For the last 35 years Artificial Intelligence (AI) has been promising much yet not delivering outside of the academic world. So what exactly is AI? There is actually a fair bit of confusion about even the definition of AI, and here I would like to take Gartner’s viewpoint, avoiding marketing terms like ‘cognitive’ in that AI stands for “Amazing Innovation”. Fundamentally, AI gives us solutions that we never thought were possible. So this is not an evolutionary change (solutions that are incrementally better) but a step change in what is even possible.
What we can say is that there are some common characteristics of artificially intelligent solutions, in that:
- They have training, validation and operational phases with the associated data
- Most solutions require large amounts of data to be genuinely useful
- There is a ‘feedback loop’, meaning that the models get ‘better’ over time
- There is a degree of unpredictability, in that the system learns by randomly exploring the solution space
Alignment of two planets
As far as buzzwords go right now I am not sure if there is a bigger one in technology circles. Again Gartner has AI and Machine Learning (ML) surfing the crest of their hype cycle. As such it appears to be the classic solution looking for a problem. However, why is this the case? After 35 years why now? The answer lies in the perfect alignment of two planets that has been created by some of the largest technology giants of our modern era.
The core business of Amazon, Google, and to a certain extent, Microsoft required them to build large scale data centre operations. These included data centres all over the world and the dedicated network infrastructure to link them together in a massively fault tolerant, low latency way required to support their core businesses, whether this is search, trading or ad-words. Having built this infrastructure have been able to diversify their business by monetising this investment, offering these services to paying third parties. To us in the cheap seats some of this technology looks like magic, as 15 years of internal innovation has suddenly been released in one go, solving problems the rest of us thought were intractable.
The data planet
AI, in its native form, requires huge amounts of data to be useful. The cloud offerings available now allow us to store and manipulate petabytes of data at a fraction of the cost of an ‘on-premise’ solution. Another problem with data storage is the need to have the data available in the right location in a timely fashion, so that any models have a chance of seeing the whole picture. The major cloud providers are racing to provide this kind of data availability (e.g. Google Big Data, Google Cloud Spanner and Microsoft Cosmos) as they go after market share.
The compute planet
As mentioned, the AI models require a data set to train the model on, and then operational data to generate intelligence from (for example predictions or diagnoses). The compute resources to train the models can be significant and with cheap computing available in the cloud, once again the cloud is an enabler of AI. In the cloud, not only are central processing units (CPUs) available, but also graphics processing units (GPUs) and tensor processing units (TPUs) as well. The operation of the models often has strong non-functional requirements, which means the ability to parallelise and scale quickly is critical.
Time to embrace the cloud
There is much talk in the press about how the advent of AI is going to replace humans in the work place with a set of machines being curated by a small set of experts. However, while there will be change, this future is unlikely. It is analogous to the pundit who recently declared that AI would result in the demise of Alpha, as we end up with a perfectly efficient market. Both these predictions rely on the fact that the normal waste and imperfections in the process are magically removed. More likely, rather than be replaced by AI or ending up as model curators, employees will actually be more empowered, and AI should thus be treated as a productivity tool. More intelligence will be available even faster, enabling humans to make better decisions, particularly in a world of unstable markets where creativity, empathy and other emotions are strong influencing factors.
The other intriguing angle to watch unfold is the role of the regulator. Today AI is used to identify trading strategies that are then fed into a standard rules engine to execute. These, strategies cannot be run by a neural net due to the unpredictability inherent in those systems. As with other disruptive technologies (e.g. Blockchain and Cloud), the regulator will take their time before opining on what is acceptable.
This is the reason that we see the most promising use cases currently emerging in areas where the regulator does not worry too much about certain methods, such as operational digitisation, know-your-customer (KYC), anti money laundering (AML) and fraud.
It is the advent of cheap compute and data engineering in the cloud that makes AI possible now. Whether it is for cost reduction, new or faster business insight or providing better customer service, now is the crucial moment for financial service firms to embrace this innovative technology.