In a conversation with astrophysicist Neil deGrasse Tyson, “Singularity” theorist and author Ray Kurzweil talked about what can and cannot be predicted in the development of the internet and AI.
In the 90s, Kurzweil reckoned that there would be a need for search engines because there’d be so much knowledge on the web that it would be impossible to find anything.
What could not be predicted, he noted, is that of 50 projects doing that, “these couple of kids in the Stanford dorm” would win, he said, referring to Google.
“I wouldn’t have to work very hard if I had made that prediction,” Kurzweil noted.
Today, it seems obvious that far more sophisticated machines are required for big data and automation, but the big question remains: who’s going to figure that out in which industries?
Who is going to make money in AI?
In a recent article, Simon Greenman, co-founder and partner at advisory firm Best Practice AI, tried to create a framework for tackling this question from a start-up perspective, comparing this year’s investment boom with a gold rush in a recent article titled: Who is going to make money in AI?
“What we’re seeing here is the startups come rushing in funded by venture capitalists (and) the corporates come rushing in to try and adopt it. So, there is a mass of companies trying to effectively mine the gold,” said Greenman speaking to MarketBrains.
Specific to finance, he named firms like Zest and Affirm as contenders to watch, pointing to characteristics such as level of investment (fundraising reached some $300 million and $700 million respectively).
“For that level of investment, it means that they have significant customer traction …those companies that can get scale are going to do well,” he added.
Q1 buy-side prediction flashback: The advancement of big data and AI adoptions on the buyside is going to provide more wealth creation opportunities for the general public in 2018. Big data and AI were once secret weapons which belonged to only the most sophisticated investors. Nowadays, however, these techniques have become much more commoditized and accessible for the broader investors through data vendors and cloud computing platforms. 2018 is the year more investors will be pressured to gradually adopt them as an essential toolkit. Big data and AI are no longer only about seeking alphas but also about driving productivities. Greater industry focus will be given to the use of big data and AI to drive efficiency in the investment processes and pass along the benefits to consumers through lower fees and more accessible fund offerings – Ichihan Tai, portfolio manager and head of data science, Tokio Marine
AI investment by the numbers
There is a tsunami of research trying to pin down how much money has been invested in artificial intelligence and how much that investment might pay off.
Estimates range from billions to trillions of dollars in terms of how much AI could increase economies in the next decade or more.
In the US, AI had another big quarter: funding to US-based artificial intelligence companies rose 21% in Q2 2018 to $2.3 billion, after a 37% rise in Q1, according to research from CB Insights and PwC.
But AI hedge funds haven’t been doing so well so far this year: AI hedge funds posted losses for the third consecutive month, down 0.60% in June. This loss brought their year-to-date return down to -3.11%, placing them behind all of the primary strategic mandates, according to Eurekahedge data.
AI (media) hype cycles
One of the most cynical assessments so far comes from Riot Research’s tech unit in a report titled, AI: show me the money.
Riot thinks that the AI market will reach $39 billion globally by the end of 2023, compared to investment of $100 billion in 2018 alone, and predicted that the field will be “littered with corpses” on the way.
Despite this cynicism overall, finance and insurance remain poised for gains because an “AI bubble burst will clear the air for sustained growth in key sectors”.
Indeed, finance and insurance are described in the report as an “area where some of the hype in terms of growth and threat to jobs is justified.”
On the latter point, there’s another set of numbers that industry professionals are warning on: how many people are actually qualified to “do” AI.
In a report from earlier this year, Canadian firm Element AI found that there are just over 20k PhD-educated researchers globally that are capable of working in AI research and applications, with some 3,000 candidates actually looking for a job.
The US had the highest concentration of researchers at 9,000, followed by the UK with just under 1,900, and Canada came in third with just over 1,000.
New graduates, narrow focus
There are plenty of startups founded by people who only understand neural networks, or deep learning, because that’s what’s being taught right now, said Pierre Haren, CEO of Causality Link, a US-based startup focusing on explainable AI in research.
Haren’s background with AI goes back some 40 years, and a company he founded, ILOG, was acquired by IBM for $340 million in 2009.
The narrow focus on one popular AI technology, deep learning, is “disastrous” he added, but somewhat typical based on AI waves that have come before.
“There will be a lot of people falling flat on their face, because there is a discrepancy between what marketing says and what the technology can do and therefore, as usual, this is long-term going to create downdrafts for the excitement,” he said.
But that won’t change industry demand for a higher level of “smart, explainable automation”, Haren added.
Q1 buy-side prediction flashback: As we progress into 2018, one of the key areas of growth will be a move from machine learning more towards deep learning/AI. Many traditional fundamental investors have been moving towards a ‘quantamental’ investing model of the last number of years, making greater use of data and self-learning algorithms in their investment decision making. A natural place to start for most of these investors was to apply machine learning algorithms to their existing data sets to help enhance investment decision making. However, as they become more comfortable with how to implement and use these models, the natural progression in 2018 will be towards deep learning and artificial intelligence (AI) models, such as that used by Google’s DeepMind when it famously beat a human expert in the board game Go. – Charles Ellis, trader and quantitative strategist, Mediolanum Asset Management
Talent mergers and acquisitions
The acqui-hire trend across all the major brokers has drawn much attention this year, but it’s unclear by what metric success will be measured by.
As far as exit strategies go, being picked up by a big firm can be very profitable: in one of the bigger deals of the year, S&P Global bought out AI analytics firm Kensho for $550 million.
But for how that could impact innovation, being acqui-hired comes with huge caveats.
In a recent article, Matthew Klein, founder of Collective2, a trading platform seeking to disrupt the hedge fund industry, asked the question: “Where are the big, fat, lazy industries we’re trying to disrupt?”
The article challenged inspirational tech-focused disruption narratives as mere fables: “In the last decade, venture capital has poured nearly one trillion dollars of cash into technology startups. In that time, we’ve had two successful David-beats-Goliath stories (Netflix, Amazon), a handful of David-still-fighting-Goliath-but-Goliath-not-giving-up stories (Uber, Tesla). And a lot of Emoji Apps and Email-Marketing-For-Other-Tech-Companies stories.”
Collective2 is not an AI company itself, rather a collection of strategies from traders around the world: it’s been around a decade, has some $90 million in assets being traded, with about 128k users whose activity varies on market conditions.
Noteworthy participants such as Jonathan Kinlay have made use of the platform for a VIX Scalper and VIX ETF Trader strategy.
Q1 buy-side prediction flashback: While AI has grown from niche concept to mainstream application, 2017 really embraced it, and Incorporating AI into and investment “theme” is a growing area. Three ways to do it: invest in companies who invest in AI; use AI to choose companies fundamentally; use AI to invest in companies with an algorithm measuring technical strength. With a growing number of corporations incorporating AI into their business, there are more companies to choose from. But actually using AI to pick companies to invest in promises to be a theme to watch as it represents a threat to the traditional fund with large teams of analysts. As a pioneer into this concept, AIEQ (EquBot ETF) had early success in 2017 in attracting assets to a niche concept. Performance by these funds will have to be there versus it’s alternative, low cost beta funds, but the idea promises to excite. – Bryan Novak, portfolio manager and partner, Astor Investment Management
Speaking to MarketBrains, Klein confirmed that AI-based strategies are increasingly popular on the platform in the last couple of years though he added a caveat emptor flag along with that.
“I know that machine learning is very hot now, so everyone’s doing machine learning to clean their bathroom toilets,” said Klein. “I have observed in many years of being in the business that it’s an arms race, and ideas and methodologies that seem to work or bear fruit early on are often arbitraged away pretty quickly.”
One thing he’s sure of: Collective2 is not going the way of acquisition. The firm has raised $1 million from angel investors and is “adamant” against the VC route.
Klein described the platform as a way of achieving the same results that the hedge fund industry is supposed to achieve but just dispensing with that framework of firms and companies so that investors and alpha generators can interact directly.
His experience of wanting to remain disruptive in the tech industry carries important lessons for AI startups being acqui-hired by financial firms, which are themselves competing with the big tech firms.
For one, he described banks as “interfaces to regulators”, an attractive benefit for an AI startup that just wants to do its own thing.
And compliance is serious business: regulators around the world have sounded the starting pistol in the race to become the global leader in artificial intelligence. In the past fifteen months, more than a dozen national strategies have been announced to promote the use and development of AI.
Banks meanwhile are motivated to scoop up the smartest people amid a skills shortage.
But Klein also noted that there’s an element of FOMO: “Everybody doesn’t want to be the stupid person that turned down Google.”
He listed the winners in AI as the “arms manufacturers”: tool providers, data vendors, cloud platforms, for examples. “The winners are the people or firms that can aggregate the most transactions or data.”
Q1 buy-side prediction flashback: 2018 will see growing demand for AI applications that can enhance asset managers’ investment outcomes and differentiate their overall product offering. Despite much marketing hype from large asset managers with big R&D budgets, AI innovation will continue to be driven by experienced AI fintech players. Key considerations will be the length and quality of the live track record of the AI investment products powered by these technologies and their rate of adoption by major asset managers. Also, investors will become increasingly more sophisticated in telling the difference between genuine AI applications that can transform the quality of investment outcomes and applications which are just repackaged existing (outdated) allocation and risk management models with slicker front-ends and customer interfaces.” – Marco Fasoli, co-founder and managing partner, A.I. Machines
Best Practice AI’s Greenman has some advice for startups going the acqui-hire route: figure out what your high value use case is as quickly as possible and test it with financial services organizations as quickly as possible. Sales cycles can be months upon months, if not years, he added.
“Those startups that are doing well and scaling in the financial services industry, they figured out their go-to market and commercial layer. They are hiring sales and business development people that come from the financial services industry and know how to navigate these bloody long, frustrating sales cycles,” he said.
One partner from an Israeli-based VC firm said to MarketBrains on the sidelines of a conference that acqui-hiring for startups is avoided: “$2-$3 million per head has no value when we want to build a billion-dollar company.”
He added that they tell their startups not to go to innovation labs because it’s a waste of a year or more. Another VC noted however that Citi’s Innovation Lab in Tel Aviv remains desirable.
For financial services firms, applying artificial intelligence or machine learning in core business functions is a matter of degrees.
Using a framework with five stages of analytics maturity, a recent research report from the
International Institute for Analytics found many companies “stuck” in phase 2, which is
characterized by implementation at a localized level, said David Alles, a VP for the firm and author of the report, speaking to MarketBrains.
As firms become more sophisticated and move through to phase 5, they become more data-driven with supportive underlying infrastructures, then graduating to use those analytics for business impact.
Insurance companies, said Alles, are a “classic” example of being localized due to hindrances such as not having direct customer contact, instead getting access via broker channels. Not having an integrated perspective on their business makes them vulnerable to disruption, he noted.
Ahead of the pack is credit card processing and payments, which rank alongside market leaders, a kind of “best in class” measure.
Successful companies share certain characteristics, explained Alles: they have a single source of truth, data-driven decision-making is embedded throughout the organization and, as a result, they are very adept at putting analytical methods to work as new technologies become available.
Some of the firms he cited as being in this class are Morgan Stanley, Goldman Sachs and Visa.
Killing the project, intelligently
But it’s still a long road ahead for the industry as a whole, and cultural change is slow, he noted, adding that between 50% and 70% of analytics project either fail or don’t meet their objectives, and AI and machine learning are going to be the same.
“You can’t cheat it, you have to go through the steps and get the capabilities and certain foundation in place, then you can go ahead and take advantage of AI and machine learning,” Alles said.
To what extent incorporating and absorbing AI startups’ products and people contributes to that remains to be seen. For now, it’s a mixed bag of experiences as a broad swathe of the financial industry figures out not just advanced technologies like AI, but also finds itself buffeted by blockchain and cryptocurrencies against an uncertain regulatory backdrop.
Causality Link’s Haren put forward a highly organic explanation with a biological basis that he sees as applicable to startups.
The human body gets more protein from dying cells in the body than it does from food.
“A big organism has to permanently kill projects and intelligently recirculate all the components, not to the molecular level but to the protein level, and that’s exactly what companies that acquire other companies have to do,” he explained.