industry will have to face the harsh reality of modernaize or disappear as we are getting deeper and deeper in the digital economy. They will need to up their game and modernize and offer value to the community and clients, not only shareholders. In the long term future, I see that the landscape will evolve to one where the financial giants of today will tranform to a different kind of organisation where they will explore and enact new business models, ones that we do not know yet how they will look like. It is unlikely that these organisations will go, but they will change dramatically, from internal operations to product and service offerings. In the era of digital everything, a lot of core assumptions needs to be revisited. And AI, along with blockchain and other emerging technologies, will accelerate that change.
Blockchain 2018 and beyond
With my fellow co-panelist from Oracle and Cognizant, we debated what are the key trends in blockchain from an enterprise applications perspective. I posited that supply chain provides us with more intricate applications that the usual suspect of financial services. Supply chain industry, a multi-trillion industry, has so many different facets and opportunities to apply blockchain technology from a track and trace element, to sovereign identify of transacting parties, fraud detection and prevention, inventory control, just-in-time delivery and dispatch and many more.
A topic that found the panel participants partly disagree as supply chain applications for blockchain need agreement and consensus amongst competing parties, hence not an easy task. However the range of applications is wide and touches different aspects of the supply chain where prior-agreement is an necessity for a network formation.
We also debated the situation with the ICO market, clearly more challenging this year than a year ago. All panelist seemed to agree that we will probably move to some sort of a STO (security token offering) in the future, as the tokenization of our global digital economy is becoming a reality.
Costs and benefits of technology innovation
Some highlights of the interview here:
- AI will keep trying to reach the level of natural intelligence, but it's a moving target;
- Entrepreneurship is easy these days because of the massive ecosystem support and community, resources and access to resources to start a business;
- AI will not replace jobs, rather, it's an opportunity for a skills' refresh, new jobs will be created, but the skills' gap needs to be addressed - access to so many online and other courses and training out there makes that easy
With my fellow co-panelists from R3, Ripple Labs, SVK Crypto, we had a lively debate on what's coming up in the blockchain world. I offered a few areas where progress has been made and could improve the state of practice:
- Regulations come clearer and workable;
- we are moving to an era of "beyond crypto" - we saw a lot of attention, and possibly failure of many projects who raised an insane amout of money using crypto-currencies; however, I see that there is a lot of interesting use cases in applications of blockchain that do not involve a "coin", or a crypto-currency;
- we are moving to other applications of blockchain and DLT, supply chain, different non-financial services consortia appear; different protocols than proof of work – we need protocols with greater efficiency and scalability
There was an interesting and strongly debated topic on the nature of blockchain protocols. I posit that the current versions are characterised by fat protocols and thin applications on top, but that's not necessarily the best way for wide scale adoption in the enterprise world. We need more interoperability between chains – more intelligence at the application layer with regards to Dapps. Ideally, we need a thin but sophisticated protocol layer with easy connectivity to enterprise infrastructure with a choice of home-grown, marketplace or bespoke apps on top.
Finally, as is normal in these sort of panels, what's the single most outlier prediction that could be a game changer: intelligent blockchains
Robo, AI and Data
With my fellow co-panelist from Societe Generale, we discussed our experiences on the use of AI technology, robo-advisors and big data in the financial services industry. One of the topics we explored, was the recent spike in the interest of young people studying, learning and practicing AI. The anticipated new cohort of PhDs in AI and ML can only be a good thing for incumbents and startups alike. As AI is becoming omnipresent in the entire food chain across the organisation, more resources and smart people would be needed.
I also opened up my experiences on interesting applications of AI in the financial services, with robo-advisors be at the top of my list. Their impact on consumers is big and opens up institutional grade investments to the masses at low cost. On the institutional side, I elaborated on the use of and emergence of alternative data. Typically these salient data, or sensory data were not a core part of investment algorithms. But their increased availability and our capacity to process big data with ML algorithms combing through to find interesting patterns and correlations, the use of alternative data is getting popular.
As with most recent AI experiences, the audience is understandably concerned about the job factor: is AI taking over or not? I made it clear, that in my experienc is all about a symbiotic relationship where the AI system lives alongside the professional investments experts and help them do their jobs better, faster and more efficient.
Big data is also in need of more diligent engineering; especially as any bias in feed data for our AI-driven systems, result in unexpected results. There needs to be a thorough data engineering, wraggling and preparation before the data we collect can be of lasting value for our AI algorithms. An interesting question from the moderator on the single largest risk when an AI system advices on an investment idea or product, got both panelist agree that we need explainable AI systems that provide a clear rationale of why a particular investment recommendation is made and what where the logical steps that led to that recommendation. This is also a regulatory requirement these days, so needs to be catered for when deploying an AI-driven investments' recommendation service.
Finally, I elaborated on the interesting use cases for AI in the asset management space; and pointed that distributions (or marketing in other industries) can be dramatically transformed by ML techniques which provide predictions for propensity to buy, likelihood to churn, hence redeem their positions in a fund, identify better opportunities for cross selling and in general provide a more pro-active platform for interaction with an asset managers' end clients. I also pointed out to the emergence of prescriptive analytics, powered up by deep neural networks and cognitive engines, that identify not only what might happen (predictive analytics), but if an event happens, what is the best course of action - this is something core in the early days of AI and planning; but the new AI techniques make it more efficient.
AI at work
Having spent over 25 years in Artificial Intelligence, I was invited to share my experiences with an eager audience of business people wanting to understand where the hype ends and reality kicks in. AI is no stranger to hype. We've been trying to build an artificial intelligence since the mid 50s and progress has been remarkable over the years. Hence, the excitment recently with the advances in computational power, availability of data for training and new and improved algorithms for machine learning. AI is finally becoming of age and finds its way in many industries and market verticals.
In my talk, I gave a brief historical reference to AI - surprisingly, a lot of young engineers and researchers in AI tend to ignore or simply don't know much about the origins of AI.Their only touch with the field is through the popular deep neural networks machine learning frameworks. But there is a lot more to AI than that. I then gave a synopsis of three simple, but powerful examples of using AI in business settings: a market analysis tool using data analytics and statistical inference techniques, a RPA (robotic process automation) use case for back-office function in the financial services sector, and an image recognition application for a waste management service in heavy industries. I pointed out to dangers of hype in AI and the importance for business audiences to be able to understand the limitations but also opportunities of AI. Some are obvious - such as does the AI system solve a real problem vs. just been trendy - others are a bit more subtle and require some technical expertise, such as how to identify and overcome bias in training data.
I advocated to the audience for a use of a toolbox approach when dealing with AI today. It is important to get organised, educated and been conversant in AI terms. Luckily, this is not as hard as it used to be in the past. There is an incredible wealth of knowledge out there in the public and resources on AI are in abundance. You just need to have an organised way of getting around it; hence the toolbox metaphor. There are elements of AI technology which are good for certain domains and tasks and others are better at different tasks. Knowing what it works, where and how is key to successful AI implementations.
Blockchain and AI
I was invited to appear at a panel on blockchain and AI , at the Blockchain Week in London, January 2018.
With my fellow c0-panelist from FlyingCarpet, we debated on the uses and potential intersection of blockchain and Artificial Intelligence (AI). My long serving history in AI - involved with researching, developing, deploying and using AI systems since the 90s - and my enteprise explorations of blockchain in recent years; enabled me to identify the following areas of intersection and synergy between blockchain and AI:
AI and be used to improve operational efficiencies in blockchain(s):
- Blockchain's consensus mechanism are slow and expensive to run (e.g., the popular proof-of-work (PoW) algorithm requires a lot of compute power, which translates to consuming enormous amounts of electricity in order to validate transactions and achieve consensus across all participants in a planetary scale network). Although progress and explorations of different, less energy intensive mechanisms have been explored, PoW remains the dominant choice. AI can help with optimizing the path to consensus and calculate (semi-) automatically the least compute-heavy, hence, less energy consumed, path to consensus.
- AI can also help with smart agents - the birth place of multi-agent systems and topologies of agents have been explored in AI since the 90s - tasked with orchestrating actions between them to achieve federated consensus at the enterprise level. This could cut the coordination time between nodes and more effectively managed network resources to validate transactions.
- AI use in blockchain can have an impact on cost, as optimized blockchains would result in more spare capacity, which then could be used for other operations or re-directed to power up AI blockchains (GPUs from mining to DNN training)
- AI's role in blokchain smart contracts is big and sadly hasn't been explored yet. AI and smart go together. So AI does have a role in smart contracts. For example, AI can enable more sophisticated triggering mechanisms for smart contracts, which go beyond externally sources trigger points through vetted oracles. There is little of no automation in monitoring and interacting with a live smart contract; AI can help there with continuously learning best execution paths and adjusting trigger points to suit the contextual changes in the surrounding environment.
- AI's role in redefining multi-modal interfaces is big. Lately, the advent and maturity of virtual agents, like chatbots, is only a first demonstration of the different modalities that could be the next user interface - go beyond keyboard and vocal commands. Yet, developing and interacting with blockchain(s) and smart contracts on-the-fly, is still done at the programming environment of software development apparatus. An AI-driven multi-modal interface could help lower the barrier of interaction with blockchain and empower new innovations to emerge.
But, Blockchain(s) can also be used to improve AI:
- Explainable AI: it's a commonly acknowledged issue with modern machine learning (ML) techniques that they suffer from a black box syndrome - too many hidden layers in mathematically elegant, but obscure to human eye networks, makes it difficult to understand and visualize rationale for decisions made by AI systems. This has a knock-on effect on trusting the AI system, regardless of the correctness of the results. Blockchain(s) can help there as their core characteristic is that of immutability and track and trace of verifiable audits. If we are in a position to record all the high level steps followed by AI algorithms in an immutable and traceable blockchain, we can then improve the explainability of the system and be able to go back in time and explain any decision ever made by that AI system. This could have an impact on the visibility of underlying algorithms and reduce training data sets bias;
- Data: most state-of-the-art AI systems rely on gigantic data feeds, the scale of which makes them affordable only by the very few tech conglomerates. But blockchain(s) can help federate and demoncratize the data sourcing for AI systems. More and better vetted data, federated and contributed by many, not selected few, resuting in better models and results as a consequence and better AI experience;
- A bit far stretched as an idea and mostly a thought experiment at the moment, but we could see the emergence of "blockchain AIs", that is, AI systems that they themselves live on blockchain. Therefore, are by the nature of blockchaion, immutable, shareable, decentralized – no central AI authority control.
- Tokenized AI: tapping a little into the inner mechanics of modern ML, blockchain technology can help tokenize the rewards' mechanism for emerging ML techniques, like reinforcement learning. Typically, these are done at the programmatic level, before the AI system is shipped out, but blockchain(s) can help tokenize the entire reinforcement learning mechanism, thus providing new innovations and economic incentives for a tryly distributed ML environment.
Blockchain and Energy
With my fellow c0-panelists Sweetbridge, Fifth9, rLoop, we elaborated on innovative uses of blockchain in the energy sector. It appears that blockchain can benefit a lot more industries than the traditional financial services use case. Energy, especially consumer energy consumption is a key area of focus for blockchain. I made the point at the panel that transactive energy is getting a lot of traction within the blockchain community and solid use cases emerge from both side of the Atlantic.
As I have had personally involved in large scale experiments with blockchain technology to make the utopia of decentralized, transactive energy trading powered by smart contracts, a reality, I backed my position with evidence of commercial interest in this space from some of the oil and energy supermajors I worked with. The future is bright for everyone with lots of (cheap) energy to keep us going and a green planet!we debated over a number of issues that shaped up blockchain in 2017.
One of the most intriguing use cases are where you can deploy smart contracts at the edge of a network to allow and empower prosumers (producers and consumbers of energy) to transact energy with fellow prosumers. This is no longer a nice idea, there are projects and early implementations of these systems at large energy companies and startups.
Blockchain State of Play
I was invited to appear at a panel on blockchain state of play in July 2o17, at the Blockchain Conference in Washington, DC.
With my fellow co-panelists from IBM, Consensys and EOS, we debated over a number of issues that shaped up blockchain in 2017. I was with Invesco at the time, representing the fund management industry's view on blockchain. It appears that following a number of years with experimental proof-of-concept (PoC) applications, in 2017 we are likely to see the first production systems in the financial services that use some sort of blockchain technology.
Post-trading settlement and reconciliation services could be made more efficient and effective using blockchain to shrink the trades' confirmation time to near real-time, when the business case to accommodate that is in place. We are also looking at blockchain technology to assist with internal operations, like for example proxy voting registry on an immutable, auditable and traceable blockchain backend.
I also pointed out to the unique characteristic of blockchain technology being a community-driven technology with the world's largest consortia built around blockchain: Hyperledger, Ethereum Enterprise Alliance and R3 the most populous and well known in the financial services industry. A common practice among incumbents in this space is to collaborate and work out nuances and improvements in the technology together with peers rather than working in antagonistic silos.
In general, my statement at the panel and to DC audience was that blockchain technology is maturing, and maturing nicely. We shall not rush things and take time to evaluate properly the benefits and cost of using blockchain as this technology will be around for quite sometime, hence, the best strategy is to take a long term, strategic view and put blockchain to work on areas where it's most applicable: speeding up confirmation of transactions among parties that have not transacted before, decentralize heavily concentrated processes at the core and build more distributed and fault tolerant systems; and evaluate the efficiencies of programmable, self-executed contracts, also known as smart contracts in blockchain jargon.