What is the time lag from Science Fiction to reality? Arthur C. Clarke wrote about geostationary satellites in 1945 and it became reality 20 years later, but Clarke’s work was more grounded in science than the Kubrick classic 2001 featuring the Hal AI computer. That was released in 1968 – 48 years ago. It is such an easy fiction trick to imagine a machine as smart as a human, but the reality of AI has been much harder. It is now getting a lot easier. AI is being democratized and is coming to a job near you. That is why this week is all about the impact of AI on different parts of the Financial Services business.
Today is a general briefing about the current state of play in AI. AI is mostly about machines doing what humans do, so the rest of the week is about specific jobs in specific Fintech domains:
- Tuesday will focus on the impact of AI on Digital Wealth Management., specifically recognizing investment opportunities before they become obvious.
- Wednesday continues the focus on the impact of AI on Digital Wealth Management., specifically “looking for the super algorithm”.
- Thursday will focus on the impact of AI on Insurance. Will AI machines replace the Actuary?
- Friday will focus on the impact of AI on Small Business Finance. Will AI machines replace the CFO?
The Current State of the Art in AI
Most of the BigTechs have AI initiatives. For example:
IBM Watson is possibly the most famous, because they did such smart marketing by using Watson in the Jeopardy game. An earlier iteration, Deep Blue, was famous for its mastery of Chess. IBM is the master at turning leading edge into big enterprise projects. To do this, they first have to ensure that the technology has passed the bleeding edge stage – to ensure that it does actually work. The Jeopardy game was a way of telling the world that AI does actually work. The fact that IBM (and so many other BigTechs are rolling out AI into the market (including TV Ads) is one indication that it is ready for prime time. IBM’s CEO, Virginia Marie “Ginni” Rometty has said she hopes Watson will create $10 billion in annual revenue within 10 years.
Watson is focused on a machine answering questions posed in natural language. This meets the Turing test, but is only one branch of AI. The commercial rollouts of Watson are starting:
- Memorial Sloan Kettering Cancer Center in partnership with the health insurance company WellPoint use Watson to make decisions about appropriate care within the constraints of the patient’s health care plan.
- Welltok’s CafeWell service is like a health concierge where customers get exercise and nutrition tips and are rewarded for healthy behavior, through lower healthcare premiums.
- Recently, IBM announced that Watson was powering a Hilton concierge called Connie.
- The North Face is using it to answer shopper’s questions such as “What do I need for a 14-day hiking trip?”
- Kayak.com wants to answer questions such as “Where should I go for a romantic vacation that includes the beach with my wife in January?”
Facebook AI Research (FAIR). One year ago, Facebook acquired Wit.AI with a specific focus on speech recognition, which is a critical part of AI. First a machine has to understand what the human is saying. Wit.ai was a Y Combinator startup that had 6,000 developers on its platform when Facebook acquired the company. This exemplifies a key trend, which is the democratization of AI. This is about small companies building on powerful platforms and that may end up empowering humans as well as replacing humans. IBM Watson also has partners who build the domain specific applications. The low hanging fruit for Facebook is voice to text for messenger – typing on a mobile phone is a consumer pain point. Facebook recently launched a more broad based AI partnership program.
Google has Deep Mind. Like IBM, they chose a game to show what their AI could do, the game of Go. DeepMind started as a UK company, founded in 2010 and was acquired by Google in 2014. They started by learning from how game developers work using a neural network. This is a more scalable way of doing AI than programming rules. For example, consider the challenge of driving a car (which Google is very focused on). You can program rules such as Red Light = Stop. That is simple, but time consuming. How do you discover the more nuanced behavior of drivers when they have to judge the probable behavior of an oncoming driver in a snowstorm at night? Observation and imitation is how humans learn. Neural networks imitate that behavior. Starting with video games was smart because it is easy to attach an observational sensor.
Open Source and the democratization of AI
The MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) has been at the bleeding edge of AI for a long time. MIT is also a big proponent of open source in all its forms. For example: their AI courseware is available for free online.
There are so many AI projects and many of them are open source. We may soon see Artificial Intelligence As A Service. This means that we are moving into a phase where it is all about the practical use cases, which is what this Fintech AI Week is all about.
Reality check – hard is easy & easy is hard
According to this research, AI is about as smart as a 4 year old. That is a provocative headline but I can still beat most 4 year olds at Chess, Go and Jeopardy. This exemplifies the AI mantra that hard is easy and easy is hard.
- Hard is easy – Chess, Go and Jeopardy.
- Easy is hard – understanding what the expression on your mother’s face means.
It’s the data stupid.
AI is closely related to Big Data. If you have all the data, the compute capacity is now available and lots of AI tools are open source. So data is the missing piece. This reliance on data is something we will encounter as we explore specific Fintech use cases later this week.
Augment or replace humans?
AI will wipe out lots of jobs. Drivers maybe the biggest job pool at risk, but there are many others. Historically new jobs have appeared as humans adapt and use the new technology to provide value added services. It is unclear how this will play out in the AI wave (but it is usually only clear in hindsight).
Jobs that humans cannot do
The easy commercial wins are in applications that replace humans (to save costs and improve predictability). However the founding principles of AI came from Alan Turing’s efforts to crack the German cipher codes during World War 2. That was tackling a job that was beyond the reach of humans. Yet when you read that fascinating story, it is the mix of human ingenuity and machine power that comes across.
There is another hard problem to crack that no human can do that is in the Fintech domain – calculate catastrophic insurance risk caused by global warming. Pity the poor Insurance and Reinsurance companies who have to work on static models of weather patterns that are now increasingly useless. This will need a merger of AI and Complexity Science and the results may also be useful in other massively complex systems such as the global financial markets, biological and neurological networks in the human body and patterns in communications networks. There are plenty of “wicked tough” problems for the best minds of our generation to work on.
Did an AI journo-robot write this research note while I went skiing? Some writing is going that way, but robotic writing as of today is only at the low end and that kind of low end writing can be done very cheaply by somebody offshore, so the automation payoff is weak. Commercial rollouts of AI need a magic quadrant of:
- Technically feasible.
- A good ROI from replacing humans and improving predictability.
The commercial rollouts being done by BigTechs such as IBM, Google and Facebook all fit this magic quadrant. There are many of those kinds of jobs in Financial Services. Those are the kind of applications we will be focused on this week.