Approximations
Tools developed in the financial industry are just approximations of the market. Some are useful approximations, such as Black-Scholes model, some are not.
Similar to a ruler, it can only measure things within its measurement error. A probability tool is just a tool that can reflect certain aspects of the reality. Probability has its own limitation for example, most real world variables does not have a concrete computable probability distribution. That made the market rather unstable if administration force the market to use probability tools.
There is a popular trend of applying machine learning tools to discover opportunities in the market. However, a crude migration of techniques from the image classification world is only going to sprinkle small waves. To really understand what is needed in order to apply machine learning to finance, one has to first understand what happened in the field of image classification.
Neural network was developed around late 1950s and it was started as a nonlinear model for classification technique. Although it has the potential of modelling any function but people find it often leads to over-fitting if one want to unleash the full potential. After a long time, there was not much invention around it. Around 2010s, many small development pieces come together, image-net, deep-net, stochastic gradient, normalization, residue-net, etc, formed a strong and powerful toolkit that can effectively solve object detection problem comparable to human level. Then people applied this advance into all kinds of gaming-AIs, including the famous Alpha Go. And it has drawn great attention from all different fields, so as financial industry.
However, to use the power of neural network in finance, it would need another 30-50 years, perhaps more, of development so that small pieces similar to image-net, deep-net, etc can be developed specifically for finance. There is no free lunch, so you can not just use tools in the image field for finance. There is no free cheap lunch, because even if you develop finance-AI after many years of hard work, it can still only approximate the financial system within its own limitation. You have to work harder to find innovative ways of modelling what finance-AI can not model.