The dearth of truly skilled data scientists has been noted as one of the major bottlenecks on the proliferation of AI and big data. In order for these technologies to achieve the reach they deserve, new tools must place their capabilities in the hands of people who may not have the expertise of a trained IT specialist.
Self service analytics platforms have exploded in the past few years. They consistently rise in both quality and feature depth. This allows non-techs to easily tap into the enormous amounts of data that many companies are just laying on. These programs let their end users carry out rote analytic tasks – building out data visualizations, creating reports, organizing data sets – without a reliance on the bonafide data specialists in the company.
Gartner recently assessed that non-technical workers using self service analytics programs will actually generate more analysis than the so-called pro data scientists. The executive suite has to be dancing in the streets – this means that companies are no longer relegated to finding that ultra rare, properly trained data specialist and paying the exorbitant salaries that they command.
But the democratization of AI driven data does not stop there. The largest cloud computing companies in the world – Microsoft, Amazon, Google – are making it a point to create tools that give non-techs the ability to build machine learning models. These sophisticated tools are aimed at developers, but those developers are not necessarily the computer geniuses you might think. One of the more notable services from Google is called Cloud AutoML, a service that makes use of machine learning to automate the construction of deep neural network image recognition programs.
Another program, DataRobot, also automates machine learning. Through DataRobot, users can upload data, focus on target variables and build out models based on the included learning algorithm templates. Users can quickly choose from hundreds of models without any need for specific training. The best models can be used to analyze data in the future.
There are also an increasing number of machine learning libraries in the open source environment with the building blocks needed for custom algorithms. Cortex from CognitiveScale is said to be the first AI model builder with a graphical user interface. Basically, there is a tool available for every level of experience and education, from unwashed newbie to seasoned developer.
The Need for the Democratization of AI
At its core, artificial intelligence is about prediction. No matter how smart they seem, computers are not able to understand things in the same way that humans do. Computers interpret data (albeit very precisely and without the memory holes of human beings) in order to create a close interpretation of a model or schema.
Although machine learning is modeled from human cognition, they are not the same thing either. A machine driven to “learn” can discern different kinds of functions from data and very precisely predict outcomes from that data. However, it cannot go outside of the algorithms that it is given. The computer cannot take risks or give an explanation for outliers. On the other hand, computers are automatic experts at the most complex forms of math that the majority of humans have only a passing knowledge of. At high levels, the probability statistics and algorithms that computers have access to can outperform the average human intuition, especially that of the untrained individual.
Basically, the market does not have the money or the patience to bring enough people up to speed on this kind of math to compete with and control computers. Why spend the tens of thousands of dollars to train a human being, wait years for the investment to pay off, and risk turnover when a computer under $1,000 can perform the same functions instantly?
The primary reason that AI is democratizing itself is because the market for IT skills is coming into a natural balance. Companies do not want to take the risk to train high level employees when computers can outperform them for a fraction of the cost. At the same time, even the most sophisticated AI driven models cannot account for all of the discrepancies in data that humans create. Companies still need someone with a working, creative brain to interpret results and feed data correctly into the machine learning algorithm.
What is the result? Demand for an easier synergy between computers and the humans who are interpreting and guiding them. There is simply too much demand in the market for the leaders in AI to ignore. Full democratization down to the point of the average consumer is only a matter of time.
Cheap AI and the Job Market
Does AI for everyone mean that no one will have a job? Although the notion of robots replacing cashiers at McDonald’s seems tenable, it is actually unfounded for the most part. The automation that AI makes more approachable has been around for quite some time. The transition to AI related technologies over traditional labor has been slow, industry experts like HBR tell us.
What’s more, AI will bring new opportunities for jobs to properly trained individuals. The democratization of AI means that “proper training” does not actually require a pH.D. in computer technology. As long as you know how to optimize the use of Microsoft Cortana, you may actually be in the running for some pretty good positions in the coming years.
There will be some job churn, say experts, because many rote tasks will no longer require humans to perform. These are mostly low skill jobs that center around data entry. However, even these people have a chance to save their positions if they incorporate democratized AI into their skill sets. As the end user tech becomes more sophisticated, employees may be able to add AI to their repertoire without having to go back to school.
What can the market expect once this new tier of employees is able to bring value to a company through simply understanding the latest UI? Experts say that we can expect a curbing of wages for data scientists and AI experts, because there will be so many more of them for companies to utilize. This will also create defined tiers between professionally trained data scientists and end user experts (EUE). While EUE may be able to handle lower level stuff, the classically trained data scientists will be used for higher level tasks.
Coming Down Off the Ivory Tower
Bringing advanced technologies down to the level of the layperson – this seems like such a noble cause at first. Digging a little deeper, however, you will find some of the other possible reasons that the big guys are so adamant about simplifying AI for the masses. Let’s take a look at some of the underlying motives and what that means for the tech industry as a whole.
There is a direct profit motive. If Microsoft, Google, Amazon, etc. deliver relevant services to the masses based on AI they develop, then they have full control of how to price those services. (Whether you buy a red cotton shirt or a blue cotton shirt, the cotton producer always gets paid.) The big guys have already perfected the subscription model service that keeps users hooked on the benefits of democratized AI without ever gaining ownership over any of the functionality.
Democratizing AI keeps new developers from searching out source code. If the big guys can lock out smaller developers from learning the basics of what makes AI tick, then they control the rate and level of innovation from then on. (You can make a White Russian from Bailey’s cream and Grey Goose vodka, but neither Bailey’s nor Grey Goose will ever give you the ability to control the ingredients in the drink. Every time you want to make a new cocktail, you have to come through Bailey’s and Grey Goose.)
Dumbing down AI future proofs the industry against further competition. When someone says they “built a PC” in the modern parlance, it does not mean that he personally soldered a motherboard together. In previous generations of computer specialists, there was no template motherboard to buy at the local shop. If you wanted to build a PC, you had to know how to solder the small parts of the motherboard together yourself. Future generations of AI users will be completely shut out of the AI creation process. They will take templates, processes and library functions that have been laid out for them. The core technology will belong to the Microsofts and Googles of the world, and it will be increasingly difficult for developers in the future to drill down into the core of the technology.
With these reasonings in mind, is democratization an all around good thing? It must be questioned. Will the underlying technology, logic and code for AI still be kept high in that ivory tower, even as the big tech guys say that they are sharing the wealth? It may be time for developers of all skill levels to take a look at the bigger picture as well as utilizing current technology to bring new innovations.
Bringing AI to the masses will certainly have consequences that have yet to be determined. Such powerful tools in the hands of the average person may lead to extreme outcomes that are both terrible and wonderful, and we should be happy to see them all. The more that we give technology to as many creative minds as possible, the faster that we can all learn from the results.
Are you interested in learning more?