Deep Knowledge Ventures, a Hong Kong based venture capital fund focused on companies developing therapies for age-related disease and regenerative medicine, has appointed VITAL (Validating Investment Tool for Advancing Life Sciences), a machine learning program capable of making autonomous investment decisions in the life science sector, to its board. VITAL uses machine learning to analyze financing trends in databases of life science companies and predict successful investments.
I would be interested in hearing everyone's thoughts on the long-term implications of AI replacing the investment Analyst and other big-data and data analysis innovations as they relate to front office M&A type roles. Will the Analyst role as we know it become obsolete? Will staffing levels be reduced going forward?
In the current stage of AI I think this is more of a publicity stunt than any real validation of AI integration into client facing roles. This is likely has a place in the future, however until AIs can do more than project possible futures based on historical data they will lack in accuracy because human nature is far more complicated than the historical trends of what ever the system analyzes.
I would echo this. In many, if not most situations, AI is able to more accurately predict possible futures than humans. However, as I said in my previous comment, there are situations in which AI algorithms will have severe limitations that would require human involvement. Complete autonomy is not possible given the current set of algorithms.
Long version: I have worked with machine learning and AI algorithms in several different industries so I think I have some ability to comment. Big data, machine learning and AI are all buzzwords now for the most part. Technically speaking, a linear regression is a very simple type of machine learning algorithm. For the most part, the mathematical foundations of machine learning and artificial intelligence are decades old. However, the hardware to support these algorithms on a large-scale is relatively new technology. Neural networks/clustering/decision trees/SVMs/etc. were all developed in the 50s/60s or even earlier. From an algorithmic perspective, there really haven't been any significant advances in machine learning and artificial intelligence. We simply have the means to put those algorithms to good use now.
For explanatory purposes, let's look at IBM's Watson super computer. Watson handily beat the best Jeopardy players of all time a few years back. For every question Watson answered, it had a confidence score, or an estimated probability that Watson's answer is correct. Usually, that confidence was quite good and over the long run, was able to beat Ken Jennings. However, if you watch the game, there were many cases in which the algorithm was not trained to handle a certain type of sentence or situation. Usually, Watson would recognize that it did not have the training to answer the question and would calculate a very low confidence score. However, I believe there was at least a few instances in which Watson was confident in its answer but was blatantly wrong.
This is the reason why AI (at least the current algorithms) will never take over an M&A analyst entirely. AI systems will always need a knowledgeable human being to do the due diligence to ensure that the AI system's prediction is sensible. In a multi-billion dollar deal, nobody is going to completely leave their decision up to a complicated mathematical algorithm that few people have the background to understand. AI may certainly lessen the need for a high-volume of analysts but could never work completely autonomously. For this, AI researchers would need to algorithmically replicate how the human brain processes information. Neural networks do this to some degree, but they are still very limited.
The short version: The current state of artificial intelligence and machine learning simply cannot replicate the intuition of a human being. On average, a computer can out-perform a human in many tasks (chess, jeopardy, etc.). However, even the most well-trained algorithms have the capacity to make terrible mistakes that even an incredibly stupid person would not make. This necessitates a knowledgeable person to ensure that the algorithm is not making a stupid mistake. With something as complex as an M&A transaction, whether or not something is stupid is not so easy.
Hopefully that helped answer your questions! PM me if you would like to discuss further.
Indeed the models can and will be wrong...... I just read a paper recently that covered errors of in models used for "back testing" and how regression overfitting produces errors in investment simulations.
Title: "Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance"
Link:: http://www.ams.org/notices/201405/rnoti-p458.pdf
I suspect for ordinary M&A analyst tasks i.e. financial modeling using structured data-sets there will be programs that will provide modeling support to the detriment of head counts.
Thanks for the input,
Cheers
Right, over-fitting is a problem for regression analysis. But that problem extends far outside the world of finance. The models I'm referring to (neural networks, support vector machines, decision trees, etc.) are significantly more adaptable than regression analysis. 99% of the forecasts these algorithms make are going to be good. It is the 1% that we have to be worried about. That's why machine learning is excellent for something like high frequency trading where you can keep "rolling the dice" so to speak and be profitable on 99% of trades.
To the science side of this, it's silly to have an artificial AI make assumptions based on a probability of financial outcomes with life sciences companies. Cash flow is quite dependent on clinical outcomes, not sure how their AI will be able to capture and reflect financial projections without making a litany of man made probabilities on clinical successes/failures.
Not saying it's impossible, I just don't believe we've captured the right "AI analytics" to take a therapeutic from the lab to the patient yet...
Now when quantum computing nears the potential for commercialization (i.e. D Wave Systems), get ready for some big change.
Agree with all of this until your last point. With respect to machine learning and artificial intelligence, there really wouldn't be any improvement here besides speed. NP-Complete problems would be solved much faster, but the actual predictions would not become more accurate if you mapped algorithms to a quantum computer. Quantum computing will definitely make huge waves in the technology world, but it does not improve our capabilities in machine learning in any sense other than efficiency.
I honestly don't know all that much about quantum computing other than the basics that you could find on wikipedia. I just know that if a regular computer and a quantum computer run the same exact algorithm for an infinite period of time, the two computers would produce the same result. Quantum computers are not inherently "smarter" than regular computers.
Whatever @"DeepLearning" chimes in with is probably reliable, as I am only a beginner with the subject myself!
The ScienceDaily website has a loads of articles to check out about Quantum Computing, I would just "google" that website.
As long as it can't bring coffee to associates analysts are secured.
This is undoubtedly a PR stunt. However, I read a study that listed the jobs that would be eliminated first, as machine learning improves. Financial analysts were one of the first.
I agree that it would be difficult for a machine to replicate the M&A advisory process...but it would be able to...for something like Equity Research. Also, if OTC derivatives are eventually forced to be listed on exchanges...most of traditional S&T would be gone, as well. Many monkeys on this board work for the Big 4. I can't imagine that auditing would be immune for very long. At the minimum, teams would be much leaner as the tools got better. This is true for all knowledge-based work, no?
@"DeepLearning" You seem to be uniquely qualified. Thoughts?
It can't destroy sales but it would certainly consolidate it. Amazon suggestions and consumer reports are good enough for me to skip listening to tv and car salesmen.
deals can also be "prepacked" for groups who know what they are doing.
Occaecati sint et rerum sunt sed quo itaque rerum. Ut ipsa id sit rem molestiae voluptas. Vitae nam amet eum dolores. Dignissimos accusantium accusantium ipsam deserunt.
See All Comments - 100% Free
WSO depends on everyone being able to pitch in when they know something. Unlock with your email and get bonus: 6 financial modeling lessons free ($199 value)
or Unlock with your social account...