In the past year, machine learning and artificial intelligence have made significant advances. Specialized algorithms, including OpenAI’s DALL-E, have demonstrated the ability to generate images based on text prompts with increasing cleverness. Natural language processing (NLP) systems are getting closer to approximating human writing and text. And some people even think that an AI has become sentient. (Spoiler alert: it didn’t.)
And as Ars’ Matt Ford recently pointed out here, artificial intelligence may be artificial, but it’s not “intelligence” — and it certainly isn’t magic. What we call “AI” depends on constructing models from data using statistical approaches developed by flesh-and-blood humans, and it can fail just as spectacularly as it succeeds. If you build a model from bad data, you’ll get bad predictions and bad results – just ask the developers of Microsoft’s Twitterbot Tay about it.
For a much less spectacular fallout, just look at our backs. Readers who have been with us for a while, or at least since summer 2021, will remember the time when we tried to use machine learning for some analytics – and weren’t exactly successful. (“It turns out ‘data-driven’ isn’t just a joke or a catchphrase,” said Danny Smith, senior product manager at Amazon Web Services, when we reached out to him for advice. “‘Data-driven’ is a reality for machine learning – or data science projects!”) But we learned a lot, and the biggest lesson was that machine learning is only successful if you ask the right questions about the right data with the right tool.
These tools have evolved. A growing class of “no-code” and “low-code” machine learning tools are making a range of ML tasks increasingly accessible by harnessing the power of machine learning analytics, which was once the sole ancestry of data scientists and programmers , and make them available to business analysts and other non-programming end users.
While the work on DALL-E is amazing and will have a significant impact on the crafting of memes, deepfakes, and other images that were once the domain of human artists (with prompts like “[insert celebrity name] in the style of Edvard Munch The Scream’), easy-to-use machine learning analytics that encompass the types of data companies and individuals create and work with on a daily basis can be just as disruptive (in the most neutral sense of the word).
ML vendors tout their products as a “simple button” to find relationships in data that may not be obvious, uncover the correlation between data points and overall results — and point people to solutions that traditional business analytics take people days, months, or more to find would uncover years through traditional statistical or quantitative analysis.
We set out to do one John Henry-esque Test: To see if some of these no-code required tools can outperform a code-based approach, or at least provide results accurate enough to make decisions at less than a data scientist’s billable hours. But before we could do that, we needed the right data—and the right question.
This article was previously published on Source link
