Voice of the DBA

Tag: AI

  • Badly Trained AI

    Most of us know that data is being used to make more and more decisions inside of all kinds of organizations from retail giants to banks to sports teams. We are constantly asked, or see reports, of data driven decisions. We often need to show some data that supports and explains the rationale for making some choice. As our populace becomes more data savvy, I expect this trend to continue.

    AI (Artificial Intelligence), and the related Machine Learning (ML), are becoming more and more widely used. From mobile phones to autos to trading systems, we regularly see new “AI capabilities” being added to products and services. No business or industry seems immune, and I’m sure many of you are seeing AI being incorporated or feeling pressure to start using some AI in your work. As you work with AI, or start to, you’ll quickly realize the importance of data in your efforts.

    This is true for the cleanliness of data, but perhaps even more important in the tagging of data sets. As Amazon learned, building an AI or ML system, is hard. They scrapped one system that was being used to rate resumes and help their recruiters sort through the volume of applications they received. Why? Because of bias.

    Apparently the system would downgrade women’s resumes for various reasons. To me, this is a perfect example of a principle I’ve had throughout my career: garbage in garbage out. In this case it’s not necessarily bad data that was the problem, but bad tagging of what was a good and bad resume, probably from the internal prejudices of a few people.

    There will be more dangers as we use ML and AI technologies in our work. It won’t be enough that we clean the raw data for training, but also that we clean and properly manage the tagging of what data sets represent the results we are looking for. Like in much of our software, it’s easy for us to only consider the happy path, to only tag those items we think are good results. That is useful, but we might also be unconsciously tagging other results as bad, which appears to have happened to Amazon.

    We can build systems that do a better, more rational job than most humans, but we need extraordinary care to ensure our training data lacks bias. Unfortunately, most people both think they’re not biased and are unwilling to spend extra resources to deeply examine the data. Two things that worry me about the future of our AI/ML systems that will inform us.

    Steve Jones

  • Self-Healing ETL

    The process of building ETL flows is cumbersome and time consuming. Many consultants have made a lot of money by building ETL processes that manage the movement of data in and out of databases, including to data warehouses. There is a lot of importance placed on consistent, quick, and accurate data loading. In fact, one of the most popular job needs these days is still for ETL developers. Even as data science grows and becomes popular, quite a bit of the work in data science is data preparation, which often involves ETL, or ELT.

    One of the struggles with an ETL process is the tedious nature of building lots of flows that are very similar. There may be a wide variety of ways to get data to move, especially as we have a crazy set of inconsistent formats in which data appears, but once we have the flow, we often repeat that for many different files or tables. I think Biml makes this better, but it’s still not simple, and there can be plenty of issues that still arise as the source inputs change.

    There was an interesting article that describes the use of AI to assist in ETL flows. While there is some basic matching up of source and target data right now, what if AI would add a few things. The idea of looking at the data you’ve matched and suggesting alterations that can potentially help with data quality is interesting. The system would have to learn what issues occur over time and what good data looks like, but perhaps this could even extend when the source data gets slightly mangled in formatting. Perhaps detecting fixed width fields that have grown and shifted all columns would be possible, and even adjust the data flow.

    I am actually more interested in detecting PII data and alerting developers, or even operations people that the ETL process needs to be changed or secured in some way. AI should be able as well to help Operations people with scheduling, even pausing or moving ETL work to other machines to better scale the flow. My view would be that any AI system ought to be looking for anomalies and detecting those, as pattern recognition is one thing AI is good at.

    I’d think this would be more of an alert capability as I’m not sure I trust an AI system to actually adjust data. ETL flows are often too important to have data mangled in an unexpected way. Perhaps that’s too cautious, as that’s what developers do all the time. They mandle data, and we find ways to recover. Ultimately, I think that having lots of logging on what a developer or AI system does is more important. This might be especially true if voice commands are used to build the flows. I could see lots of ambiguities and mistakes from business analysts that are trying to describe how to build an ETL flow.

    Steve Jones

    The Voice of the DBA Podcast

    Listen to the MP3 Audio ( 4.2MB) podcast or subscribe to the feed at iTunes and Libsyn.

  • AI Regulators

    With the GDPR now being enforced in the European Union, there are plenty of companies that are getting concerned about the potential fines from regulatory authorities if they aren’t complying with the law, or at least, making an attempt. There certainly is leeway for regulators to adjust fines or give warnings if a company is making efforts to comply. This has likely contributed to the work inside many organizations to move towards compliance.

    There are likely some companies that might not worry, since there are relatively few regulatory employees and many companies. There are lots of complaints coming in, which could easily overwhelms the relatively small staff in each EU country. Complaints might not be investigated in a timely manner or even lost because of the workload. The problem will likely get worse as more consumers complain about data processing practices. I don’t expect regulatory authority staffing to increase, so I’m sure only the most aggregious or complained about companies will get caught.

    There is one way to help amplify the capabilities of the relatively small staffs reviewing complaints. There are researchers in the EU Institute in Florence that are are working with consumer organizations to create AI programs that can help by performing some of the work. The initial thrust is to evaluate privacy policies of companies. If there are issues, the software doesn’t assess a fine, but it does alert a human to perform additional checks.

    In one sense, this is exactly what computers can do well. They amplify the capabilities of humans by doing a piece of the work. We can build systems, whether traditional programmed ones or AI based applications, that handle a piece of the work that requires lots of human labor. Once initial evaluations are made, a human can review the work and make more refined judgments.

    The danger, to me, is that humans will be lazy. They’ll start to trust the AI systems as authorities and use less of their own judgment, mostly because it’s just easier. I could see these systems evolve over time to actually train humans involuntarily. New employees would initially trust the AI results, learning from the AI rather than teaching it and constantly evaluating its effectiveness.

    I think AI can really help improve the way that we accomplish work in many ways, but it should be regularly audited and approached with some skepticism. There certainly needs to be some sort of supervisory group overseeing the program that isn’t involved in the outcome. We should be sure that the goals and results from any AI system continue to be focused on what we want to achieve, and that we transparently define those goals for anyone impacted. Otherwise we might end up having AIs evolve in ways that are counter to the original purpose.

    Steve Jones

    The Voice of the DBA Podcast

    Listen to the MP3 Audio ( 3.5MB) podcast or subscribe to the feed at iTunes and Libsyn.

  • AI Regulators

    With the GDPR being enforced in the European Union, there are plenty of companies that are getting concerned about the potential fines from regulatory authorities if they aren’t complying with the law, or at least, making an attempt. There certainly is leeway for regulators to adjust fines or give warnings if a company is making efforts to comply.

    Other companies might not worry, since there are relatively few regulatory employees and lots of companies. There are lots of complaints coming in, which could easily overwhelms the relatively small staff in each EU country. The problem will likely get worse as more consumers complain about data processing practices.

    There is one way to help amplify the capabilities of the relatively small staffs reviewing complaints. There are researchers in the EU Institute in Florence that are are working with consumer organizations to create AI programs that can help. The initial thrust is to evaluate privacy policies of companies. If there are issues, the software doesn’t assess a fine, but it does alert a human to perform additional checks.

    In one sense, this is exactly what computers can do well. They amplify the capabilities of humans by doing a piece of the work. We can build systems, whether traditional programmed ones or AI based applications, that handle a piece of the work that requires lots of human labor. Once initial evaluations are made, a human can review the work and make more refined judgments.

    The danger, to me, is that humans will be lazy. They’ll start to trust the AI systems as authorities and use less of their own judgment, mostly because it’s just easier. I could see these systems evolve over time to actually train humans involuntarily. New employees would initially trust the AI results, learning from the AI rather than teaching it and constantly evaluating its effectiveness.

    I think AI can really help improve the way that we accomplish work in many ways, but it should be audited and regularly approached with some skepticism by some sort of supervisory group. We should be sure that the goals and results from any AI system continue to be focused on what we want to achieve, and that we transparently define those. Otherwise we might end up having AIs evolve in ways that are counter to the original purpose.

    Steve Jones