The AI Renaissance

Candles, typewriters, telegraphs, and calculators have something in common. They have all been radically improved by technology. And if switching to electricity and ultra-high-speed Internet has been a complete no-brainer in the past, the adoption of Artificial Intelligence (AI) is not only gaining the lion’s share in the data revolution but becoming the backbone – the operating system – of the modern era. 

Even in the aftermath of the pandemic, data is growing at exponential speed while our brainpower remains at best fixed. More than ever, we need synthesisers, decision-makers able to put together the correct information at the right time and think critically about it. And this is where we now stand. Just recently, math, biology, investing, and space exploration have all seen AI lead scientists to numerous breakthroughs and discoveries. Here we recap, explain and look at how all of this is taking place – retracing the footsteps of what is de facto an AI Renaissance.

It’s About Time! At 65, AI is Finally Coming Of Age

February 1956. While the communist party was celebrating its 20th Congress in Moscow, the famous ‘Heartbreak Hotel’ of Elvis Presley entered the US charts for the first time. Later the same year, just before Dwight “Ike” Eisenhower was about to become the 43rd president of the United States, the world’s most influential brainstorming session was taking place in Hanover, New Hampshire. 

There, a workshop gathered a dozen Mathematicians, Physicists, and Computer Scientists to discuss their ideas, opinions, and visions about the future of technology and computer science. Although no one could have imagined how significant their work would have become for future generations, it was during that period that the tech revolution we are witnessing today effectively started – the first time the term “Artificial Intelligence” was coined.

As one of the founding fathers of the workshop, the legendary American Computer Scientist John McHarty wrote in the proposal:

We propose a 2-month, 10-man study of Artificial Intelligence be carried out during the summer of 1956 at Dartmouth College. The study is to proceed based on the conjecture that every aspect of learning or any other feature of intelligence can, in principle, be so precisely described that a machine can be made to simulate it.”

Exhibit 1 - Participants of the 1956 Dartmouth Summer Research Project on Artificial Intelligence in front of Dartmouth Hall. From right to left, Claude Shannon (front right), John McCarthy (back right), Marvin Minsky (center), Ray Solomonoff (front left), and Nathaniel Rochester (back left).

In this sense, the first attempts to use AI quickly delivered many remarkable discoveries previously unthinkable. In the span of a few years, as computers started solving problems in algebra, learning checkers strategies, and speaking basic English, all of a sudden large amounts of funding were poured into AI research. 

However, the frenzy did not last very long. The initial breakthroughs, experts suggest, had only been a sort of low-hanging fruit: low computing power and a general lack of data haunted researchers that struggled to deliver on the original high expectations. Consequently, less than twenty years later, both the US and UK governments cut off exploratory research in AI, officially starting what historians now call the “AI Winter”. 

Nevertheless, AI remained one of the most sought-after topics, especially in Academia. This resulted in an impressive amount of theoretical research that started to be applied with increasing success as soon as data and computing power grew. This exponential growth, well described by the famous Moore’s Law in Exhibit 2, completely turned the cards upside down, making the training of AI models more efficient and less costly. 

Empirical evidence of Moore’s Law: the number of transistors on integrated circuit chips from 1970 to 2018. Source:
Exhibit 2 - Empirical evidence of Moore’s Law: the number of transistors on integrated circuit chips from 1970 to 2018. Source:

And in the end, when we look at the bigger picture, we see that there is also an additional element that has been crucial for getting AI out of computer science laboratories and in the daily workflow of billions of people. Simply put, the AI we ended up developing has been very different from the one depicted in sci-fi movies. 

As we will see, AI’s most widespread applications are narrower (and more reliable) than we might think. Instead of machine domination, we rather entered into a more profound human interaction with these tools, helping us solve problems that previously would have required a tremendous amount of effort.

Start integrating AI into your investment process with the most advanced no-code solution.

Fill out the form to book a demo of Sphere®, and learn how it can bring value to your investment decisions.

Book A Custom Demo

By submitting you are agreeing to the Terms of Use and Privacy Policy
Thank you! Your submission has been received.
Oops! Something went wrong while submitting the form.

The Hammer and Anvil: Why AI is ‘Moneyball for Everything’

The hundreds of dystopian Hollywood movies about technology have made the idea of a superintelligent AI deep-rooted in our culture to a certain extent. This is the case, for example, of Stanley Kubrick's famous “2001: A Space Odyssey” or James Cameron’s 1984 blockbuster movie “The Terminator”, to the point that hardly a day goes by without someone drawing comparisons between its robots and the – supposed – current applications of machine learning. 

Human intelligence’s complex nuances that make us unique have never been put on the line by AI developments to a closer look. For example, when Garry Kasparov lost against IBM supercomputer in 1997, or when Google’s Deepmind defeated the world’s best player of Go, it was just computers showing us their edge at processing data, perform complicated calculations, and solve complex equations. No more, and no less than that.

Ironically, every time we experience this kind of event, we all win by a little. And it is not about being overly optimistic about the impact of technology. It is about rethinking and deeply understanding where humans and machines add value. Who is the hammer, and who is the anvil in this game – and never forget about it. 

As humans, our job is to use our imagination and creativity to build machines that either automate routine tasks or allow us to go beyond what we thought possible. Whether it is about inventing the wheel or a turbo-charged space rocket, the machines’ destiny is and always will be the one of being enablers. On the other hand, humans are synthesisers: we must put together the correct information at the right time, think critically about it, and make important choices wisely. 

And this is where the whole AI and Machine Learning ecosystem enters the stage to help decision-makers create value by extracting insights from vast amounts of data. Quoting the famous CNBC reporter, Eamon Javers, “think of it [The AI] as Moneyball, but not just for baseball, for everything”. 

In this sense, AI’s role is similar to how predictive analytics rocked baseball twenty years ago, inspiring the famous movie directed by Bennet Miller. The film tells how in 2002, Oakland A’s General Manager Billy Beane completely transformed the way teams were built, using data to uncover the hidden value of baseball players. 

To achieve this, he reshaped his organisation’s culture towards one of unbiased and data-driven assisted decision-making. In this sense, he pioneered data to eliminate biased factors, such as players’ age, appearance, or personality, and truly unlock value.

And even if reality might be more complicated than that, this is what AI is about in today’s complex world. On the one hand, it helps us make sense of complexity by turning data into actionable insights. On the other hand, it is critical to mitigate typical human biases – like overfitting – ensuring more rational, reliable, and scientific results. 

Exhibit 3 - Jonah Hill plays a recent Ivy League graduate who explains to Beane (Brad Pitt) how data can help find undervalued baseball players.

Don't Underestimate the Power of ‘Weak’ AI

And if the analysis of Big Data eventually led Coach Beane to 20 consecutive victories, forever changing the face of baseball, today the unprecedented power of AI is enabling decision-makers across the most diverse industries to unlock value and reach unseen heights.

In this sense, current AI applications are directed to solve specific problems. They belong to the “Weak AI” world instead of the “Strong” one, in more technical jargon.

Strong AI – also called Artificial General Intelligence – is a field of AI research focused on creating something capable of thinking, learning, and solving problems like humans. On the contrary, Weak – or Narrow – AI reflects a more practical use of this technology to target computationally complex but well-defined problems.

This is the case, for example, of laboratory testing in which Weak AI is increasingly used to automate repetitive and time-consuming tasks. This allows lab technicians to focus on more sensitive testing issues, such as analysing abnormal or critical samples and offer more valuable insights and recommendations.

Indeed, this radical shift has given a massive boost to research, and the dramatic impact of the recent pandemic has only accelerated the transition towards greater technological integration. Over the past few years, a multitude of achievements occurred. And still, last year can be considered truly exceptional, as we witnessed some astonishing results in a variety of sectors: astronomy, mathematics, biology – and finally investing.

Inside SpaceX’s CrewDragon AI systems 

For instance, AI has dramatically reshaped space exploration. The impact of technology in this sector is nowhere more evident than by looking at the evolution of spacecrafts’ internal dashboards. Exhibit 3 compares the 1985 Space Shuttle Atlantis with the 2020 SpaceX Crew Dragon. Gone are the tens of knobs and switches. Astronauts now use large, futuristics touch screens.

This is possible because the Crew Dragon is run by AI software designed to assist space navigation. Indeed, in May 2020, a sophisticated AI autopilot steered the cone-shaped Crew Dragon up until 20 meters from the International Space Station. In this sense, it contributed to making space exploration safer and more efficient, potentially detecting any dangers in lengthy space missions, such as changes in the spacecraft atmosphere or sensor malfunctions.

Remarkably, the adoption of AI software has further enhanced the role of Mission Control centres. They have become the critical touchpoints between hardware and astronauts, responsible for checking and guiding the mission’s overall state. 

Exhibit 4 - The inside dashboard of the Space Shuttle Atlantis (1985) and Crew Dragon (2020)

How AI Cracked Partial Differential Equations for Good

AI tools have also proven extremely valuable to overcome significant mathematical hurdles like solving Partial Differential Equations (PDEs). And even if the name might sound scary at first, almost all investors know at least one: the famous Black-Scholes equation.

PDEs are incredibly complex but extremely precise at describing changes over space and time. This makes them valuable to understand and model dynamic systems’ behaviour, from option prices to planetary motion, from air turbulence to plate tectonics.

Until recently, solving PDEs was computationally intensive and time-consuming. However, in October of 2020, a newly developed deep-learning technique from Caltech University  achieved a 30% lower error rate, and was 1,000 times faster than previous computational methods. 

Exhibit 5, taken from the original paper, gives an example of this new technique’s accuracy. The upper row shows four snapshots of a fluid’s motion starting from a situation of initial vorticity; the bottom one shows how the neural network predicted the fluid would move, with an almost identical result.

Because these equations can be applied to a multitude of problems and fields – sound and heat transfer, diffusion, electrostatics, electrodynamics, fluid dynamics, elasticity, general relativity, and quantum mechanics – such a breakthrough promises a potentially immense impact.

Exhibit 5 - AI Predictions (bottom) compared to the Actual Movement of Fluid (top)

Predicting the 3D Protein Structure with AI

Ever wondered how all life depends on about twenty amino acids? In late 2020, a group of computer scientists made an incredible leap forward in answering that question. 

The team from Google DeepMind was able to tell the protein’s 3D structure by using AI. This process is extremely complex, as illustrated in Exhibit 6, because of the multitude of potential combinations of amino-acid sequences, folds, and tridimensional shapes.

Until recently, to determine protein structures experimentally, lengthy and complex techniques such as X-ray crystallography and cryo-electron microscopy were needed. Instead, DeepMind’s program was able to apply deep-learning techniques to predict the final structure of a target protein sequence by incorporating structural and genetic data with physical and geometric constraints.

This discovery is a real game-changer for life sciences, medicine, and bioengineering. It will empower an entirely new generation of biologists to find the answers to more advanced questions, like why misfolded proteins cause diseases or how to artificially create a protein that breaks down plastics.

Indeed, accurately predicting how a protein folds would vastly accelerate our understanding of cells, thus enabling quicker and more advanced drug discovery and design. For instance, in early 2020, this technology was used to predict the structures of a handful of SARS-CoV-2 proteins that had not yet been determined experimentally, saving time, effort, and money for the development of vaccines.

Exhibit 6 - How amino acids determine the 3D structure of proteins. Source: DeepMind

Get the Sunblock Out. The ‘Summer’ of AI in Investing is Already Here

What worked for AI in other industries is increasingly gaining traction also in the world of investments. Here, its applications are becoming more and more frequent, mature and reliable, to the point that AI is quickly becoming a fundamental bedrock of modern portfolio management. 

In this sense, AI advancements in investing appear as a necessary evolution rather than a complete revolution of existing techniques. From the development of unbiased investment strategies to the construction of tailor-made and diversified portfolios, they are giving investors a chance to finally close the gap between theory and empirical evidence.  

Unbiased Research Brings Reliable Results

Finance has been definitely one of the most exciting and challenging fields in which to apply quantitative techniques. However, although tempting, quant tools have often failed to deliver on their promise. And if their firepower is considerable, what has held their mass adoption has been the inability to anticipate, or rather be aware of, their fragilities. 

As we recalled in our recent paper “The Right Call”, this happens for several reasons: a reliance on outdated statistical techniques, not taking into account real-world peculiarities (i.e., trading costs), or merely flawed models caused by human bias. Consequently, it is very easy to stumble upon false positives or, worse, wrong decisions due to a naive interpretation of empirical findings.

In this context, AI is setting a new gold-standard for developing reliable, unbiased, and controllable investment strategies, overcoming these limits. Backed by a solid economic rationale, AI can tackle time-consuming and bias-prone tasks, understanding how to extract investment signals from markets’ background noise. 

Combining technology, research, and expertise is what truly enables investors to build investment strategies that exploit specific market anomalies and avoid irrational market movements. And while all of that happens, researchers in Mission Control remain in charge of supervising and managing the entire process, ensuring that everything runs smoothly - just like SpaceX engineers.


AI-Driven Means 100% Customizable

Another critical aspect is fuelling this tech evolution in investing: AI models bring together scalability and personalisation. 

Far from being a one-size-fits-all solution, AI models can efficiently be tailored to include not only a target risk-return profile but a whole set of client requirements, constraints or investment objectives.

For instance, AI redefines the frontiers of how advisors and private bankers cater to their clients’ investment portfolios. In this sense, many large financial institutions provide their advisors with scalable solutions to facilitate building custom portfolios and reports, incorporating demographic data – like age, profession, or family status – and specific investment objectives, including preferences on Factors, Investing or ESG screenings, for instance.

To a certain extent, this is marking a natural step forward from the era of traditional model portfolios that rarely finds a perfect fit to a scalable “one-client, one-portfolio” approach that can accommodate the whole spectrum of the client’s actual investment preferences.

Continuous Learning Makes AI Antifragile

To a certain extent, the ‘killer app’ of AI models relies on their capability to quickly adapt when markets change. And combined with the two previously-mentioned features – unbiasedness and customisation – this makes AI models build portfolios that, at any given time, have the best chances to perform in the current market environment. 

Indeed, adaptability is deeply ingrained in the DNA of machine learning techniques, as we can see from Exhibit 7. Thanks to this, they uncover hidden relationships in data and adapt as soon as new information becomes available. Put differently, continuous learning ensures that AI can follow the gradual evolution of market dynamics as soon as they tend to unfold.

And even when unforeseeable outliers occur – such as the Covid-19 market sell-off or the recent GameStop frenzy – they bring crucial insights into the current market regime’s health. In this sense, AI models are definitely what Nassim Taleb would call “Antifragile,” meaning that the more they are subject to heterogeneous events, the better they can learn and, the more robust they become over time. 

Clearly, this offers enormous advantages for asset and portfolio managers, who can avoid unpleasant surprises by building portfolios that are better positioned to navigate increasingly complex financial markets.

Exhibit 7 - Human Supervision and Continuous Learning: How Well-Trained AI Models Adapt to Current Market Scenarios

Is this a Rebirth, Revival, or Renaissance of AI?

When it comes to AI, 2020 can be really considered a significant inflection point. After decades of research and stimulus, AI is finally delivering on its promise to scale up our ability to connect the dots and cut through complexity like never before.

And yet, as the American Hedge Fund manager Paul Tudor would say, “No human is better than a machine, but no machine is better than a human with a machine”. Indeed, this transition from statistical reasoning to a new era of assisted decision-making is also the result of the progress achieved in the field of more transparent AI models. 

As we recently discussed in a paper entitled “Building AI that Investors Can Rely On”, with the right skills, expertise, and tools, AI models instead of a black become a glass box. And this has turned to be extremely valuable to navigate the gradual unfolding of the markets in an unbiased, scientific and rational manner. 

This flourishing of AI does not limit itself to investing. Indeed, it has matured into almost an infrastructure to navigate a world that is growingly interconnected and complex, with an impact that resembles the one brought by electricity, steam power, and the Internet. A modern operating system – a new Renaissance – set to become the next building block of technology, science, and investing.

Start Integrating AI Into Your Investment Process