Here\u2019s the problem: at my normal rate of approximately two conversations with Chief Information Security Officers (CISOs) per month, the \u201cpresent discounted value\u201d of the information gathered never quite brings this picture into clear focus \u2013 this is where the RSA Conference comes to the rescue.Lessons learned from RSA Conference 2018If you work in the cybersecurity industry, April was marked on your calendar as the month the now mammoth RSA Conference that took place in San Francisco. While this is always a time to catch up with former colleagues who also work in the security business, it also presents a unique opportunity to meet with many CISOs in a short span of time. I personally took part in approximately 12 meetings with CISOs over a span of three days.Here are some of the things I learned from those meetings:The notion that a strategy built primarily around preventing bad things from entering an environment is effective seems to have finally been put to rest. The rebalancing of investments around true defense-in-depth \u2013 arraying your prevention technology to take care of the known bad and having tools and people ready to deal with threats still making it through \u2013 is well underway.CISOs are still experimenting with how much preventive effort is enough and when to begin doubling down on their detection and response capabilities. My description of this to them seems to resonate: prevention is noise-reduction while detection is signal amplification. There is no perfect answer for what mix produces the clearest signal.The hype surrounding artificial intelligence seems to have abated somewhat. Most CISOs view Artificial Intelligence (AI) as an automation technique and are now asking the right question: \u201cwhat AI will actually do for me?\u201dThere\u2019s muddled use of the terms \u201cartificial intelligence\u201d, \u201cdata science\u201d and \u201cmachine learning\u201d. AI can exist without Machine Learning (ML) \u2013 remember \u201cexpert systems\u201d of a couple of decades ago, which were a very different form of AI. Data science need not use ML \u2013 it\u2019s just the use of algorithms to extract meaning out of data. ML is having an algorithm learn from data rather than having a human encode the logic. There\u2019s still plenty of work for humans to set up the data, possibly label it, select features for the algorithm and tune it to achieve the desired results \u2013 in other words, it\u2019s not magic.Lots of companies are experimenting with data lakes and data ponds (and maybe even data puddles). The precise end goals of those experiments are not very clear. They usually start with collecting a lot of different kinds of data in a Hadoop or Elasticsearch cluster and then unleashing data scientists on those clusters. The data scientists hired for each problem domain tend to have no background in that domain \u2013 e.g. an ex-physicist is hired to do data science for cybersecurity. And each problem domain has at most 1-2 data scientists working on it.I did not hear a single customer claim success in real-time detection of cyberattacks with their cybersecurity data lakes, but there were examples of them being able to perform better investigations and forensics after-the-fact. And even before RSA, I have heard of several companies declaring their first foray into the crossroads of data science and cybersecurity a failure \u2013 either because they never were able to get value out of it or because it became impossible to maintain the value over time. Against this backdrop, retention of data scientists is also an issue.3 predictions for RSA Conference 2019It\u2019s interesting to consider what 2019 will bring. Let me take a crack at three predictions:The prevention vs detection balance will have matured. More investment in products and people will have shifted from the prevention column to the detection column with emphasis on commercial products that leverage data science to improve the performance of the prevention side and to wrestle meaning out of data for the detection side.Cybersecurity data science experiments will continue \u2013 with some companies still going through their first iteration and others starting on their second one. This experiment is mirrored by other parts of their organizations also trying to extract value out of data science. And retention of data scientists will continue to be an issue as their skills are in high demand in almost every industry and working on endeavors that do not appear to be succeeding leads to unhappy employees.Like the rather unfocused-SIEM projects of a decade ago that eventually became focused on specific use cases, security teams will learn to focus their cybersecurity data science efforts on tractable problems for which they have the necessary data and skills. Crawl, walk and, eventually, hope to run.All in all, it was a super productive week. I would need to log about tens of thousands of air miles to have as many quality meetings with this number of CISOs. That would involve a lot of time in airports, plenty of jet lag and lots of bad airplane food.Come to think of it, maybe we can just fast forward to RSA Conference 2019.