Several machine learning based investment models are already deployed and functioning.
Some of them focus on technical analysis and apply investment models based upon daily fluctuations in the market. Having the ability to observe patterns, provide quick feedback, and adjust course, these algorithms take positions, and exit from positions, while accumulating trading profits and simultaneously learning from each transaction.
Then there are other models that are based upon the fundamentals of the stock. By understanding the underlying value drivers of a firm (e.g. profits, cash flows, cost of capital etc.) these AI based systems learn to trade and/or manage a portfolio,
The one that I found of great interest is the one developed by SignalFire. This algorithm is designed to pick the best startups. This is the most unique implementation since it uses nonstandard methods of finance. In other words, it is not exclusively patterned after standard investment theory (even though it may use a lot of it) and instead uses millions of data points. The algorithm has selected 8 companies as the best startups to invest in. You can read the story here.
The good part: You can pick the best startups (as determined by the algorithm). It learns and with each interaction becomes smarter. Arguably, the venture capital investment industry will benefit from rapidly eliminating the potential losers, removing investment bias, and focusing on the ones that matter.
The bad part: Assuming that the datasets used to train the system were probably composed of existing success stories (and failure stories i.e. what not to pick), we are essentially codifying that success is achieved by the presence of certain variables – and since the existing success in the society is not a representative of all potential success into the future, the patterns of success codified will be a representation of the current model of success. In other words, if the patterns of success indicate that the successful enterprises are formed by people of a certain ethnic group and sex, who graduated from a certain university, and who live and work in the Silicon Valley – the algorithm may arbitrarily become prejudice toward someone living in Arkansas who didn’t attend a certain university. That person will not even get a chance to present her business plan to the venture capitalist industry. Lastly, the employment impact will be negative on the traditional investment analyst jobs in the VC industry.