Welcome Davide. Let’s start from your background and the most significant moments in your career.
In brief, after my first Master in Boston, I enrolled for an MSc in Finance & Investment Management at London Business School, which I attended part-time while working as a derivatives trader in the City. I was one of the first to apply neural networks to trading at a commercial level – Neural networks, or connectionist systems, are at the heart of AI and are computing systems that are inspired by the biological neural networks that constitute animal brains, able to « learn » to perform tasks by considering examples, generally without being programmed with any task-specific rules. Fundamentally, they group and interpret unlabelled data and classify it based on similarities. I had five top-of-the-line PCs – at the time the best available but providing less than 1/1000th of the computing power of a normal smartphone today – to reach a satisfactory computational level and develop the network. It cost a great deal of money compared to today. You could have bought a small apartment in London with that money at the time.
This neural network worked well, and by the age of 25 I was managing a significant amount of money for the time, although not much if compared to today’s $100 billion-sized funds. Still, at that time it was unusual to have a hedge fund trading over US$1bn in derivatives. Eventually, I sold the business and moved to the Centre for the Study of Financial Innovation, where I became Director of Studies. In those years, around 1995, the World Wide Web had just started, and we conducted the first European study on the impacts of the Internet on the financial system.
Later, together with a number of others, we created a Venture Capital firm to leverage the considerable opportunities the Internet and WWW world were offering, specialising in identifying, advising on and building companies in this space. I got out of that just before the 2001 crash, purely by chance rather than by any foresight, lucky I guess, did several years of independent strategy consulting and then ultimately joined CSC (Computer Sciences Corporation), now DXC Technology, focusing on banking and capital markets for Northern Europe.
Tell us about your interest in Artificial Intelligence, which led you to Macrina Investment Management and how the company works.
In 2015 I moved into the AI world. Together with other industry experts we created an advisory board, crafting the basis for Macrina. The company supports and helps AI and Robotics companies grow. We provide both advisory and investment services but do not manage money directly, rather supporting those that do in a number of ways. Many of those involved in this type of investments invest in many firms every year – some times hundreds – knowing that only a few companies they invest in will make enough to cover costs and generate profit. Conversely, we analyse over 600 companies but provide support to only two or three.
Macrina also operates mainly through equity financing, and we are currently helping a very innovative company, Kensai, raise US$1M, a relatively small amount. We have been working with them for almost a year, and they have developed a way to speed up AI computational power by over 90%, resulting in a solution that can automatically generate reports on sentiment across social media, news and private databases – over 1.2 billion data points – in real time. In the next six months, this will also be available for video and speech analysis, thereby creating even further real market opportunities with a number of uses. We conservatively believe this company could be worth in the region of US$250M in 3 to 5 years. It takes a lot to find the really good companies, and it is for the long haul. Nothing comes easily.
In which industries do you invest and what do you look for in your targets?
Macrina does not invest directly. We focus on identifying strong opportunities and management teams in the AI and Robotics space, and more AI than Robotics, to be perfectly honest. They need to have a really deep understanding of AI technically, have a great product or use case for the technology they have developed or adopted, and involve a strong team that can make it happen. The team is essential.
Robotics, on the other hand, represents a more niche investment area, as a robot is a piece of dextrous metal, where physical engineering is just as, if not even more important than, the AI that may control it. Therefore Robotics companies have two challenges instead of one: The engineering and the AI. They are therefore harder to differentiate and represent a greater risk from an investment perspective.
Overall, we seek companies that are addressing markets which are naturally growing, preferably with their technology having potential application across industries, or being adaptable to a variety of use cases and target markets. That way, their target market and product diversification, especially as they grow, is not as limited. Kensai is a great example of this.
However, the key element is always the target’s management team. They have to be good, really good, work well together, have a strong vision and be willing to work hard. They should preferably also have built or tried to build other companies before, even if they failed at it. Failure just means that they will not make the same mistakes again. We like that, as long as it occurred before the venture we would be backing!
For instance, in a few cases, it can happen that we invest considerable time and effort in a company which we subsequently decide to drop, without taking it to investors or engaging longer term as a result of our ultimate evaluation of the management. If the management team doesn’t fully understand the market or have the right vision and approach, the fact that they may have a great technology is not enough. It is generally better to back a good and very hard working management team with a mediocre technology, as the technology can be improved, while personalities, vision and work ethic tend to be harder to change.
From a macro-level point of view, which sectors do you reckon are impacted the most by AI?
AI is currently at a “nascent” status, even though it may not seem it, it is at the “steam engine” stage. Impacts can be notable, yet they remain minimal compared to what my children’s generation will witness. Currently, the biggest impacts are in media advertising and targeting, health care and the financial industry. For example, today AI can analyse X-rays more efficiently than doctors, while pharmaceutical companies have developed in-house solutions to speed up and optimise drug testing processes up to 10,000 times. In financial services, a lot of support is now provided by AI systems for trading and fraud detection, while in marketing and targeting the likes of Google and Facebook are but prime examples of how AI is being monetised.
Additionally, AI is also being increasingly used for customer service and customer segmentation, both of which are hugely impacting on areas like credit decisions, insurance payouts and in general profiling. This is in both risk profiling terms, but also in terms of opportunity profiling for cross and upselling to existing clients. At the retail level there is also an increasing use of AI to target customers on purchasing decisions at “the moment of want”, say when they are looking at a particular store or browsing in a specific mall or shopping centre. This will only increase as AI becomes further integrated into almost all devices, from cars to smartphones, to watches to clothing and all household appliances.
So, are financial institutions being reshaped and transformed by Artificial Intelligence?
Definitely, banks are being “commoditised” in the sense that they do not differentiate anymore on the specific product they offer, but tend to compete on cost. Back to when I worked for CSC, our clients such as Barclays, RBS, UBS, and others were asking us to help in lowering costs in the provision of products and services that everyone had. Those who succeed would retain a certain margin based on prevailing interest rates, market trends and other factors, but it remained that the final total Return on Investment was and is not a variable truly in the hands of banks. The only thing financial institutions can influence and lower is their operating cost.
Therefore, like any specific system that reaches a minimum operating cost, banking becomes a utility, just like deciding which electricity or water supplier to sign with for supplying your house. In this, an AI solution that analyses risk profiles will undoubtedly lower costs both for the bank and their customer. However, I have my doubts as to whether, in the longer term, AI will actually assist incumbent banks. I think it will make them increasingly vulnerable, not just from new entrants with no legacy effects, but to non-bank entities assuming bank-like functions within the economy, paid for by other services they provide or through the better monetisation of data they collect on their customers. Banks only have banking services as an income source.
Many start-ups and fintechs are emerging, some, for example, aimed at providing better lending solutions. Yet, they often struggle to make progress and disrupt the financial world. Why?
Indeed, the only advantage of a start-up is to have lower operating cost through new technology. There are generally no real new innovative and disruptive products to be brought into the market able to incentivise customers to change bank. Banking has been around a long time, and its traditional role within economies is well defined, at both retail and wholesale levels. This means banking services and products are, by their very nature, not innovative. They have mostly been explored and perfected.
At the same time, the core processes of incumbent banks are often based on obsolete systems that no one knows how they work anymore. The documentation and people that initially built them are long gone. This is at the heart of the so-called “legacy effect”. Quite a few parts of code underlying core banking technologies in current use are still written in COBOL, a 1960/70s programming language. Nevertheless, these systems are critical to the bank’s operations and remain unchanged as it is very high risk and potentially disruptive to try and change them. This results in structural higher costs to be borne by incumbent banks.
This is the advantage newcomers – “challenger banks” as they call them in the UK – have. However, it is the only real advantage. The fact that they can package standard banking offerings in new ways, to make them more readily accessible via smartphones or other platforms, or that they are able to react to new market ideas or demand more cost efficiency ultimately comes down to their ability to use newer technology. It is not about changing the underlying function of a bank.
The fact that not many have been successful attests to two issues: one is the fact that they have not done enough to differentiate themselves from incumbents, possibly because it is very difficult to do so and the second is the fact that there is a considerable brand risk associated with banking in people’s minds. When you deposit your money somewhere, you want to know it is safe and the general perception is that if a bank has been there a long time, it will continue to be there and your money will be safe. This is irrespective of the existence of government compensation schemes.
In general, there are opportunities for newcomers to gain market share, especially in the retail sector. For example, Metro Bank in the UK has grown consistently from scratch in the last three to four years. Nevertheless, the fact remains that new technologies carry costs and established financial institutions are good enough to invest and remain competitive. Lastly, big financial institutions can count on a consistent reputational advantage.
How strong are links between governments and companies in developing AI solutions?
This is a really interesting topic. First, we have to take a step back and contextualise your questions: AI suffers from a problem given by the intrinsic competitive advantage it can provide. Let’s take the extreme case of a “singularity”, or the moment at which an AI starts to be able to make autonomous and independent decisions. Some call it the time when computers will become conscious. In brief, AI would acquire the capacity to generate new code to improve itself, autonomously and without human supervision.
Any company or government who is the first to achieve such “singularity” will automatically be able to develop solutions that are inherently better than anyone else as the AI would provide answers and approaches no human mind would be able to compete with. According to a recent study, such an AI could conceivably be able to conduct the equivalent of 10,000 years of research and ideas made by 10 top minds in about a week. This would leave whoever comes second well behind! It is likely that AI, at least at this level, will be a “winner take all” type of scenario. That can be very troubling.
At the same time, achieving such AI would open multiple possibilities, from developing solutions for cancer, to new ways to tackle scarcity and even arrive at solving energy efficiency issues. Naturally, it follows, that whomever owned such an AI, or could claim rights to its inventions and outputs, would ultimately be in a hugely advantageous position compared to anyone else. And it is also the case that those who will own these technologies will increase their advantage over time. As long as they have enough processing power, within a month they would have the equivalent of 40,000 years of advantage and so on. Taking this to its abstract logical conclusion, they would end up owning everything.
Now, it may not be the case that a “singularity” is ever reached, but the same logic applies at “singularity minus one” or “minus 1000”, or whatever. This is what governments and companies around the world are gradually realising, accounting for why both the US and China are now hell-bent on gaining dominance in this space. It is not just a competition for technology, it is a competition that can be ultimately seen as being for everything! Nothing like this has ever occurred in the history of man, and it will be a great challenge to see how it is all managed. There are no precedents.
Technological advancements in this field are already achieving unexpected results. For instance, it was reported that at Facebook two distinct AI programs assigned specific tasks and connected to each other to execute them, started to autonomously create a new language in which to complete their tasks. This was neither pre-programmed or part of the original tasks the AI had been given, but arose as the AIs tried to make what they were doing faster and more efficient. For this, both AIs had to be shut down immediately. What will happen when the AIs control the switch?
As a further example, a third generation AI has beaten the World Champion of GO – a very complex strategy board game popular in Asia. At Google’s DeepMind – who created this AI – work is now ongoing on what could be considered the ninth generation. Consider that each generation tends to improve on a logarithmic scale – so from the third to the ninth, things get better by 10 to the 6th times – the AIs now available in some research areas are already incredibly powerful. So much so that certain general AI applications, given their very advanced level of behaviour, have been withheld from general release by some companies to avoid scaring public opinion. Although this may not make the news, I can assure you this has been the case in more than one instance in the past year alone.
It follows that governments are very interested in AI technology, as is industry. However, the extent to which they collaborate is – for the moment at least – less than one would think, especially in the west. But this is changing rapidly.
In your opinion, which countries are favoured in the global race to AI supremacy?
There is little doubt to anyone who actually looks at the matter in detail – goes to the places and speaks to those involved – that China leads or will soon lead the race in every aspect. Firstly, it has a stated, well defined and centralised government policy on becoming the number one country in AI. This is hugely important as it means the industry and related research institutions are coordinated and can count on high levels of investments, amounting to $150 billion per year, and an unmatched growth rate in the AI commercial sector. By comparison, the USA invested around $70 billion while the EU, which cannot be considered to be competing globally, officially invested only €20 million last year mainly in England and France. It is also the case that no major country in the west has a stated, explicit and clear government policy on AI and how the industry is to be supported or made to grow.
Also critical is that China generates over 300 thousand IT professionals every year, ten times EU levels. With a much bigger population of 1.3 billion people, China is able to produce 10 to 15 times more engineers and computer science experts every year. This is now really starting to show as, for example, in 2018, China published more academic articles on AI than the US, indicating a shift in where academic innovation and research is primarily conducted. In the future, this is also likely to increase, as China is the only country that has introduced AI related studies at junior school levels as part of the national curriculum.
As a final point, China also benefits from fewer restrictions on data, and data is the fuel of AI. Chinese AI companies have a considerable advantage in this respect, and make full use of this by focusing on the implementation of AI to everyday issues, achieving much faster market distribution of AI solutions and applications. The result is that, although China may not be 100% ahead in all measures, it soon will be. And remember, such rankings are always backward looking, so China may actually already be ahead in more areas than are generally reported.
Considering all we have discussed so far, which implications can be derived for humanity and what kind of human – AI relationship will be in place?
The Industrial Revolution consisted in replacing human physical activity with machines, redefining the role our mental capabilities had within economies and society. Our value add within economic function changed from contributing physical strength and physical work, to contributing mental work and ideas. The problem that now arises is that the eventual ubiquitous applications of AI will mean that our mental capabilities are displaced. While we had our minds to fall back upon to add value once our physical strength was replaced with the steam engine and ultimately automation, what do we fall back upon when AI and the advances they will produce displace the value add our minds and ideas currently have? What would then be left for us to do? In the most optimistic scenarios, humans could perhaps just relax and enjoy the arts. I wish it were that simple!
More immediately, we are also increasingly witnessing other problems, such as attributing our own biases to AIs and so, in a sense, exonerating ourselves in the process. This occurs when AI solutions are fed real data and then accused of being biased in their conclusions – such as hiring decisions or risk attributions – wrongly implicating, on the one hand, the existence of an opinion which AI actually do not have, and at the same time freeing people from facing the responsibility they have in such biases. The reality is that data derives from us and contain our own bias – it has nothing to do with the algorithm or the AI concerned. Instead, a better approach we could consider is to ask ourselves: Is our aim to create AI solutions able to reflect such biases, or should we use this feedback to improve ourselves?
Lastly, a central problem modern society is facing is the so-called “income distribution problem”. Since the beginning of the Internet in the 90s, the spread between those who have and those who have not has increased dramatically on a global level and – although poverty has decreased overall – today we witness a structural division in which the eight wealthiest people in the world own the equivalent of the bottom 50% of the global population. Is that sustainable?
Add to that the fact that it will likely be the top 1% of global society that will most benefit financially from AI and its wider adoption. As a result, this gap is likely not only to grow, but to “solidify” structurally, especially in western economic systems, and you have the basis for an inevitable systemic level issue across all political and geographic domains. Current issues arising from mass immigration are but one initial symptom of this. In short, advanced AI applications generate an advantage for those that control, own or can claim the economic benefits AIs generate that will only serve to widen this income distribution gap, with all the social consequences that this can bring. This is a real challenge at the heart of all capitalist systems, and I am not sure the redistributive means available within these systems – or the capacity to develop new ones – are sufficient to resolve the problem. This is actually the top problem of our age, especially in the west.
One last question: which skills should a young enthusiast need to develop to have a better impact in a highly technological industry such as the AI world?
More than ever, practical expertise is required: Being able to program in Python and similar languages used in AI, and to understand the main AI architectures, are now key skills. They will become even more so in the future. One can be a good manager, but if you look for instance at Google, Microsoft, Facebook and many others, you will notice that all the founders are programmers. Only at a later stage did they bring in additional good managers. Although an expert level in programming is not required, coding is important in the sense that it provides fundamental notions enabling one to understand how AI works at a logical and architectural level. In addition, I would suggest learning Mandarin as the future of AI – and all that will go with it and come from it – is increasingly in China. Overall, AI is this century’s top area of opportunity, allowing those who have the capability of understanding such concepts to exploit very profitable futures. The alternative is, to my view at least, considerably less positive.
Gianluca Biancardi & Giorgio Guarrella.