Voice of the DBA

Tag: machine learning

  • Disclosing the How

    Many of us have written a computer program of some sort. Maybe just a script to complete a task, maybe complex software the clients interact with. For all of us, I would guess that we’ve been asked to do something, or we’ve asked ourselves to build a program, and we use that request as a specification for our code.

    When we do this, how many of us get this right the first time? I’m sure some do, and the simpler or narrower the task, the more likely we get it right. However, we also sometimes make mistakes and realize that our specification wasn’t correct or that we interpreted the specification incorrectly. Even when I write code for myself, sometimes I realize I haven’t described when I need clearly enough, even in my own head.

    That happens when I control the entire process. When someone else is my client, there are plenty of ways in which we can make mistakes, especially in decoding the specification. As we use more and more software to make decisions for us and automate our world, I wonder how often we’ve assumed the software works in a way different than it actually does. With machine learning and artificial intelligence programs, I think there is an even greater chance of a misunderstanding.

    Recently I saw that in New York City there is a proposal to disclose how software is used in hiring decisions. Humans are barred from discriminating by law, but what about software? What if there are bias that can come out in the software based on how someone enters data?

    I think there are certainly potential issues in having software evaluate individuals, though I do think software will do a better job than many people, once we’ve worked out the bugs and we know how we want it to work, and then we can compare that to how the algorithm produces results.

    These are tough issues, and difficult projects. I also think it’s important that we analyze the data that is used in testing software, to be sure that it doesn’t contain its own biases. Ultimately I think having public data sets that are evaluated by many individuals is important. I could see a reference set of data and the expected results, or some range of results being required to ensure software meets the goals we want as a society. This, along with the disclosure of how algorithms work will make our use of software better over time.

    Steve Jones

    Listen to the podcast at Libsyn, Stitcher, Spotify, or iTunes.

  • AI Can Improve Healthcare

    It’s early days, but I found this article to be very interesting. In this case, an AI model was trained to use some patient feedback on knee pain, as opposed to a previous methodology. The old methodology did work, but it had some problems with some populations.

    I haven’t thought that much of our ML/AI (machine learning/artificial intelligence) work is particularly smart. The algorithms learn well and they can match or out-perform humans, but these systems are really mimicking what humans do. They can be more reliably and definitely more scalable, but it’s doing what we humans do, not often leaping ahead.

    Often we train these models based on previous data and results from human experts. However, often what we think of as expert advice, what many people accept, is flawed in and of itself. Humans often work with a small set of data and experiences. They find patterns and create a solution that works, but not always as well as we’d like. Especially as the solution is applied to a wider variety of situations.

    In this case, researchers looked at alternative methodologies, and used AI/ML to test whether a different solution might be better. This isn’t going to replace the current methodology for now, but it might get more doctors and researchers to rethink how they approach this particular issue.

    This might be one area where AI/ML truly help humans move forward. By looking for gaps, oversights, and other problems in our existing methods, the computer might spur humans to make new leaps that help us drive forward.

    Steve Jones

    Listen to the podcast at Libsyn, Stitcher, Spotify, or iTunes.

  • The Degradation of the Turing Test

    The Turing Test from Alan Turing was proposed as a test of an intelligent system. Could a human determine if the other party in a conversation was a machine? This was an interesting way of imagining how powerful a computer might grow and the types of answers it might give to a human. Interestingly enough Turing didn’t argue about the correctness of the answers, just that they appeared to be from a human.

    In some sense, I wonder how many people would have been fooled by the GPT-3 bot on Reddit. It posted comments for a week to a variety of threads. You can look through the posts by the “thegentlemetre” user, but this one caught my eye, and as I read it, I was surprised how much this looks like things I’ve seen posted on the Internet.

    Is this AI bot intelligent? I don’t know, but I do think that the quality of comments and posts on all sorts of threads and articles seems to go down over the years. Maybe it’s humans that are becoming worse at online communications rather than computers are getting smarter. Really, I think both things are happening.

    AI/ML systems are getting better at mimicking what humans do, and I suspect that in many cases, especially in small samples, they can fool many people, perhaps most. That’s disconcerting, especially as I already feel many people are a worse version of themselves online, without feedback and social cues directly available. Having bots add to the volume of poor communications and comments doesn’t seem useful to society in general.

    While I do think that AI systems can dramatically help us with mundane tasks and tedious work, I also think there can be problems if they become rigid in their actions, without allowing for some flexibility. Humans have discretion, and while they might not use it fairly, or even in ways that their organizations approve of, but they are flexible. Seeing these posts, I wonder if the AIs can learn to be flexible as well. I think they can be.

    Steve Jones

    Listen to the podcast at Libsyn, Stitcher or iTunes.

  • The Computing Revolution

    One of the keynotes at this year’s Build conference was on the Future of Tech. This was a mix of live talk and recorded pieces from Kevin Scott, CTO at Microsoft. This talk looks at the innovation of tech, with lots of AI/ML, but also with the idea that data is fundamental to the future. While he notes that there are different eras in computing as we have breakthroughs, one of the fundamental things that will change the world is the data explosion.

    We constantly push through boundaries and past constraints that limited us just a few years ago. I can still remember a professor in college bragging about the 32MB of RAM in his Solaris workstation, at a time when many of us had 1,2, or maybe 4MB. I think about that each time I get a new device. In a generation of my life, we’ve grown an order of magnitude, and pushed into a new world where I don’t have an exponential level of computing on my desk, but in my pocket.

    We might appreciate or worry about the growth of data, and it’s implications, or we may get excited and embrace it. Either way, big data is coming, and it continues to come to many organizations. While most of us aren’t pushing 10TB+ databases, some of us are, and few of us think in MB anymore. Makes you think that perhaps the default settings in SQL Server for MB in some cases are silly. Surely a MB is a like a penny (or pence in the UK), take one, leave one, but they don’t really count. Not until there are hundreds of them, and even then, maybe only a thousand is worth much mental effort.

    I do agree that the big win with the Internet, and with many pieces of software, isn’t the platform, but the ways in which people take advantage of the platform. I think that the creativity of developers, taking advantage of incredible computing on mobile, is what has created an amazing revolution. However, it’s not really just the creative software, but the understanding and use of data at scale that has really enabled software to change the world. Access to lots of data, whether this is people wanting car rides, sharing their thoughts and images, or just the bits that make up the music of the world are the ways in which software can become incredibly popular. Also, this is how things might get incredibly dangerous.

    I do think that AI has the chance to become an inflection point in computing, helping to create more applications that will work with humans in new ways that harness our knowledge, ethics, and vision with the computing power that continues to grow in incredible ways.

    The keynote is an interesting watch. It’s about 45 minutes, but if you want to dream a bit, it’s worth your time.

    Steve Jones

    Listen to the podcast at Libsyn, Stitcher or iTunes.