5 Stages of Autonomy

By Andreas Roell


Last week’s post had the purpose of deciphering the various approaches to trading in the financial markets. It was stimulated by the difficulty that the financial industry community has with classifying the emerging mechanics of algorithmic trading, the category AlphaTrAI is part of.


In the first blog, I was describing algorithms-at-the-core traders like us as autonomous traders for descriptive purposes.This week, I want to go a step further and dissect the state of autonomy for firms like AlphaTrAI.


How autonomous is the current state of possibilities? What does autonomy truly mean? Perhaps even how to ask smart questions to firms like us to truly identify how far on the spectrum one is.


To make this post exciting for you, there is a chance that I might shock you with my perspective despite the fact that I am the chief executive officer of an algorithms-at-the-core shop.


First things first. The best way to think about this discussion is to align it to the commonly adopted classification of autonomy in automobiles. 


Level 0 – No Automation
This describes your everyday car. No bells and whistles. Just your ordinary cruise control to help with long-distance driving and minimize the risk of a speeding ticket from a lead foot. Almost all cars today will offer Level 0 autonomous technology.


Level 1 – Driver Assistance
Here we can find your adaptive cruise control and lane-keep assist to help with driving fatigue. Adaptive cruise control will keep a safe distance between you and the car ahead of you by using radars and/or cameras to automatically apply to brake when traffic slows, and resume speed when traffic clears. Lane keep assist will help nudge you back into the lane should you veer off a bit. These systems will assist drivers but still require the driver to be in control. You can find Level 1 autonomy in most cars today.


Level 2 – Partial Automation

This is where it gets a bit more interesting. Although the driver must have hands on the wheel and be ready to take control at any given moment, level 2 automation can assist in controlling speed and steering. It will help with stop-and-go traffic by maintaining the distance between you and the vehicle in front of you, while also providing steering assist by centering the car within the lane. These features are a godsend for commuters! Tesla Autopilot, Volvo Pilot Assist, Audi Traffic Jam Assist are some examples of Level 2 autonomous capabilities.


Level 3 – Conditional Automation
Level 3 autonomous vehicles are capable of driving themselves, but only under ideal conditions and with limitations, such as limited-access divided highways at a certain speed. Although hands are off the wheel, drivers are still required behind the wheel. A human driver is still required to take over should road conditions fall below ideal. In some cases, certain carmakers, like Mercedes, are offering these capabilities but have them limited for a maximum amount of hands-off-the-wheel periods of time.


Level 4 – High Automation
K.I.T.T.? Is that you? Not just yet. Level 4 autonomous vehicles can drive themselves without human interactions (besides entering your destination) but will be restricted to known use cases. We’re not too far from seeing driverless vehicles out on public roads. Though regulations constrict its availability, Waymo has developed and is in the process of testing Level 4 vehicles capable of driving themselves in most environments and road conditions. If there were no regulations or legal obstacles, you’d likely see more level 4 vehicles on the road today! This is pretty much what Tesla has been offering to premium model drivers, while the government has not fully provided clearance for it.


Level 5 – Full Automation Super Pursuit Mode!

At Level 5 autonomy, we arrive at true driverless cars. Level 5 capable vehicles should be able to monitor and maneuver through all road conditions and require no human interventions whatsoever, eliminating the need for a steering wheel and pedals. Combine a Level 5 autonomous car with a capable voice assistant and you’ll have your very own K.I.T.T.! Although many of the technological components exist for an artificially intelligent car today, due to regulations and legal battles, Level 5 vehicles are probably still many years away. Until then, we’ll have to settle for partial automation.


So let’s take this to our world, financial trading.


Level 0 – No Automation
This describes your typical retail investor. 100% manual approach to trade decisions and executions.


Level 1 – Investor Assistance
Here we can find human traders are taking advantage of software and data analytics to inform their manual decision-making and trading. This is likely your typical technical trader that uses floor and ceiling breakthroughs via a software-related setup, for instance. This is more prevalent in some institutional investment firms.


Level 2 – Partial Automation

Humans take advantage of automated alerts, notification systems to drive faster and more comprehensive trade decision making.


Level 3 – Conditional Automation

Trading is automated but only under ideal conditions and with limitations. Here is where we typically find our traditional concepts of Quantitative Traders. Investment decisions were created by humans via a so-called rules-based approach. As such, a human investment manager will be required to have constant supervision and step in for manual decisions or adjustments, if the market conditions fall below ideal conditions. A key element of this level is the fact that this human-created, rule-based automated decision approach does not have the ability or has extremely limited conditional self-learning. As such, the scope of automation is limited to narrow and historical-based market conditions.


Level 4 – High Automation

In this level, self-learning of market conditions and self-judgment by algorithms is introduced to trade decisions. Algorithms and models are not only able to recognize the familiarity of conditions based on previous market states but are starting to create awareness and concept of modified or even new market states. As such, Level 4 algorithmic executions display an ability for finding new trade signals and resulting trade decisions on their own. These characteristics require significantly less human intervention as the models are able to cover a significant amount of ever-evolving market conditions. If developed properly, these level 4 algorithms have the ability to also have a strong awareness of making decisions under varying market conditions in relation to risk-reward profiles of its trades. Ultimately, creating an automated risk control mechanism. This is a significant deviation from Level 3, as in this incremental level of algorithmic automation, the trade positions can show more of a healthy, situational weighting of risk exposure versus the all or nothing positions at Level 3. Just for clarity, Level 4 still requires human oversight as the potential for course correction is a sustaining, new market condition that cannot be figured out by algorithms on their own. A great example of this “blind spot” is a systems’ inability to deal with rising bond yields after many historicals had reduced yields in their playbook.


Level 5- Full Automation

The ultimate vision! Turn on that magic switch, give the algorithms the objectives you like to accomplish and sit back with a cocktail and let it run. Here, algorithms evolve into such a general state of awareness and ability to recognize not only the changing market conditions but most importantly are able to navigate a brand new state of markets that it has never seen before. In such an environment, the algorithms were originally programmed by human interceptors but done in a way where the general approach to signal detection and risk and reward was established in an ever learning and general AI fashion so that any future intervention will not be necessary. Humans are solely supervisory in nature and only interfere during emergency situations, such as moment-based crises like a War.


The reality of the current state of algorithmic advancements overall and specifically how it is being applied is that the vast majority of algo-traders fall into Level 3. There is obviously nothing wrong with systematic trading but it is important in my mind for everybody to understand the level of human versus algorithmic controls. With that unique benefits and disadvantages come to fruition that investors need to buy into when allocating their capital. We at AlphaTrAI classify ourselves at a Level 4 of autonomy and are confident about the remaining synergies between humans and algorithms. For us, it is a matter of pursuing a Level 5 world as our core belief states non-emotional and highly scalable trading decisions by algorithms have a stronger sustainable effect on investment returns.